Interaction of NDRG1 and TGM2 modulates DNA replication and repair

Abstract

In tumor cells, DNA replication is constantly challenged by endogenous and exogenous sources, referred to as replication stress, and various pathways have evolved to mitigate this stress in cancer. We recently identified a novel extracellular matrix-induced DNA repair pathway involving NDRG1 (N-myc downstream regulated gene 1). Matrix-induced signaling results in NDRG1-dependent protection from chemotherapy-induced replication stress. To uncover further mechanistic details of NDRG1-mediated effects on DNA replication, we identified Transglutaminase 2 (TGM2) as a novel NDRG1 binding partner. TGM2 is an acyltransferase that catalyzes Ca(2+)-dependent protein modifications. This interaction was enriched upon chemotherapy-induced replication stress and also upon ECM-induced signaling. Our data show that TGM2 depletion significantly slows replication fork progression, and this phenotype is dependent on TGM2 catalytic activity and its nuclear localization. Our study further identified a putative NDRG1-TGM2 binding site and show that the physical interaction between NDRG1 and TGM2 is required for efficient DNA replication.

Introduction

DNA replication is a fundamental cellular process that results in the accurate duplication and transmission of genetic information to daughter cells. Faithful DNA replication is essential to maintain genomic stability. In mammalian cells, DNA replication occurs at individual replication origins along the genome that form replication forks. During DNA replication, a multiprotein machinery known as the replisome, coordinates the unwinding of the DNA double helix, and replication of the two DNA strands [1]. DNA replication forks are frequently challenged by endogenous and exogenous sources, including but not limited to nucleotide depletion, lesions in the DNA, secondary DNA structure, transcription complexes, and oncogene-induced stress [2]. These challenges to replication forks result in DNA replication stress, which is defined as the slowing or stalling of replication fork progression or DNA synthesis [2]. Replication stress can have negative consequences such as increased genomic instability and DNA damage [2]. During replication stress, the replicative helicase continues to unwind the parental DNA double helix while the polymerase has stalled. This results in the formation of stretches of single-stranded DNA (ssDNA), which acts as a physiological intermediate at replication forks that triggers the replication stress response [2]. ssDNA at stalled replication forks gets coated by replication protein A (RPA) and ssDNA-RPA serves as a platform to recruit and activate ataxia telangiectasia and Rad3-related (ATR) checkpoint signaling, which helps to mediate the effects of replication stress by stabilizing and remodeling stalled forks and arresting the cell cycle [3].

Replication stress has been linked to tumorigenesis due to oncogene activation, which is known to drive uncontrolled cell proliferation, deregulated DNA replication, and genome instability [4]. Additionally, replication stress has been identified as a crucial vulnerability of cancer cells that has the potential to be therapeutically targeted. Currently, there are several clinical trials investigating the therapeutic efficacy of targeting kinases that coordinate the replication stress response such as ATR, CHK1 and WEE1 [5]. Furthermore, in several cancers including pancreatic ductal adenocarcinoma, chemotherapeutic drugs are the standard of care, and many chemotherapies target DNA replication by inducing replication stress [6]. Unfortunately, cancer cells can frequently overcome chemo-induced replication stress, underscoring the need to better understand the replication stress responses in cancer cells to develop more effective therapeutic strategies [5].

Our recent work has identified a novel extracellular matrix-induced resistance mechanism involving NDRG1 (N-myc downstream regulated gene 1). Our data revealed that cancer associated fibroblasts secrete matrix proteins that in turn activate an intracellular signaling cascade that results in phosphorylation of NDRG1 at Thr346 site. Phosphorylated NDRG1 in turn was required to protect tumor cells from chemotherapy-induced replication stress and DNA damage (bioRxiv 2025.01.22.634323). To uncover further mechanistic details of NDRG1-mediated DNA repair and replication, we performed several biotin proximity-labeling technique (BioID) screens in the nuclear and cytoplasmic cell fractions and identified Transglutaminase 2 (TGM2) as a novel binding partner of NDRG1.

TGM2 is a multifunctional transamidating acyltransferase that catalyzes Ca(2+)-dependent protein modifications. TGM2 catalyzes the crosslinking of proteins, and is involved in several cellular processes including apoptosis, extracellular matrix scaffolding, and intracellular signal transduction [7]. Furthermore, TGM2 has also recently been shown be highly expressed and contribute to poor prognosis and chemoresistance in several cancer types, including pancreatic and colorectal cancers [8, 9]. TGM2 has been known to localize to the extracellular space, cytoplasm, and the nucleus for 40 years, however the nuclear functions of TGM2 have only begun to be explored relatively recently [10]. Some recent studies have shown that nuclear TGM2 plays an epigenetic regulatory role through inducing post-translational modifications called serotonylation on histones [11, 12]. Another recent study has shown that nuclear TGM2 interacts with topoisomerase IIα (TOPOIIα) to promote DNA double strand break (DSB) repair in lung cancer cells [13]. To our knowledge, TGM2 has not yet been studied in the context of replication stress. Therefore, this study aimed to investigate the role of TGM2 in DNA replication and in the NDRG1-dependent DNA replication and repair pathway.

We found that the interaction between NDRG1 and TGM2 was regulated by replication stress-inducing conditions, including gemcitabine and hydroxyurea treatment as well as extracellular matrix signaling, and that this interaction was modulated by SGK1-mediated phosphorylation of NDRG1 in pancreatic cancer cells. Depletion of TGM2 resulted in impaired replication fork progression, and this phenotype was dependent on TGM2 catalytic activity. The nuclear localization of TGM2 was critical for its role in DNA replication. We identified a putative NDRG1-TGM2 binding interface and found that disruption of the physical interaction between NDRG1 and TGM2 resulted in slowed DNA replication. Mechanistically, we found that the interaction between NDRG1 and TGM2 promotes the nuclear localization of TGM2, providing a functional link between the interaction and replication fork regulation. Taken together, our work has uncovered a novel protein interaction that modulates DNA replication fork dynamics and the cellular response to replication stress. Understanding how this pathway integrates extracellular signals with replication stress responses may inform future efforts to target replication-associated vulnerabilities in cancer cells.

Materials and Methods

Cells Lines

SW1990 (CRL-2172) and HPAC were a kind gift from Dr. Nada Kalaany. U2OS, and HEK293T cells were purchased from ATCC. MDAMB-468, MDAMB-231, MCF7, CAOV-3, OVCAR-3, KURAMOCHI cells were a kind gift from Dr. Alex Toker. All cell lines were tested Mycoplasma negative by using the MycoAlert Detection Kit (Lonza) regularly and STR authenticated in the last three years. Cells were maintained in DMEM supplemented with 10% FBS, 50 IU/ml penicillin, and 50 μg/mL streptomycin, and new vials thawed after ~30 passages.

Plasmids, Oligonucleotides, and Cloning

All oligonucleotides used for cloning can be found in Supplemental Table 2. The coding sequence of WT NDRG1 (bioRxiv 2025.01.22.634323) was subcloned into mycBioID2-pBABE-puro (kind gift from Kyle Roux Addgene #80900, no longer available on Addgene) via BamHI and EcoRI, allowing for the expression of the C-terminal Myc-tagged NDRG1-fused to BirA (R118G). LentiCRISPR v2 was a gift from Feng Zhang (Addgene plasmid# 52961; http://n2t.net/addgene:52961 ; RRID:Addgene_52961[14]), the NDRG1 KO construct was created by annealing the gNDRG1-KO2 sense and antisense oligonucleotides (Supplemental Table 2), followed by subcloning in the LentiCRISPR v2 backbone. Constructs for TGM2 KO were created by annealing the gTGM2-KO1, gTGM2-KO2, gTGM2-KO3 sense and antisense oligonucleotides (Supplemental Table 2), followed by subcloning in the LentiCRISPR v2 backbone. The control constructs were created by annealing the gAAVS1 sense and antisense oligonucleotides (Supplemental Table 2), followed by subcloning in the LentiCRISPR v2 backbone. Coding sequence of CRISPR-resistant NDRG1 with mutated PAMs was ordered from GenScript, and further subcloned into pLenti6/V5-p53_wt p53 (gift from Bernard Futscher, Addgene plasmid #22945 [15]) by cutting out the coding sequence of p53 and pasting the coding sequences of NDRG1 variants into via XbaI and SpeI restriction sites, allowing the generation of the C-terminal V5-tagged NDRG1-WT construct. The pLenti6-V5-NDRG1-WT construct was used to generate pLenti6-V5-NDRG1-Q117A and pLenti6-V5-NDRG1-Q117E mutant constructs via site directed mutagenesis (QuikChange Lightning Multi Site-Directed Mutagenesis Kit, Agilent Technologies #210518) using the NDRG1-Q117A and NDRG1-Q117E primers (Supplemental Table 2). Coding sequence of TGM2 was obtained from pGST-PSP-hTG, a kind gift from Jeffrey Keillor and Ray Truant (Addgene plasmid #85470 [16]), by PCR amplification using primers flanked with SpeI and XhoI restriction sites (Supplemental Table 2). SpeI and XhoI digestion of the TGM2 PCR fragment and pLenti6/V5-p53_wt p53 plasmid, followed by ligation allowed for the generation of C-terminal V5-tagged TGM2-WT construct. Site directed mutagenesis (Q5 Site-Directed Mutagenesis Kit, NEB #E0554) was performed on the pLenti6-V5-TGM2-WT construct using the TGM2-CRISPR-res primers to create TGM2-gKO3-CRISPR-resistant rescue constructs. TGM2 catalytic mutants were generated via site directed mutagenesis (QuikChange Lightning Multi Site-Directed Mutagenesis Kit, Agilent Technologies #210518) using the TGM2-C277S and TGM2-Y516F primers (Supplemental Table 2). Similarly, TGM2 Ca²⁺ binding site mutants were generated via site directed mutagenesis (QuikChange Lightning Multi Site-Directed Mutagenesis Kit, Agilent Technologies #210518) using the TGM2-Ca1 and TGM2-Ca3A primers (Supplemental Table 2).

HA-tagged TGM2 and NDRG1 constructs were generated using the pICE-HA-MRE11-WT construct as a backbone, a kind gift from Sebastien Britton & Patrick Calsou (Addgene plasmid #82033 [17]). TGM2 WT sequence was PCR amplified from pGST-PSP-hTG, a kind gift from Jeffrey Keillor and Ray Truant (Addgene plasmid #85470 [17]), using primers flanked with AgeI and KpnI restriction sites (Supplemental Table 2). AgeI and KpnI digestion of the TGM2 PCR fragment and pICE-HA-MRE11-WT plasmid, followed by ligation allowed for the generation of N-terminal HA-tagged TGM2-WT construct. To generate pICE-HA-NDRG1-WT plasmid, human NDRG1 sequence was PCR amplified from pET28a-TEV-NDRG1 plasmid, a kind gift from Venla Mustonen and Salla Ruskamo [18] using primers flanked with AgeI and KpnI restriction sites. AgeI and KpnI digestion of the NDRG1 PCR fragment and pICE-HA-MRE11-WT plasmid, followed by ligation allowed for the generation of N-terminal HA-tagged NDRG1-WT construct.

To generate GST-tagged constructs, GST gene fragment was ordered (Integrated DNA Technologies). To generate a negative control pICE-GST plasmid, a NotI and KpnI flanked GST sequence containing a C-terminal flexible linker (GSAGSAAGSGEF) and stop codon was generated by amplifying the GST gene fragment using two sequential rounds of PCR with the NotI-GST-STOP-KpnI primers. The GST-stop PCR fragment and pICE-HA-TGM2-WT plasmid were digested using NotI and KpnI, and ligated to generate the pICE-GST plasmid.

To generate pICE-GST-TGM2 and pICE-GST-NDRG1 plasmids, a NotI and AgeI flanked GST sequence containing a C-terminal flexible linker (GSAGSAAGSGEF) without a stop codon was generated by amplifying the GST gene fragment using two sequential rounds of PCR with the NotI-GST-AgeI primers (Supplemental Table 2). The GST PCR fragment, pICE-HA-TGM2-WT, and pICE-HA-NDRG1-WT plasmids were digested using NotI and AgeI and ligated to generate pICE-GST-TGM2-WT and pICE-GST-NDRG1-WT plasmids.

To generate TGM2-NLS and TGM2-NES constructs, SpeI and XhoI flanked TGM2 fragments containing C-terminal NLS or NES sequences were generated by amplifying the TGM2 sequence from pLenti6-V5-TGM2-WT (PAM mutated) using two sequential rounds of PCR with the TGM2-NLS/NES primers (Supplemental Table 2. TGM2-NLS and TGM2-NES PCR fragments, and pLenti6-V5-TGM2-WT were digested using SpeI and XhoI, and ligated to generate pLenti6-V5-TGM2-NLS and pLenti6-V5-TGM2-NES plasmids. Transient knockdown of the TGM2 and NDRG1 were achieved by transfecting cells with smartpool ON-TARGETplus Human TGM2 siRNA (Horizon Discovery, L-004971-00-0005), ON-TARGETplus Human NDRG1 siRNA (Horizon Discovery, L-010563-00-0005).

Production of Lentiviruses and stable cell lines

For the generation of lentiviruses, 70-80% confluent HEK-293T cells were co-transfected with psPAX2 (a gift from Didier Trono, Addgene plasmid # 12260), pVSVg (gift from Joan Brugge), and one of the following vectors: LentiCRISPR v2 with subcloned gNDRG1 KO-1,-2,-3, gTGM2 KO-1,-2,-3, gAAVS1 and pLenti6-V5-NDRG1-WT or pLenti6-V5-NDRG1-Q117A/E, and pLenti-V5-TGM2-WT or pLenti-V5-TGM2-NLS/NES or pLenti-V5-TGM2-C277S/Y516F, or pLenti-V5-TGM2-Ca1/Ca3A by Lipofectamine 3000 (Thermo Fisher). Media was collected over a period of 24 to 72 hours post-transfection in 24-hour batches, pooled, centrifuged and filtered through a 0.45 μm filter. LentiCRISPR v2 infected cells were cultured in the presence of Puromycin (1 μg/ml) for 96h. KO cells were later infected with pLenti-TGM2 or pLenti-NDRG1 expressing viral particles, followed by selection with Blasticidin (15 μg/ml) for 7-10 days.

Western blot

Cells were lysed in RIPA buffer (BP-115, Boston Bioproducts) containing complete protease inhibitor cocktail tablet (Roche). Cells were then sonicated, and centrifuged at 14,000 × g for 10 minutes at 4°C. Proteins (20 μg per sample) were separated by electrophoresis on 4–12% pre-cast Tris-Glycine gels (Invitrogen) and transferred to nitrocellulose membranes. The membranes were blocked in 5% milk in Tris-buffered saline with Tween 20 (TBST), and incubated with the following antibodies: TGM2 (#3557, CST), NDRG1 (#9485S, CST), Vinculin (#13901, CST), GAPDH (#2118, CST), Myc (#2276, CST), Tubulin (#ab7291, Abcam), HA (#2367, CST), GST (#sc-138, Santa Cruz Biotechnology), H3 (#4499, CST), pATR-S1989 (#58014, CST), pATR-S428 (#2853, CST), ATR (#2790, CST), pChk1-S317 (#12302, CST), pChk1-S345 (#2348, CST), Chk1 (#2360, CST), RPA70 (#2267, CST), Cyclin E2 (#4132, CST), pH3 (#9701, CST). Appropriate secondary antibodies (peroxidase-conjugated IgG, CST) were used. The ECL Kits (Thermo Fisher) were used for signal detection. The Westerns were imaged by Amersham Imager 600. Images were imported to ImageLab software (Biorad).

Co-immunoprecipitation

For co-IPs performed on whole cell extract, cells were lysed in RIPA buffer (BP-115, Boston Bioproducts) containing complete protease inhibitor cocktail tablet (Roche). Cells were then sonicated, and centrifuged at 14,000 × g for 10 minutes at +4°C. For co-IPs performed on nuclear fraction, cells were incubated in hypotonic buffer (20 mM Tris-HCL pH8.3, 10mM NaCl, 3mM MgCl2, protease inhibitor tablets) on ice for 10 minutes. Cells were scraped and 10% NP40 detergent was added to the cell suspension. Cells were vortexed and centrifuged at 14,000 × g for 10 minutes at +4°C to isolate cytoplasmic fraction (supernatant). Pellet was suspended in RIPA buffer (BP-115, Boston Bioproducts) containing complete protease inhibitor cocktail tablet, sonicated, and centrifuged at 14,000 × g for 10 minutes at +4°C to obtain nuclear fraction. Cells were incubated with hypotonic buffer (20 mM Tris-HCL, pH 8.3, 10 mM NaCl, 3 mM MgCl₂, protease inhibitor tablets) on ice for 10 minutes. Cells were scraped and NP40 was added at a final concentration of 0.5% to cell suspension. Cells were vortexed and centrifuged at 14,000 × g for 10 minutes at +4°C to isolate cytoplasmic fraction (supernatant). Pellet was suspended in RIPA buffer (BP-115, Boston Bioproducts) containing complete protease inhibitor cocktail tablet, sonicated, and centrifuged at 14,000 × g for 10 minutes at +4°C to obtain nuclear fraction. For GST IP, 400 mg protein per sample was incubated with 50 μL of glutathione beads (#78602, Thermo Fisher) for 2 hours at +4°C while rotating. For anti-NDRG1 IP samples, 400 mg protein per sample was incubated with 2 μL NDRG1 (#9485S, CST), or 0.5 μL IgG Rabbit negative control (#2729, CST) antibodies per sample for 2 hours at +4°C or 1 hour at room temperature while rotating. 60 μL of Protein A/G beads (#78609, Thermo Fisher) were added to each sample and incubated overnight at +4°C or 1 hour at room temperature while rotating. After incubation, beads were washed three times with RIPA buffer (BP-115, Boston Bioproducts) using magnetic stand, and beads were boiled in 2x SDS loading buffer for 8 minutes. Magnetic stand was used to separate beads, and samples were loaded in 4–12% pre-cast Tris-Glycine gels (Invitrogen) and Western blot was performed.

Immunofluorescence

Cells were plated on coverslips and grown for 24 hours. Cells were fixed with 4% PFA for 15 min at RT and permeabilized with 0.1% Triton-X 100 in PBS for 10 min at RT. Coverslips were blocked in 5% bovine serum albumin (BSA, BP1600-100, Fisher) for 1 hour at RT. Coverslips were incubated with primary antibodies at +4°C diluted in antibody dilution buffer (1% BSA, 0.3% Triton X-100 in PBS). Primary antibodies used: rabbit NDRG1 polyclonal antibody (1:100, Cell Signaling), mouse Myc-tag antibody (#2276, CST), mouse TGM2 antibody (1:100 CUB 7402, Invitrogen), rabbit NDRG1 antibody (1:100 #9485, CST). On the next day, coverslips were washed three times in PBS and incubated with secondary antibodies in antibody dilution buffer for 1hr at RT. Secondary antibodies used: Alexa-Fluor 488 goat anti-mouse IgG (A11029, Invitrogen), Alexa-Fluor 546 goat anti-rabbit IgG (A11010, Invitrogen). Coverslips were then washed three times in PBS and incubated with bisbenzimide (1:5000, Hoechst stain, Sigma-Aldrich) for 10 min RT and cells were mounted. Microscopy was performed using a Zeiss LSM 880 Upright Confocal System, 63x PlanApo oil immersion objective and appropriate filter sets for Hoechst 405, Alexa Fluor 546, and Zen2009 software or Keyence BZ-X800 immunofluorescent microscopes. Images were quantified using CellProfiler [19].

Proximity Ligation Assay

The NDRG1-TGM2 interaction was detected using Duolink PLA Kit (DUO92102, Sigma-Aldrich) according to the manufacturer’s protocol. Cells were plated on coverslips, grown for 24 hours, then treated for 16 hours either with 2 mM HU, 1uM gemcitabine or vehicle. Cells were fixed with 4% PFA for 15 min at RT. Coverslips were permeabilized with 0.1% Triton-X 100 in PBS for 15 min at RT. Coverslips were blocked in Duolink Blocking Buffer. The coverslips were incubated with the primary rabbit NDRG1 polyclonal antibody (1:100, Cell Signaling) and mouse TGM2 antibody (1:100 CUB 7402, Invitrogen) diluted in Duolink Antibody Diluent overnight at +4°C. On the next day, cells were incubated with Duolink probes, ligation solution, and amplification solution. After the final washes with buffer B, cells were incubated with Phalloidin (#A12379, Invitrogen) for 40 min RT. Finally, cells incubated with bisbenzimide (1:5000, Hoechst stain, Sigma-Aldrich) for 10 min RT and cells were mounted. Microscopy was performed using a Zeiss LSM 880 Upright Confocal System, 63x PlanApo oil immersion objective and appropriate filter sets for Hoechst 405, Alexa Fluor 546, and Zen2009 software or Keyence BZ-X800 immunofluorescent microscopes. Confocal images were recorded in a Z-stack, further processed via ‘maximum intensity projection’ tool provided by Zen 2009 software. PLA foci were quantified using modified speckle analysis pipeline from CellProfiler [19].

AlphaFold Multimer Prediction

AlphaFold multimer prediction was performed using Cosmic² platform [20]. Human full length (1-394) NDRG1 (#Q92597) and human catalytic core (140-468) TGM2 (#P21980) protein sequences were obtained from Uniprot and uploaded to Cosmic². AlphaFold2 tool was selected, and task was created using full_dbs database and multimer model. Output models were analyzed using PyMol software.

BioID Pull-Down Experiments

NDRG1-myc-BioID2-pBABE2-puro construct was used to generate retroviral particles further used for transduction of SW1990 and HPAC cell lines. mycBioID2 construct alone (Addgene #80900) was used as a negative control. BioID labelling and sample preparation was performed exactly as described in Roux et al., 2018 [21]. Briefly, three 15 cm dishes for SW1990 (in 0% FBS), and five 15 cm dishes for HPAC (in 10% FBS), at 80% confluency were incubated with 50 μM biotin for 18 hours, washed twice with PBS, and cell pellets were collected. SW1990 BioID was performed in non-conditioned media (0% serum DMEM) or CAF-conditioned media. To obtain CAF-conditioned media, CAFs were plated onto 15 cm paltes, and at 90% confluency, media was changed from 10% serum DMEM to 0% serum DMEM. 3-4 days later, media was harvested, spun down, and sterile filtered through 0.22 μm filter. For the drug treatment, in the HPAC BioID, HPAC plates were treated with 2 μM gemcitabine overnight prior to sample collection. For SW1990, cells were lysed with RIPA buffer (BP-115, Boston BioProducts) containing benzonase (E1014, Sigma Aldrich), sonicated, and the lysate was centrifuged at 14,000 x g for 15 minutes to remove debris. For HPAC, cells were collected, and nuclear fractions were prepared by incubating the pellets in hypotonic buffer (20 mM Tris-HCL pH8.3, 10mM NaCl, 3mM MgCl2, protease inhibitor tablets) on ice for 10 minutes. Cells were scraped and 10% NP40 detergent was added to the cell suspension. Cells were vortexed and centrifuged at 14,000 × g for 10 minutes at +4°C to isolate cytoplasmic fraction (supernatant). Pellet was suspended in RIPA buffer (BP-115, Boston Bioproducts) containing complete protease inhibitor cocktail tablet and bensonaze, sonicated, and centrifuged at 14,000 × g for 10 minutes at +4°C to obtain nuclear fraction. Later, lysate concentration was measured, and 1 mg of total protein was used for streptavidin-pull down using Dynabeads^™ MyOne^™ Streptavidin C1 (65001, Thermo Fisher,). For HPAC, only nuclear fraction was used for pull down. The beads were incubated with lysate for 16 hours at +4°C, followed by a series of extensive washes and subsequent LC-MS/MS analysis. On-bead trypsin digestion and LC-MS/MS analysis was performed by BIDMC Mass Spectrometry core (for SW1990) and by the Taplin Biological Mass Spectrometry facility (HMS, Boston) for HPAC.

BioID SAINT Score Analysis

The analysis was performed using the The Contaminant Repository for Affinity Purification (CRAPome) as a resource for the analysis of interaction proteomics data (https://reprint-apms.org) as described in [22]. SAINT (Significance Analysis of INTeractome) is a statistical algorithm designed to assess the confidence of protein-protein interactions. It analyzes the spectral counts of proteins identified in the BioID pull down experiment, comparing them to negative control samples (e.g. data from samples expressing the BirA* tag). The SAINT score is a probability score, ranging from 0 to 1, where a higher SAINT score indicates a greater statistical likelihood of protein to be a true interactor of the bait protein (NDRG1), while a lower score suggests the protein might be a non-specific binder or contaminant [23]. The experimental data were formatted according to the file formatting tutorial required by the analysis pipeline and uploaded. Next, H.Sapiens was chosen as the organism, and proximity-dependent biotinylation as an experiment type, using spectral counts (SPC) as quantitation type. Later, the SAINT score analysis was performed, top scoring targets are presented in Figure 1. All the BioID hits are shown in Supplemental Table 1.

Figure 1. TGM2 is a novel binding partner of NDRG1.

A. Schematic showing proximity-dependent labeling (BioID) screen performed in pancreatic cancer cells (left) and BioID construct used expressing NDRG1-myc-BioID2 fusion protein (right). B. Immunofluorescence showing localization of NDRG1-myc-BioID2 fusion protein in SW1990 and HPAC cells. Scale bar: 100 μm. DAPI (blue), Myc (green), NDRG1 (red). C. Western blot showing expression levels of control-myc-BioID and NDRG1-myc-BioID constructs (Continued) in SW1990 and HPAC cells. NDRG1 blot shows endogenous NDRG1 (lower band) and the NDRG1-myc-BioID2 construct (upper band). D. Tables displaying highest-scoring interaction partners from BioID screen performed in whole cell extracts (WCE) of SW1990 cells (left), and nuclear fraction of HPAC cells untreated, or treated with 2 μM Gemcitabine or 2 mM hydroxyurea (HU) for 16 hours (right). Mass spectrometry data were filtered using the SAINT score analysis pipeline. E. HEK293T cells were transfected with indicated tagged overexpression constructs and GST pulldown was performed using HEK293T cell lysate to detect binding between GST-TGM2 and HA-NDRG1 and probed with the indicated antibodies. F. HEK293T cells were transfected with indicated tagged overexpression constructs and GST pulldown was performed using HEK293T cell lysate to detect binding between GST-NDRG1 and HA-TGM2. G. Endogenous co-immunoprecipitation with NDRG1 antibody was performed using SW1990 WCE, to detect NDRG1 interaction with TGM2. IgG rabbit control antibody was used as a negative control for immunoprecipitation (left). Quantification (right) shows TGM2 signal in the NDRG1 immunoprecipitate normalized to TGM2 input levels for each condition. Fold change was calculated relative to the negative control. Each dot represents an independent biological replicate (N=3), with bars indicating mean. H. Endogenous co-immunoprecipitation using NDRG1 antibody was performed in U2OS WCE, to detect NDRG1 interaction with TGM2. IgG rabbit control antibody was used as a negative control for immunoprecipitation (left). Quantification (right) shows TGM2 signal in the NDRG1 immunoprecipitate normalized to TGM2 input levels for each condition. Fold change was calculated relative to the negative control. Each dot represents an independent biological replicate (N=3), with bars indicating mean. I. Endogenous co-immunoprecipitation using NDRG1 antibody was performed in SW1990 nuclear fraction, to detect NDRG1 interaction with TGM2. IgG rabbit control antibody was used as a negative control (top); Quantification (middle) shows TGM2 signal in the NDRG1 immunoprecipitate normalized to the amount of immunoprecipitated NDRG1 for each condition. Fold change was calculated relative to the negative control. Quantification shown is from the representative experiment displayed. N=2 biological replicates; blot showing the controls for the fractionation of the SW1990 lysates (bottom). Input blot in the top panel for NDRG1 is a representative image from their nuclear fractionation shown in the bottom panel. J. Endogenous co-immunoprecipitation using NDRG1 antibody was performed in U2OS nuclear fraction, to detect NDRG1 interaction with TGM2. IgG rabbit control antibody was used as a negative control (top); Quantification (middle) shows TGM2 signal in the NDRG1 immunoprecipitate normalized to the amount of immunoprecipitated NDRG1 for each condition. Fold change was calculated relative to the negative control. Quantification shown is from the representative experiment displayed. N=2 biological replicates; blot showing the controls for the fractionation of U2OS lysates (bottom). Input blot in the top panel for NDRG1 is a representative image from their nuclear fractionation shown in the bottom panel. K. RNAseq expression data from The Cancer Genome Atlas (TCGA) PanCancer Atlas and Genotype-Tissue Expression (GTEx) projects were used to determine TGM2 expression in normal vs. tumor samples in pancreatic cancer (PAAD), kidney adenocarcinoma (KIRC), stomach adenocarcinoma (STAD), ovarian cancer (OV), and breast cancer (BRCA). One-way ANOVA was used to test statistical significance. Log₂FC cutoff = 0.6, P-value: <*0.01. L-P. Patients were split into two cohorts with the cutoff at the median expression level (high: top 50% expression vs low: bottom 50% expression) based on TGM2 TCGA RNAseq data, and patient overall survival is plotted. (L) Pancreatic cancer (PAAD), (M) Kidney adenocarcinoma (KIRC), (N) Stomach adenocarcinoma (STAD), (O) Ovarian Cancer (OV), (P) Breast cancer (BRCA). Log-rank test was used to test statistical significance.

Chemicals, Reagents and Drugs

Biochemicals and enzymes were of analytic grade and were purchased from commercial suppliers. Gemcitabine (S1714, Selleckchem) was reconstituted in DMSO and used at a final concentration of 1 μM. ZDON (616467, Sigma-Aldrich) was reconstituted in DMSO and used at a final concentration of 50 μM. GK921 (HY-12337, MedChemExpress) was reconstituted in DMSO and used at a final concentration of 1 μM. SYBR^™ Gold Nucleic Acid Gel Stain (S11494) was from Invitrogen. 5-chloro-2′-deoxyuridine (CldU) (C6891), 5-iodo-2′-deoxyuridine (IdU) (I7125), DAPI (D9542), Hydroxyurea (H8627), Thymidine (T1895), 5-ethynyl-2′-deoxyuridine (EdU) (900584), Laminin (L2020), Fibronectin F1141 were from Sigma. Propidium Iodide (11348639001) was from Roche. Restriction enzymes were from NEB. Small molecule SGK1 inhibitor BLU6340 was made by Blueprint Medicines, Cambridge, MA, USA.

DNA Fiber Assay

DNA fibers were performed as described previously [24]. Briefly, cells were sequentially pulsed with two thymidine analogues 50 μM CIdU (Sigma, C6891) and IdU 150 μM (Sigma, 17125) with 2xPBS washes in between. When fork stalling was required, cells were pulsed with CldU, washed with PBS, treated with 2 mM HU for 1 hour, washed with PBS and pulsed with IdU. Cells were then trypsinized and resuspended in PBS. 2.5 μL of cell suspension was pipetted on the top of SuperFrost plus slides (#48311-703, VWR). After 4 min, 7.5 μL spreading buffer (0.5% SDS, 200 mM Tris-HCl pH 7.4, 0.5 mM EDTA) was mixed with the cells for additional 2 minutes. Two glass slides were made per condition per experiment. Slides were tilted at 15 degrees to allow DNA fibers to run down the glass slide. Later, the fibers were air dried and then fixed in 3:1 methanol:acetic acid solution for 2 min, followed by the 2.5M HCl treatment for 30 minutes and 3% BSA/PBST blocking for 1 hour. Primary antibody incubation was performed for 1hour with anti-CIdU (ab6326 Abcam, 1:100) and anti-IdU (BD-347580, 1:20). Following three washes with PBS, fibers were stained with appropriate secondary Alexa-Fluor conjugated antibodies for 30 minutes, washed, air-dried and mounted. Slides were imaged with Zeiss LSM 880 Upright Confocal System, 40x or 63x PlanApo oil immersion objective. Measurement of replication structures was performed manually using Fiji. Pooled data from at least N=2 biological replicates are shown.

Drug treatments and IC50 calculations

For dose dependent drug treatment experiments, cells were plated onto 96 well plates (7.5x10³ per well) and were allowed to adhere. The next day, gemcitabine or HU was administered. For gemcitabine, a range of drug concentrations corresponding to a log 1/3.16 step down from the highest indicated concentration was administered using the Tecan D300e digital dispenser. For HU, a range of drug concentrations corresponding to a log 1/3.16 step down from the highest indicated concentration was administered manually. 48h or 72h after, end-point cell viability assays were performed with Presto Blue HS (Invitrogen), values were normalized to corresponding DMSO-treated or untreated control wells and plotted in GraphPad Prism software. Cell response to drug concentrations was analysed using non-linear regression to fit the data to log(drug) vs. response (variable slope) curve.

Flow Cytometry

SW1990 cells were plated on 6 cm plates in 10% DMEM. Cells were pulsed with 10 μM EdU for 1 hr before harvesting via trypsin digestion and fixation with ice cold 70% ethanol overnight. Later the cells were subjected to Click chemistry using Click-iT^™ EdU Alexa Fluor 647^™ Flow Cytometry Assay Kit according to manufacturer’s instructions (Thermo Scientific C10419). Cells were then incubated with propidium iodide (Roche) and with 192 μg/ml RNase A (Sigma-Aldrich) for 30 min at RT before analysis to measure total DNA content. Data acquisition was performed using Beckman Coulter’s CytoFLEX LX Flow Cytometer at BIDMC Flow Cytometry Core and later analysed using FlowJo software. After gating for single cells, the percentage of cells in S phase was determined by the percentage of cells that incorporated EdU. Gates were established on EdU negative cells to quantify the percentage of cells in G1 (low PI) and G2 (high PI).

Statistical Analysis

GraphPad PRISM 7-10 was used for statistical and visual analyses. Sample size and error bars are reported in the figure legends. P-values less than 0.05 were considered significant. For simple comparison of means, data were first checked for normality; if distributed normally data were then analyzed with paired or unpaired Student t test. If not distributed normally, data were analyzed using Mann-Whitney U test.

Ethics approval and consent to participate:

No animals were used in this study. No participants were included in this study. Patient data was downloaded from a publicly available cBioPortal database and does not require a separate IRB approval.

Results

Identification of a novel interaction between NDRG1 and TGM2

To uncover novel mechanisms by which NDRG1 impacts the DNA damage response (DDR), we performed a biotin proximity labeling experiment (BioID), which takes advantage of a promiscuous biotin ligase fused to the protein of interest (NDRG1) to allow for the biotinylation of endogenous proximal proteins, to identify putative binding partners of NDRG1 [21]. NDRG1 was cloned with the APEX fusion in its C-terminus (Figure 1A). We performed two independent BioID experiments in the pancreatic cancer cell lines, SW1990 and HPAC. After confirming that NDRG1-Myc-BioID2 was expressed at appropriate subcellular locations (Figure 1B) and migrated at the correct predicted molecular weight (Figure 1C), we proceeded with the BioID experiments. The SW1990 BioID was performed using whole cell extracts (WCE) in the absence or presence of CAF-conditioned media in low serum, whereas the HPAC BioID was performed in 10% serum using nuclear fractions of cells grown in the absence or presence of gemcitabine, a chemotherapeutic drug commonly used for PDAC. Biotin was added to the cells for 18 hours, after which the lysates were collected, and the biotinylated proteins captured with streptavidin beads and identified by mass spectrometry. Data were analyzed using the SAINT score analysis pipeline to identify high-confidence interactors [23]. To identify NDRG1-interacting partners shared between the two independent BioID experiments, we compared candidates with a SAINT score ≥ 0.9 across all experimental conditions using a Venn diagram analysis (Supplemental Figure 1A). This analysis identified three proteins shared between SW1990 and HPAC datasets, in addition to NDRG1. Notably, TGM2 was the only shared interactor whose interaction with NDRG1 was selectively increased under gemcitabine treatment in the HPAC BioID (Figure 1D). In contrast, TGM2 displayed a relatively high basal SAINT score in the SW1990 BioID dataset in the absence of treatment. Differences in culture conditions between the two experiments (serum-deprived versus serum-containing media) may contribute to the elevated basal interaction observed in SW1990 cells. Taken together, these data indicate that TGM2 may be regulated in a replication stress-associated manner, supporting its specific functional relevance in the replication stress response (Figure 1D). These data indicate that TGM2 interaction may be regulated in a replication stress-dependent manner. To validate the NDRG1-TGM2 interaction, we performed co-immunoprecipitations (co-IP) in HEK293T cells over-expressing tagged GST-TGM2 and HA-NDRG1 constructs and observed the pulldown of HA-NDRG1 with GST-TGM2 (Figure 1E). We were also able to pulldown HA-TGM2 using GST-NDRG1 in HEK293T cells (Figure 1F). Next, we performed endogenous co-immunoprecipitations to validate the endogenous interaction between TGM2 and NDRG1. Since HPAC cells had low levels of endogenous NDRG1 protein expression (Figure 1C), we selected two cancer cell lines with higher NDRG1 and TGM2 expression levels, and moderate to high basal levels of replication stress, for detailed mechanistic studies (U2OS and SW1990) (Figure 1G–J, Supplemental Figure 1B–C). Using co-immunoprecipitation, we were able to detect the interaction between TGM2 and NDRG1 in both the whole cell extract (WCE) and nuclear fractions of U2OS and SW1990 cells (Figure 1G–J), as well as in an additional pancreatic cancer cell line (HPAC), and ovarian cancer cell lines (CaOV-3, OVCAR-3 and KURAMOCHI) (Supplemental Figure 1D).

To investigate whether the expression or interaction of NDRG1 and TGM2 correlates with basal replication stress levels and cellular status, we examined the expression of these proteins and key replication stress markers (pATR, pChk1, pRPA) under basal and HU-treated conditions across a panel of cancer cell lines encompassing various tumor types (osteosarcoma, pancreatic, breast, and ovarian cancer) (Supplemental Figure 1B–D). The expression of NDRG1 and TGM2 showed a weak yet inconsistent trend with the basal expression levels of replication stress markers (i.e. correlation was not universal across all cell lines tested) (Supplemental Figure 1B–C). Furthermore, we observed no correlation between NDRG1 and TGM2 expression and chemoresistance towards fork stalling drugs, HU and gemcitabine (Supplemental Figure 1B–C, E–F). The NDRG1-TGM2 interaction was detected across various cell lines tested and did not correlate with replication stress signature, suggesting that the interaction does not strictly depend on the constitutive level of replication stress (Supplemental Figure 1D).

We next wanted to investigate the role of TGM2 in the context of solid tumors. We used the TCGA and GTEx database to determine the expression levels of TGM2 in normal and tumor samples in multiple solid tumors, including pancreatic adenocarcinoma (PAAD), kidney renal cell carcinoma (KIRC), stomach adenocarcinoma (STAD), ovarian cancer (OV), and breast cancer (BRCA) [25]. We found that TGM2 expression in pancreatic adenocarcinoma, kidney renal cell cancer and stomach adenocarcinoma was significantly enriched in tumor samples compared to normal samples (Figure 1K), whereas its expression was not significantly enriched in ovarian cancer or breast cancer (Figure 1K). Interestingly, pancreatic tissue had substantially lower baseline TGM2 expression compared to many other epithelial tissues (Figure 1K), which likely reflects the unique cellular composition and physiological function of the pancreas. TGM2 expression is known to be induced by stress- and injury-associated pathways, largely absent in healthy pancreatic tissue [26]. In contrast, pancreatic ductal adenocarcinoma is characterized by extensive desmoplasia, fibroblast activation, and ECM remodeling, which are conditions that strongly induce TGM2 expression [27]. Indeed, we also found that high TGM2 expression was correlated with poor overall survival (OS) in pancreatic cancer patients using both a median expression cutoff (Figure 1L) and a biologically informed cutoff defining the lowest 20% of tumors based on TGM2 expression levels observed in normal pancreatic tissue (Supplemental Figure 1G) [25]. Both stratification approaches revealed that high expression of TGM2 significantly correlated with poor OS in pancreatic cancer whereas it did not significantly correlate with OS in the other tumor types (Figure 1M–P). We did, however, observe a modest trend towards poor OS in the high TGM2 expression cohort in kidney renal cell cancer and stomach adenocarcinoma (Figure 1M–N) [25].

NDRG1-TGM2 interaction is enriched upon treatment with fork-stalling drugs and ECM signaling, and is regulated by SGK1-mediated NDRG1 phosphorylation

We recently discovered that NDRG1 is a novel DNA repair protein that has a direct role in DNA replication and repair by regulating replication fork homeostasis [7]. Since we observed that the interaction between NDRG1 and TGM2 was enriched upon gemcitabine treatment in our BioID experiment (Figure 1D), we hypothesized that the NDRG1-TGM2 interaction may be regulated by or linked to the replication stress response. To validate the enrichment of the NDRG1-TGM2 interaction under replication stress, we performed proximity ligation assays (PLA) in U2OS and SW1990 cells treated with HU or gemcitabine. We observed a significant increase in both nuclear and cytoplasmic PLA foci in HU and gemcitabine treated cells compared to untreated cells (Figure 2A–C). To further assess this interaction biochemically, we performed co-immunoprecipitation assays in U2OS and SW1990 cells treated with HU or gemcitabine. Quantification of co-immunoprecipitation across independent biological replicates revealed a consistent trend toward increased NDRG1-TGM2 interaction following replication stress, although this increase did not reach statistical significance with the available number of replicates (Figure 2D–E).

Figure 2. NDRG1-TGM2 interaction is enriched upon fork stalling and ECM signaling, and is dependent on SGK1-mediated NDRG1 phosphorylation.

A. Proximity ligation assay (PLA) using antibodies against NDRG1 and TGM2 was performed in U2OS and SW1990 cells treated with 2 mM HU and 1 μM gemcitabine for 24 hours. Scale bar: 10 μm. DAPI (blue), Phalloidin (green), PLA (white). B. CellProfiler was used to quantify nuclear (left) and cytoplasmic (right) PLA foci/cell in U2OS cells. Mann-Whitney test was used for significance, N=3 biological replicates. Red line represents mean. >65 cells quantified per condition. ****<0.0001. C. CellProfiler was used to quantify nuclear (left) and cytoplasmic (right) PLA foci/cell in SW1990 cells. Mann-Whitney test was used for significance, N=3 biological replicates. Red line represents mean. >141 cells quantified per condition. ****<0.0001, ***<0.001. D. Endogenous co-immunoprecipitation with NDRG1 antibody was performed in U2OS cells treated with 2 mM HU and 1 μM gemcitabine for 24 hours. IgG rabbit control antibody was used as a negative control for co-immunoprecipitation shown (left). Quantification (right) shows TGM2 signal in the NDRG1 immunoprecipitate normalized to TGM2 input levels for each condition. Each dot represents an independent biological replicate (N=3), with bars indicating mean. E. Endogenous co-immunoprecipitation with NDRG1 antibody was performed in SW1990 cells treated with 2 mM HU and 1 μM gemcitabine for 24 hours. IgG rabbit control antibody was used as a negative control for co-immunoprecipitation shown (left). Quantification (right) shows TGM2 signal in the NDRG1 immunoprecipitate normalized to TGM2 input levels for each condition. Each dot represents an independent biological replicate (N=3), with bars indicating mean. F. Endogenous co-immunoprecipitation with NDRG1 antibody was performed in SW1990 cells grown in polystyrene plates (adhered) or ultra-low adhesion plates (suspended), in the absence or presence of 20 μg/mL fibronectin (FN) and 20 μg/mL laminin (Lam) as indicated. IgG rabbit control antibody was used as a negative control for co-immunoprecipitation (left). Quantification (right) shows TGM2 signal in the NDRG1 immunoprecipitate normalized to the amount of immunoprecipitated NDRG1 for each condition. Fold change was calculated relative to the corresponding untreated condition within each culture condition (adhered or suspended). Quantification shown is from the representative experiment displayed. G. Endogenous co-immunoprecipitation with NDRG1 antibody was performed in SW1990 cells treated with 250 nM BLU6340 overnight. IgG rabbit control antibody was used as a negative control for co-immunoprecipitation. N=2 biological replicates. Quantification (right) shows TGM2 signal in the NDRG1 immunoprecipitate normalized to TGM2 input levels for each condition. Fold change was calculated relative to the untreated condition. Quantification shown is from the representative experiment displayed.

Because NDRG1 functions downstream of extracellular matrix (ECM) signaling via the integrin/Src/FAK/SGK1 signaling cascade, we next investigated whether this signaling axis also regulates the NDRG1-TGM2 interaction. Co-immunoprecipitation revealed that ECM signaling enhanced the interaction between NDRG1 and TGM2 (Figure 2F). Furthermore, to determine whether the SGK1-mediated phosphorylation of NDRG1 influenced the NDRG1-TGM2 interaction, we performed co-immunoprecipitation for NDRG1 and TGM2 in the presence of the SGK1 inhibitor, BLU6340. Inhibition of SGK1 using BLU6340 reduced the NDRG1-TGM2 interaction in SW1990 cells, suggesting that SGK1-mediated phosphorylation of NDRG1 contributes to the regulation of this interaction (Figure 2G). Interestingly, this regulatory effect of SGK1 inhibition on the NDRG1-TGM2 interaction was cell-type-specific. We observed a similar reduction in the interaction in the other pancreatic cancer cell line HPAC; however, in U2OS cells, which are an osteosarcoma cell line, SGK1 inhibition resulted in the enriched interaction between TGM2 and NDRG1, suggesting that upstream regulatory mechanisms influencing this interaction may differ across cellular contexts (Supplemental Figure 2A–B).

TGM2 is a novel regulator of DNA replication

Our previous work has shown that NDRG1 has a direct role in DNA replication and repair (bioRxiv 2025.01.22.634323). Specifically, NDRG1 was found at the replication fork, and depletion of NDRG1 resulted in the perturbation of normal replication and defective stalled fork restart (bioRxiv 2025.01.22.634323). Because both gemcitabine and HU cause fork stalling, and since we observed that the NDRG1-TGM2 interaction was enriched upon treatment with these fork stalling drugs, we hypothesized that the interaction might be important for NDRG1’s role on DNA replication and repair. Of note, a few recent studies have explored the potential role of TGM2 in DNA repair. For example, nuclear TGM2 was shown to be involved in DSB repair by interacting with TOPOIIα [13]. Another study demonstrated that TGM2 overexpression alleviated DNA DSBs and inhibited cell death [28]. However, to our knowledge, the direct role of TGM2 in DNA replication has not been explored before. Therefore, we first wanted to determine whether TGM2 played a role in normal DNA replication in unperturbed conditions. To do this, we performed a DNA fiber assay which allows for the investigation of replication fork dynamics at single-molecule resolution (Figure 3A). To assess the role of TGM2 in normal replication, we used TGM2 targeted siRNAs (Figure 3B). We then performed the DNA fiber assay in TGM2 knockdown and control cells by pulsing the cells with CldU followed by IdU. TGM2 knockdown cells had significantly shorter IdU track lengths compared to control cells, suggesting that TGM2 is required for normal replication fork progression (Figure 3C, E). To assess the role of TGM2 in stalled fork recovery, cells were pulsed with CldU, then treated with HU for 1 hour to induce fork stalling followed by a 40 min IdU pulse. TGM2 knockdown cells had significantly shorter IdU track lengths compared to control cells, suggesting that TGM2 is also required for stalled fork recovery (Figure 3D–E). To further validate these results, we performed the DNA fiber assay in TGM2 CRISPR knockout cell lines. We observed a similar replication defect in U2OS and SW1990 CRISPR knockout cell lines generated using three independent guide RNAs (Figure 3F–J).

Figure 3. TGM2 is required for normal replication fork progression and stalled fork recovery.

A. Schematic showing DNA fiber assay protocol (left) and pulse labeling scheme using 5-chloro-2′-deoxyuridine (CldU) and 5-Iodo-2’-deoxyuridine (IdU) (right). To measure normal replication (reg), cells were pulsed with CldU for 20 minutes, followed by IdU for 40 minutes. To measure stalled fork recovery, cells were pulsed with CldU for 20 minutes, followed by HU for 1 hour, followed by IdU for 40 minutes. B. Western blot showing knockdown efficiency for cells treated with control and TGM2-targeted siRNA used in C-D. C. DNA fiber assay measuring normal replication (reg) in U2OS cells transfected with control or TGM2-targeted siRNA. IdU track lengths were quantified for >200 double-labelled fibers. D. DNA fiber assay measuring stalled fork recovery (HU treated) in U2OS cells treated with control or TGM2-targeted siRNA. IdU track lengths were quantified for >200 double-labelled fibers. E. Representative fiber images from C-D. F. Western blot showing expression of TGM2 in U2OS and TGM2 CRISPR knockout (KO) cell lines. G. DNA fiber assay measuring normal replication (reg) in U2OS TGM2 KO cells. IdU track lengths were quantified for ≥135 double-labelled fibers. H. DNA fiber assay measuring stalled fork recovery (HU) in U2OS TGM2 knockout (KO) cells generated using CRISPR. IdU track lengths were quantified for ≥135 double-labelled fibers. I. DNA fiber assay measuring normal replication (reg) in SW1990 TGM2 KO cells generated using CRISPR. IdU track lengths were quantified for ≥200 double-labelled fibers. J. DNA fiber assay measuring stalled fork recovery (HU) in SW1990 TGM2 KO cells. IdU track lengths were quantified for ≥160 double-labelled fibers. K. Western blot showing expression levels of TGM2 in U2OS and SW1990 TGM2-WT, TGM2-NLS, TGM2-NES rescue cells. L. Immunofluorescence images showing localization of TGM2-WT, TGM2-NLS and TGM2-NES in U2OS cells, which were stained with anti-TGM2 (red) and DAPI (blue). Scale bar: 2 μm. M. DNA fiber assay measuring normal replication in U2OS rescue cells. IdU track lengths were quantified for ≥200 double-labelled fibers. N. DNA fiber assay measuring normal replication in SW1990 rescue cells. IdU track lengths were quantified for ≥200 double-labelled fibers. For all DNA fiber experiments, Mann-Whitney test was applied to determine significance. ****<0.0001, ***<0.001, **<0.01, n.s.: non-significant. Red line represents median IdU track length. Pooled data from at least N=2 biological replicates are shown.

To investigate whether changes in cell cycle distribution might be responsible for the observed effect of TGM2 on replication, we first assessed whether TGM2 depletion influenced cell cycle distribution. Flow cytometry analysis, performed on unsynchronized cells, revealed no significant difference in the distribution of cells throughout the cell cycle phases between control and TGM2 KO cells (Supplemental Figure 3A). Additionally, we used a double thymidine block to synchronize cells at the G1/S boundary and performed Western blot analysis for cell cycle markers to determine the speed of progression through S phase. We found that control and TGM2 KO cells progressed through S phase at similar speeds, further confirming that TGM2 does not influence cell cycle timing (Supplemental Figure 3B). Furthermore, we performed the DNA fiber assay to directly assess the replication fork speed of control and TGM2 KO cells at each time point after release from the double thymidine block. We found that TGM2 KO cells exhibited slower replication compared to control cells throughout S phase, suggesting that the effect of TGM2 on replication speed were independent of its effects on cell cycle distribution or timing (Supplemental Figure 3C).

Given our observation that the loss of TGM2 resulted in slower replication and impaired stalled fork recovery, we hypothesized that TGM2 depletion might also sensitize cells to fork-stalling drugs. However, we observed no significant difference in sensitivity towards HU or gemcitabine in SW1990 and U2OS TGM2 KO cells compared to control cells (Supplemental Figure 3D–E). This suggests that, despite its role in maintaining replication fork homeostasis, TGM2 does not influence chemoresistance towards fork stalling drugs in these cell lines.

Since TGM2 is known to localize both to the cytoplasm and nucleus and have distinct functions depending on its sub-cellular localization, we wanted to understand whether the localization of TGM2 was important for its role in DNA replication. We cloned TGM2 rescue constructs that either had a c-Myc nuclear localization (NLS) sequence (PAAKRVKLD) or a leucine-rich nuclear export signal (NES) sequence (LPPLERLTL) at its C-terminus and added these to the TGM2 KO cells and analyzed these cells by DNA fiber assay. Only the TGM2-WT and TGM2-NLS rescue constructs were able to rescue the replication defect phenotype, while the TGM2-NES rescue cells phenocopied the TGM2 KO cells (Figure 3K–N). Taken together, these data suggest that TGM2 acts in the nucleus to impact both, normal fork progression and stalled fork recovery.

TGM2 catalytic activity is required for its role in DNA replication

TGM2 is an enzyme that catalyzes transamidation reactions that can result in the formation of covalent bonds between peptide bound glutamine residues and primary amines [29]. TGM2 has been shown to act on its substrates both outside the cell, such as collagen and fibronectin, and inside the cell. Some studies have demonstrated that TGM2 transamidating activity is also active in the nucleus [30, 31]. Since TGM2 and NDRG1 interact in the nucleus, and nuclear TGM2 is required for its role in DNA replication fork progression and stalled fork recovery, we hypothesized that the catalytic activity of TGM2 could be required for its role in DNA replication. To test this hypothesis, we treated cells with TGM2 catalytic inhibitors, ZDON or GK921, and subjected the cells to DNA fiber assay (Figure 4A). Treatment with the TGM2 catalytic inhibitors resulted in significant shortening of IdU track lengths under normal replication conditions and HU-induced stalled fork conditions (Figure 4B–E). To further validate these results and rule out any off-target effects of these inhibitors, we cloned catalytically inactive C277S and partially inactive Y516F TGM2 mutant constructs, and generated stable rescue cell lines expressing these mutant TGM2 constructs in the TGM2 KO background for U2OS and SW1990 cells (Figure 4F–G). Using these cell lines, we performed the DNA fiber assay and observed that the TGM2 catalytic mutant rescue cells were not able to rescue the replication defect phenotype (Figure 4H–K). TGM2 has several Ca²⁺ binding sites in its catalytic domain, and binding of Ca²⁺ is required to induce the catalytically active conformation of TGM2 [32]. Therefore, as an orthogonal method to validate whether the catalytic activity of TGM2 was required for its role in DNA replication, we mutated two of the Ca²⁺ binding sites of TGM2 and assessed the replication phenotype of these mutants in U2OS and SW1990 cells (Figure 4L). The Ca²⁺-binding mutants were unable to rescue the replication phenotype, further suggesting that the catalytic activity of TGM2 was required for its role in DNA replication (Figure 4M–N).

Figure 4. TGM2 catalytic activity is required for its role in DNA replication.

A. Table showing the mechanism of action for TGM2 catalytic inhibitors used in experiments shown in 4B-E. B. DNA fiber assay measuring normal replication in U2OS cells treated with TGM2 inhibitors. Cells were treated with 50 μM ZDON and 1 μM GK921 overnight prior to pulsing with nucleotide analogs as shown in Figure 3A. IdU track lengths were quantified for ≥176 double-labelled fibers. C. DNA fiber assay measuring stalled fork recovery in U2OS cells treated with TGM2 inhibitors as in B. IdU track lengths were quantified for >200 double-labelled fibers. D. DNA fiber assay measuring normal replication (reg) in SW1990 cells treated with TGM2 inhibitors as in B. IdU track lengths were quantified for >200 double-labelled fibers. E. DNA fiber assay measuring stalled fork recovery (HU) in SW1990 cells treated with TGM2 inhibitors as in B IdU track lengths were quantified for >200 double-labelled fibers. F. Table showing catalytic transamidation activity of TGM2 mutants used in Fig. 4G–K. G. Western blot showing expression levels of TGM2 in U2OS and SW1990 TGM2 catalytic mutant rescue cells. GAPDH was used as a loading control. H. DNA fiber assay measuring normal replication in U2OS TGM2 catalytic mutant rescue cells. IdU track lengths were quantified for ≥152 double-labelled fibers. I. DNA fiber assay measuring stalled fork recovery in U2OS TGM2 catalytic mutant rescue cells. IdU track lengths were quantified for ≥178 double-labelled fibers. J. DNA fiber assay measuring stalled fork recovery (HU) in SW1990 TGM2 catalytic mutant rescue cells as in H. IdU track lengths were quantified for ≥200 double-labelled fibers. K. DNA fiber assay measuring normal replication in SW1990 TGM2 catalytic mutant rescue cells as in I. IdU track lengths were quantified for >200 double-labelled fibers. L. Western blot showing expression levels of TGM2 in U2OS and SW1990 TGM2 calcium-binding mutant rescue cells. GAPDH was used as a loading control. M. DNA fiber assay measuring normal replication (reg) in U2OS TGM2 calcium-binding mutant rescue cells. IdU track lengths were quantified for ≥200 double-labelled fibers. N. DNA fiber assay measuring normal replication (reg) in SW1990 TGM2 calcium-binding mutant rescue cells as in M. IdU track lengths were quantified for >200 double-labelled fibers. For all DNA fiber experiments, Mann-Whitney test was applied to determine significance. ****<0.0001, ***<0.001, **<0.01, *<0.05, n.s.: non-significant. Red line represents median IdU track length. Pooled data from at least N=2 biological replicates are shown. O. Endogenous co-immunoprecipitation with NDRG1 antibody was performed in SW1990 cells treated with 50 μM ZDON or 1 μM GK921 overnight. IgG rabbit control antibody was used as a negative control for co-immunoprecipitation. P. Endogenous co-immunoprecipitation with NDRG1 antibody was performed in HPAC cells treated with 50 μM ZDON or 1 μM GK921 overnight. IgG rabbit control antibody was used as a negative control for co-immunoprecipitation.

To assess the effect of the TGM2’s catalytic activity on the NDRG1-TGM2 interaction, we performed co-immunoprecipitation for NDRG1 and TGM2 in the presence of the TGM2 inhibitors, GK921 or ZDON, in SW1990 and HPAC cells. We found that treatment with the allosteric small molecule inhibitor, GK921, had very little to no effect on the interaction (Figure 4O–P). In contrast, treatment with the competitive inhibitor, ZDON, reduced the NDRG1-TGM2 interaction (Figure 4O–P). Since ZDON is a peptide-based inhibitor that irreversibly binds with the active site cysteine (C277) of TGM2, the reduction in binding observed might suggest that the inhibitor masks a potential binding interface between TGM2 and NDRG1 that may be located near the TGM2 active site. Because GK921 is an allosteric inhibitor that binds away from the active site and did not affect the interaction, these data suggest that the catalytic transamidase activity of TGM2 itself may not be involved in regulating its interaction with NDRG1.

To determine whether TGM2 catalytic activity was important for chemoresistance, we performed cell viability assays and assessed IC50 responses of SW1990 cells towards HU in the presence of the TGM2 catalytic inhibitors, ZDON and GK921. ZDON had no observable effect on cell viability at any of the concentrations tested. In contrast, GK921 displayed inherent toxicity starting from 0.5 μM, but even at non-toxic concentrations, GK921 did not sensitize cells towards HU (Supplemental Figure 4A–B). Given our observation that TGM2 is important for replication fork homeostasis, we next investigated whether loss of TGM2 would create a synthetic lethality in the absence of key DDR kinases, ATM and ATR. We did not observe any sensitization towards HU in the presence of TGM2 inhibitors combined with either ATM or ATR inhibitors (Supplemental Figure 4C–F). Finally, we tested for synergistic sensitization between TGM2 inhibition and SGK1 inhibition and again observed no sensitization toward HU (Supplemental Figure 4G–H). Altogether, these data suggest that TGM2 catalytic transamidase activity does not influence chemoresistance towards fork stalling drugs.

NDRG1 and TGM2 likely act in the same pathway to influence DNA replication

Given our observation that NDRG1 and TGM2 interact and are both required for normal replication and stalled fork recovery, we wanted to explore whether these proteins function together or independently to influence DNA replication. To test the functional relationship between NDRG1 and TGM2, we performed the DNA fiber assay in NDRG1 KO cells that were treated with control or TGM2-targeted siRNA (Figure 5A). We observed that IdU track lengths were shortened to the same extent in both siTGM2-treated cells and NDRG1 KO cells (Figure 5B–C). When TGM2 and NDRG1 were both depleted, we observed no further IdU track length shortening compared to the single depletion conditions (Figure 5B–C). We confirmed this result by performing the reciprocal experiment in TGM2 KO cells treated with NDRG1-targeted siRNA and observed the same phenotype where the depletion of both TGM2 and NDRG1 did not result in any further IdU track length shortening compared to the single depletion conditions (Figure 5D–I). This suggests that TGM2 and NDRG1 function in the same non-redundant pathway to modulate DNA replication.

Figure 5. TGM2 and NDRG1 act in the same pathway to influence DNA replication.

A. Western blot showing NDRG1 KO and TGM2 knockdown efficiency of cells used for fiber assays in B-C. GAPDH was used as a loading control. B. DNA fiber assay measuring normal replication (reg) in U2OS AAV (control) or NDRG1 KO cells treated with control or TGM2-targeted siRNA. IdU track lengths were quantified for >200 double-labelled fibers. C. DNA fiber assay measuring stalled fork recovery (HU) in U2OS AAV (control) or NDRG1 KO cells treated with control or TGM2-targeted siRNA. IdU track lengths were quantified for ≥193 double-labelled fibers. D. Western blot showing TGM2 KO and NDRG1 knockdown efficiency of cells used for fiber assay in E-F. GAPDH was used as a loading control. E. DNA fiber assay measuring normal replication (reg) in U2OS AAV (control) or TGM2 KO cells treated with control or NDRG1-targeted siRNA. IdU track lengths were quantified for >200 double-labelled fibers. F. DNA fiber assay measuring stalled fork recovery (HU) in U2OS AAV (control) or TGM2 KO cells treated with control or NDRG1-targeted siRNA. IdU track lengths were quantified for >200 double-labelled fibers. G. Western blot showing TGM2 KO and NDRG1 knockdown efficiency of cells used for fiber assay in H-I. GAPDH was used as a loading control. H. DNA fiber assay measuring normal replication (reg) in SW1990 AAV (control) or TGM2 KO cells treated with control or NDRG1-targeted siRNA. IdU track lengths were quantified for ≥173 double-labelled fibers. I. DNA fiber assay measuring stalled fork recovery (HU) in SW1990 AAV (control) or TGM2 KO cells treated with control or NDRG1-targeted siRNA. IdU track lengths were quantified for ≥200 double-labelled fibers. J. Alphafold2 predicted multimer model generated using Cosmic² for human TGM2 catalytic domain (residues 140-468) shown in green, and human NDRG1 full length protein shown in blue. Cosmic² output models were visualized in PyMol. Left: Zoomed out to visualize whole multimer with TGM2 catalytic triad colored in red and white box highlighting NDRG1-TGM2 interaction site. Right: Zoomed in to white box to focus on NDRG1-TGM2 interaction site. K. Endogenous co-immunoprecipitation using NDRG1 antibody in U2OS cells expressing V5-tagged WT, Q117A, or Q117E NDRG1 constructs. N=2 biological replicates. L. Endogenous co-immunoprecipitation using NDRG1 antibody in SW190 cells expressing V5-tagged WT, Q117A, or Q117E NDRG1 constructs. M. DNA fiber assay measuring normal replication (reg) in U2OS NDRG1 Q117 mutant rescue cells. IdU track lengths were quantified for ≥124 double-labelled fibers. N. DNA fiber assay measuring normal replication (reg) in SW1990 NDRG1 Q117 mutant rescue cells as in M. IdU track lengths were quantified for ≥177 double-labelled fibers. For all DNA fiber experiments, Mann-Whitney test was applied to determine significance. ****<0.0001, ***<0.001, **<0.01, n.s.: non-significant. Red line represents median IdU track length. Pooled data from at least N=2 biological replicates are shown. O. Representative immunofluorescence images of SW1990 cells expressing V5-tagged WT, Q117A, or Q117E NDRG1 constructs. Cells were stained with anti-TGM2 (green), anti-NDRG1 (red), and DAPI (blue). Scale bar: 10 μm. P. CellProfiler was used to quantify the integrated intensity of TGM2 in the nucleus and cytoplasm, and the ratio of nuclear:cytoplasmic TGM2 was calculated and plotted for each cell. N=2 biological replicates. >270 cells quantified per condition. ****<0.0001, **<0.01, *<0.05, n.s.: non-significant. Red line represents median.

To determine whether the physical interaction between NDRG1 and TGM2 was required for their role in DNA replication, we aimed to specifically disrupt the binding between NDRG1 and TGM2 and assess the effect on DNA replication. To do this, we used AlphaFold2 to predict the binding site between NDRG1 and TGM2 and identified a putative interaction site between TGM2-M252, and NDRG1-Q117 (Figure 5J). The fact that the M252 residue is located within the catalytic core of TGM2 aligned with our earlier finding that the active site inhibitor, ZDON, reduced interaction between TGM2 and NDRG1 (Figure 4O–P). Since the catalytic activity of TGM2 was important for its function in replication fork homeostasis (Figure 4B–N), we avoided mutating M252 to preserve its enzymatic function. Instead, we focused on mutating the Q117 residue on NDRG1. We generated stable SW1990 and U2OS cells expressing NDRG1-WT, NDRG1-Q117A or NDRG1-Q117E constructs in the NDRG1 KO background. We then used these cells to perform co-immunoprecipitation experiments and found that the NDRG1-Q117A mutant was still able to bind to TGM2, whereas the charge-altering NDRG1-Q117E mutant disrupted the binding between NDRG1 and TGM2 (Figure 5K–L). We then subjected these cells to the DNA fiber assay. The binding-competent NDRG1-Q117A mutant successfully rescued the DNA replication phenotype, whereas the binding-deficient NDRG1-Q117E mutant was unable to rescue the phenotype and phenocopied the NDRG1 KO cells (Figure 5M–N). Despite the functional effect of the interaction on replication fork dynamics, disrupting the interaction did not influence chemoresistance towards fork stalling drugs. The NDRG1-Q117E mutant cells displayed no significant change in IC50 to HU or gemcitabine compared to NDRG1-WT or NDRG1-Q117A cells (Supplemental Figure 5). These data collectively suggest that the physical interaction between NDRG1 and TGM2 is likely required for their roles in maintaining normal DNA replication efficiency and stalled fork recovery.

Since we showed earlier that TGM2 nuclear localization was critical for its role in replication fork homeostasis, we asked whether TGM2’s interaction with NDRG1 influenced its subcellular localization. Therefore, we assessed the nuclear localization of TGM2 in cells expressing the NDRG1-Q117 mutants. In cells expressing the NDRG1-Q117E mutant, nuclear TGM2 was significantly reduced compared to in cells expressing NDRG1-WT or the NDRG1-Q117A mutant (Figure 5O–P). These data suggest that the physical interaction between NDRG1 and TGM2 regulates the nuclear import or nuclear retention of TGM2.

Overall, our study described for the first time a novel function of TGM2 in DNA replication and reveals the importance of an unexpected protein interaction between TGM2 and NDRG1 in maintaining DNA replication efficiency. These findings reveal a stress-responsive mechanism that supports replication homeostasis in cancer cells and advance our understanding of how extracellular signals are integrated with replication and repair pathways.

Discussion

We recently discovered NDRG1 as a protein involved in resolving chemotherapy-induced DNA damage and replication stress. Here, we aimed to further understand the mechanistic details of NDRG1-mediated effects on DNA replication by studying a novel NDRG1 binding partner, TGM2. We found that the NDRG1-TGM2 interaction was enriched upon treatment with HU and gemcitabine, suggesting that the interaction may play an important role in regulating the response to replication fork stalling. We also found that the interaction is regulated by ECM signaling and SGK1-mediated NDRG1 phosphorylation at Thr346 in pancreatic cancer derived cell lines. Although TGM2 has been shown to play a role in DSB repair, TGM2 has never been studied in the context of DNA replication and stalled fork recovery. We observed slower replication fork progression and stalled fork recovery in the absence of TGM2, showing that TGM2 plays a role in replication fork homeostasis. We further demonstrate that this novel function of TGM2 is dependent on its catalytic activity, and its nuclear localization. Our data show that NDRG1 and TGM2 act in the same pathway to influence DNA replication and show that disrupting the physical interaction between NDRG1 and TGM2 leads to defects in DNA replication, suggesting that NDRG1 and TGM2 could both contribute to joint regulation of DNA replication. Finally, we show that disruption of the NDRG1-TGM2 interaction results in reduced nuclear localization of TGM2, suggesting that NDRG1 binding may facilitate or stabilize TGM2 nuclear localization. This provides a potential mechanistic link between the physical interaction of these proteins and the nuclear functions of TGM2 in DNA replication.

We identified TGM2 as a putative binding partner of NDRG1 through two independent BioID experiments performed in two different pancreatic cancer cell lines, SW1990 and HPAC. TGM2 was the only interactor that was identified as a high scoring interaction partner across both experiments, that was upregulated in response to gemcitabine treatment in the HPAC BioID. The interaction was basally high in the SW1990 but not in the HPAC BioID, and we noted that this could have been attributed to the differences in growth conditions between the two experiments (serum deprivation in the SW1990 and complete media in the HPAC experiments). Additionally, intrinsic differences between the two pancreatic cancer cell lines may also contribute to differences in baseline replication stress and stress responses. Although both SW1990 and HPAC harbor oncogenic KRAS (G12D) mutations and TP53 alterations, SW1990 carries a TP53 deletion resulting in complete loss of p53 expression, whereas HPAC harbors a TP53 missense mutation, which may still produce mutant p53 proteins with context-dependent activities. These differences in TP53 status may therefore contribute to cell-line specific regulation of replication stress pathways and could potentially partially explain the differences in baseline and stress-responsive interaction observed between NDRG1 and TGM2 in the SW1990 and HPAC BioID experiments.

TGM2 localizes to multiple subcellular compartments, and we find that the nuclear localization of TGM2 is required for its function in DNA replication. In response to replication stress-inducing drugs, we observed increased interaction between NDRG1 and TGM2 in both the cytoplasmic and nuclear compartments. The enhanced cytoplasmic interaction may represent an upstream regulatory step rather than a direct contribution to DNA replication. One possibility is that cytoplasmic binding of NDRG1 may facilitate the trafficking of TGM2 to the nucleus. Consistent with this idea, disruption of the NDRG1-TGM2 interaction reduced the nuclear localization of TGM2, suggesting that NDRG1 may promote or stabilizes the nuclear pool of TGM2 of TGM2 under replication stress. Thus, the cytoplasmic interaction may reflect an early regulatory step that enables efficient nuclear localization and function of TGM2 during the replication stress response. To determine whether TGM2’s effect on DNA replication could be attributed to change in cell cycle distribution, we performed cell cycle analysis and DNA fiber assays on synchronized cells. We observed that TGM2 depletion does not influence cell cycle distribution, and that TGM2 KO cells have slower replication at timepoints all throughout S-phase, suggesting that the replication defect observed in TGM2 KO cells is not attributable to alterations in cell cycle distribution or S-phase progression, but instead reflects defects in replication fork dynamics.

TGM2 is known to act on substrates that contain glutamine and primary amine residues. Some known extracellular substrates include ECM proteins such as fibronectin, collagen, and laminin, and the most well-established nuclear substrate for TGM2 are histones. TGM2 has been shown to modify all four mammalian core histones (H2A, H2B, H3, H4) through its catalytic crosslinking activity, resulting in the incorporation of primary amines into histones [11]. We observed that TGM2 catalytic activity is required for its effect on DNA replication using three orthogonal approaches: small molecule TGM2 inhibitors, mutation of the TGM2 catalytic residues, and mutation of the TGM2 calcium binding sites (Fig. 4). The substrate of TGM2 catalytic activity that is involved in regulating DNA replication still remains unknown, and it will be interesting to identify and further characterize the nuclear substrate(s) of TGM2 that are required for its role in DNA replication. It is possible that the substrate(s) of TGM2’s DNA replication function could be histones since it is known that one mechanism of fork stabilization during replication stress involves chromatin compaction and histone modifications [33]. Another possibility is that NDRG1 could be a potential substrate of TGM2. Our Alphafold model for the predicted NDRG1-TGM2 multimer highlights a putative interaction between TGM2-M252 and NDRG1-Q117. Since TGM2 is known to act on substrate proteins containing glutamine residues, it is possible that NDRG1 could be a potential substrate protein of TGM2, and TGM2 catalytic activity may result in a post-translational modification on NDRG1 that could regulate its function. Further investigation into the nuclear TGM2 substrate(s) that are important for regulating DNA replication will provide new insights into TGM2 biology.

TGM2 has been explored as a potential therapeutic target for cancer since it’s been shown to be involved in several pathways including epithelial to mesenchymal transition, metastasis, drug resistance, and angiogenesis. Furthermore, TGM2 is highly expressed in several cancer types including pancreatic cancer, gastric cancer and kidney cancer, and its expression has been associated with poor prognosis in these cancers [34]. However, the role of TGM2 in cancer has been controversial with some studies demonstrating that TGM2 has a tumor promoting role and other studies showing that TGM2 has a tumor-suppressive function. It is likely that different cell types and different cancers respond differently to TGM2 inhibition [34]. In this study, we performed the majority of our mechanistic experiments using a pancreatic cancer cell line (SW1990), and osteosarcoma cell line (U2OS). We observed that perturbation of TGM2 similarly impaired replication fork progression and stalled fork recovery in both cell lines, suggesting that the role of TGM2 in regulating DNA replication may extend beyond pancreatic cancer. However, we also observed lineage dependent differences in the upstream regulation of this interaction. Specifically, in pancreatic cancer cell lines, SW1990 and HPAC, the inhibition of SGK1-mediated phosphorylation reduced the NDRG1-TGM2 interaction, whereas in U2OS cells, we observed the opposite response to SGK1 inhibition. These results suggest that while the downstream function of the NDRG1-TGM2 interaction in DNA replication may be conserved, the upstream signaling pathways regulating NDRG1-TGM2 complex assembly may be cell-type specific. We further examined the relationship between replication stress markers and expression of TGM2 across a panel of cell lines, and observed a general trend where cell lines with higher expression of replication stress markers also had higher expression of TGM2 (Supplementary Figure 1B–C), suggesting that TGM2 expression may be associated with higher replication stress cellular states (Supplemental Figure 1B–C). In addition, analysis of normal tissues revealed substantial tissue-specific differences in basal TGM2 expression (Figure 1K), further highlighting that TGM2 function and regulation are likely context dependent.

TGM2 inhibitors are currently in development and are showing promise as potential anti-cancer therapeutic agents [34]. Our study reveals a novel function of TGM2 in regulating DNA replication under conditions of replication stress and highlights the need for further investigation into how this function contributes to replication fork stability and genome maintenance.

Implications:

This study reveals a previously unrecognized nuclear function for NDRG1 and TGM2 in regulating DNA replication fork stability and recovery and uncovers a stress-responsive mechanism that supports replication homeostasis in cancer cells and advance our understanding of how extracellular signals are integrated with replication and repair pathways.

Data availability:

All data supporting the findings in this study are available within the Article, it’s Supplementary Information, Source Data files and from the corresponding author upon reasonable request. Cell lines from ATCC and Addgene plasmids are under materials transfer agreements. RNAseq datasets for figure 1K and figure 1L–P are from TCGA PanCancer Atlas and GTEx studies (http://xena.ucsc.edu).