Targeting IGF Perturbs Global Replication through Ribonucleotide Reductase Dysfunction

Type I insulin-like growth factor receptor (IGF1R) is a receptor tyrosine kinase (RTK) expressed by most common cancers, and promotes proliferation, cell survival, and invasion. These effects are mediated via multiple signalling pathways including PI3K–AKT–mTOR and MAPK kinase–extracellular signal-regulated kinases (MEK-ERK; ref. 1). IGF1R is required for cellular transformation by oncogenes including SV40 and RAS, and can itself induce transformation when overexpressed (2). We and others found that IGF1R overexpression associates with clinical radioresistance in breast and prostate cancer (3, 4), with evidence that IGF1R targeting sensitizes to DNA-damaging cytotoxic drugs and ionizing radiation (IR) in vitro and in vivo (5–7). We further reported that IGF1R inhibition delays repair of IR-induced DNA double-strand breaks (DSB) and inhibits homologous recombination (HR) and nonhomologous end-joining (NHEJ) in repair reporter assays (7). The ability of IGF1R to regulate repair of exogenous DNA damage appears independent of its well-characterized ability to protect from apoptosis (7), and may require IGF-regulated interaction between the docking protein insulin receptor substrate-1 (IRS-1) and RAD51 (8), and IGF-induced activation of PI3K–AKT–mTOR and MEK–ERK pathways, both reported to mediate radioresistance (9, 10).

While this evidence supports a role for the IGF axis in the response to IR-induced DNA damage, IGFs have not previously been implicated in maintaining replication integrity in the absence of exogenous damage. We previously reported that IGF1R–inhibited or -depleted tumor cells accumulate γH2AX foci, representing initial evidence of endogenous DNA lesions (11), but their origin was unknown. Foci formed by γH2AX indicate H2AX phosphorylation at damage sites by PI3K-like kinases ataxia telangiectasia mutated (ATM), ataxia telangiectasia and Rad3-related protein (ATR), and DNA-dependent protein kinase (12). ATM is activated by diverse DNA structures, in addition to IR-induced DSBs, while ATR responds to extended single-stranded DNA (ssDNA).

Here, we identify the source and functional consequences of endogenous DNA lesions in cells where IGF signaling is suppressed. We discover unanticipated regulation by IGF1R of ribonucleotide reductase (RNR) function and dNTP supply; as a consequence, IGF axis blockade compromises completion of global DNA replication, delaying replication fork progression and allowing under-replicated DNA to progress through mitosis. We also reveal that replication stress in IGF-inhibited cells is maintained at a tolerable level by ATM, preventing conversion of single-strand breaks (SSBs) into DSBs. Thus, replication stress in IGF-inhibited cells associates with a state of reciprocal codependence on ATM that can be exploited in therapy.

Breast cancer cell lines HCC1143, MCF7, T47D, and ZR-75–1 were obtained in 2015 from Anthony Kong (Institute of Cancer and Genomic Sciences, University of Birmingham, Edgbaston, Birmingham, United Kingdom), HCT15 cells in 2018 from Walter Bodmer (Department of Oncology, University of Oxford, Oxford, United Kingdom), BXPC-3 in January 2020 from Eric O'Neil (Department of Oncology, University of Oxford, Oxford, United Kingdom), ATM WT and ATM-null H322 cells in 2019 from Sebastian Nijman, Nuffield Department of Medicine, University of Oxford (Oxford, United Kingdom) and KPL1 (2016), LoVo, SK-CO-1, immortalized ATM WT MRC5-SV2 in 2018 from the European Collection of Authenticated Cell Cultures and ATM^−/− AT5BIVA fibroblasts in 2018 from The Coriell Institute for Medical Research, MiaPaCa-2 in March 2020 from the ATCC, paired ATM-proficient and -deficient MiaPaCa-2 in March 2020 from Duncan Jodrell and Frances Richards, University of Cambridge (Cambridge, United Kingdom). Cell cultures were cryopreserved at early passage and used within 20 passages. Cell lines were Mycoplasma-free (MycoAlert, Lonza Rockland Inc.) and cell line identity was authenticated by STR genotyping (Eurofins Medigenomix Forensik GmbH).

Gene knockdowns were performed as described in refs. 7 and 13 using 2 to 3 separate siRNAs to deplete each target, including 3 IGF1R siRNAs (siIGF1R_1, _2, _3; where no numbering is stated, siIGF1R_1 was used.

Cell viability, doubling time, spheroid, and cell death assays were performed as described in Supplementary Methods.

Western blotting, immunofluorescence and flow cytometry were performed as described (7, 13) using antibodies in Supplementary Table S1.

DNA fiber assays were performed as detailed in Supplementary Methods. Fork velocity was converted to kb/minute based on tract length (μm) and duration of incubation with 5-chloro-2′deoxyuridine (CIdU) and 5-iodo-2′deoxyuridine (IdU; 20 minutes each).

Nucleotide assay was performed as described in Supplementary Methods and ref. 14.

Reverse-transcription-quantitative PCR (qPCR) was performed following extraction of RNAs and reverse-transcription using Pure Link RNA Mini RNA extraction kits (Ambion) and SuperScript III First-Strand Synthesis SuperMix (Invitrogen), amplifying cDNAs using primers listed in Supplementary Methods with Sybr Green PCR Mix (Applied Biosystems) on a 7500 Fast RT-PCR System (Applied Biosystems).

MCF7 cells were transfected with pcDNA 3.1 (#V79520, Invitogen) or pcDNA3.1 RRM2 (Addgene, Plasmid #13796) using Lipofectamine 3000 (Invitrogen). On day 2, cells were siRNA-transfected, reseeded on day 3 into 96-well plates, and on day 4 luciferase assays were performed (ONE-Glo EX Luciferase assay, Promega). Values were expressed as relative light units per μg protein.

MCF7 cells were transfected with pcDNA 3.1 (#V79520, Invitrogen) or pcDNA3.1 RRM2 (Addgene, Plasmid #13796) using Lipofectamine 3000 (#L3000001, Thermo Fisher Scientific). After selection with 800 μg/mL G418 (Sigma-Aldrich) for 30 days to obtain stable clones, RRM2 expression was tested by Western blot analyisis.

Two screens were performed in duplicate as detailed in Supplementary Methods. In brief, the first screen tested five ER⁺ breast cancer cell lines with a 60-compound custom library at 100 nmol/L, 1 μmol/L, and 10 μmol/L alone or with 1 μmol/L xentuzumab. Screen replicate correlation was 0.97 and the screen Z-Factor calculated using the positive control for viability inhibition (PLK1 inhibitor BI-2536) was 0.64 at 100 nmol/L, 0.68 at 1 μmol/L and 0.72 at 10 μmol/L, indicating high-quality screens. Second, ATM WT MRC5-SV2 and ATM^−/− AT5BIVA fibroblasts were tested against a 188-compound kinase chemogenomic set. The screen Z-Factor calculated using the positive control for viability inhibition (camptothecin analogue SN-38) was 0.63 and 0.81 at 0.1 μmol/L for ATM WT and mutated cells, respectively, and 0.83 and 0.81 at 1 μmol/L, indicating a high-quality screen.

Alkaline and neutral comet assays were performed as described in Supplementary Methods. Comets were visualized at ×100 magnification using an epifluorescence microscope (Ni-E, Nikon) and Komet v.5.5 software (Andor), analyzing 100 comets per condition and quantifying tail DNA as % total DNA.

In vivo experiments were performed under UK Home Office approved Project Licence PPL 30/3395 and PIL I9BC08CD7. Before submitting to the UK Home Office, the Project Licence was reviewed and approved by the Oxford University Animal Welfare and Ethical Review Board, in line with the Oxford University Policy on use of animals in research (https://www.ox.ac.uk/news-and-events/animal-research/university-policy-on-the-use-of-animals-in-scientific-research). Animals were treated and tissues processed for IHC as described in Supplementary Methods, using antibodies listed in Supplementary Table S1.

GraphPad Prism v7 was used to perform Student t test to compare means of two groups, one-way ANOVA to compare >2 groups, two-way ANOVA to compare two variables, ≥2 groups. All tests were two-sided and P < 0.05 (*), P < 0.01 (**), and P < 0.001 (***) were considered significant.

To explore the role of IGF1R in protection from endogenous DNA damage, we used IGF1R gene silencing and IGF neutralizing antibody xentuzumab (15). Initially, we detected increased γH2AX foci in xentuzumab-treated MCF7 breast cancer cells, at xentuzumab concentrations that inhibited IGF1R and AKT-S6 phosphorylation (Fig. 1A and B). ERKs were not inhibited in MCF7 although were suppressed in other breast cancer cell lines (Supplementary Fig. S1A). Xentuzumab also caused concentration-dependent increase in lesions marked by 53BP1 (Fig. 1C), and cyclin A-53BP1 costaining indicated that both xentuzumab and IGF1R depletion increased 53BP1 signal in G1 (cyclin A-negative) cells (Fig. 1D). These findings suggest appearance in G1 of 53BP1 nuclear bodies. These lesions form to protect incompletely replicated DNA from erosion during mitosis (16), and their presence suggests a state of replication stress (17). This finding prompted us to perform DNA fiber assays, revealing significantly shortened DNA tracts in IGF1R–depleted MCF7 (Fig. 1E and F) and KPL1 cells (Supplementary Fig. S1B). Replication fork speed was also reduced by IGF1R depletion (Fig. 1G), with comparable reduction in tract length and fork speed within 6-hour exposure to xentuzumab (Fig. 1H and I), suggesting a direct effect of IGF blockade. There was no increase in stalled or collapsed forks, identified as CldU (red) only signal or CIdU tracts followed by short IdU (green) tracts (Supplementary Fig S1C). We also measured sister fork tracts, which extend in opposite directions from a unique replication origin. Sister forks progress at similar rates under physiologic conditions or during replication stress not due to a physical impediment, and at different rates if one fork encounters obstructing DNA lesions (17, 18). We found no significant sister fork asymmetry upon IGF1R depletion (Fig. 1J), suggesting that replication stress originated from global reduction in replication. Supporting the presence of unreplicated ssDNA, we detected increased S33 phosphorylation of replication protein A (RPA32) in IGF1R depleted cells (Supplementary Fig. S1D).

If DNA replication was indeed incomplete, ssDNA regions should trigger activation of the checkpoint kinase ATR (19). Therefore, we assessed S345-CHK1 phosphorylation, a marker of ATR activation at the replication fork, finding enhanced CHK1 phosphorylation in IGF1R–depleted and xentuzumab-treated MCF7 and KPL1 cells (Fig. 2A, Supplementary Fig. S1E and S1F). ATR activation should inhibit origin firing (12), and indeed newly fired origins were reduced after IGF1R depletion (Fig. 2B). IGF1R depletion also induced accumulation of cells in S-phase, detectable in asynchronous cultures (Supplementary Fig. S2A). We also depleted IGF1R in the context of cell-cycle synchronization using nocodazole, which had the predicted effect of G2-M arrest (Supplementary Fig. S2B). In cells transfected with control or IGF1R siRNA prior to nocodazole treatment, there was significant S-phase accumulation (Fig. 2C and D) associated with reduction in the G₂–M population (Supplementary Fig. S2B) representing further evidence of activation of ATR and the intra-S checkpoint. IGFs are well-known to promote G₁–S and G₂–M checkpoint progression via cyclin upregulation, CDK activation, and Rb phosphorylation (1). Our data show that by regulating fork progression, IGFs also influence the rate of transition through S-phase, and reveal hallmarks of replication stress when IGF1R is nonfunctional.

Having identified replication fork delay and shortened but symmetrical sister forks in IGF1R–depleted cells (Fig. 1E, F, and J), we hypothesized that IGFs regulate the supply of substrates for replication. Quantification of deoxyribonucleoside triphosphates (dNTP) revealed significant reduction in dATP content in IGF1R–depleted MCF7 cells, while dTTP and dGTP were unchanged and dCTP undetectable (Fig. 3A; Supplementary Fig. S3A and S3B). In HCT15 colorectal cancer cells, which proliferate more rapidly than MCF7, dATP content was reduced by IGF1R depletion as in MCF7, while dCTP, dGTP, and dTTP were unaltered (Fig. 3B). These differential changes led us to investigate RNR, which converts ribonucleotide diphosphates to deoxyribonucleotides, the rate-limiting step for dNTP production (20). We noted that a microarray study in MCF7 cells had generated an IGF gene signature that associates with adverse outcome in breast cancer, and included upregulation of cyclins, components of the replication machinery and the regulatory subunit M2 (RRM2) of RNR (21). In the same cell line, we found that RRM2 mRNA and protein were upregulated by IGF1 and downregulated by xentuzumab and IGF1R depletion (Fig. 3C–F). Although IGF axis blockade induced significant RRM2 downregulation, we speculate that residual RRM2 protein accounts for lack of effect on dCTP, dGTP and dTTP (Fig. 3A and B). IGF1R depletion did not affect catalytic subunit RRM1 or alternative regulatory subunit p53R2 (Fig. 3E; Supplementary Fig. S3C and S3D). RRM2 was also downregulated in IGF1R–depleted KPL1, HCC1143, ZR-75–1, and T47D breast cancer, DU145 prostate cancer, and U2OS osteosarcoma cells (Fig. 3G; Supplementary Fig. S3E). The magnitude of RRM2 downregulation induced by IGF1R depletion (60%–80% of levels in control-transfectants) appeared unrelated to cell line proliferation rate, quantified as doubling time (Supplementary Fig. S3E, right).

To investigate regulators of RRM2 expression we assessed the principal IGF1R effectors MEK-ERK and AKT. After 3–24 hours, IGF-induced ERK activation was clearly detectable in MCF7 cells (Fig. 3H), in contrast to the absent ERK response in Fig. 1B, perhaps reflecting the shorter IGF stimulation (15 minutes) used in Fig. 1B to test effects of xentuzumab on IGF1R phosphorylation, which is likely to be transient. MEK inhibitor trametinib and AKT inhibitor AZD5363 were confirmed to inhibit ERK and S6 phosphorylation respectively, and both agents suppressed IGF-induced RRM2 upregulation (Fig. 3H and I). We recently reported that IGF1R is recruited to proximal promoters to influence gene transcription (13), but our chromatin immunoprecipitation sequencing (ChIP-seq) did not detect IGF1R at the RRM2 promoter (Supplementary Fig. S3F), suggesting that IGFs regulate RRM2 via canonical signaling, not via noncanonical nuclear translocation. The RRM2 promoter contains motifs for multiple transcription factors (www.genecards.org/cgi-bin/carddisp.pl?gene=RRM2), including E2F1 and FOS/JUN dimer AP-1, and we investigated the contribution of these candidate RRM2 regulators. E2F1 depletion downregulated RRM2, and IGF1R:E2F1 codepletion achieved more profound RRM2 downregulation than was achieved by depleting either factor alone (Supplementary Fig. S3G), suggesting that IGF1R regulates RRM2 via different pathway(s) from E2F1. JUN depletion induced minor RRM2 downregulation (to 67%–72% of control levels), and IGF1R:JUN codepletion achieved no greater RRM2 downregulation than depletion of IGF1R or JUN separately (Fig. 3J). This preliminary result suggested JUN as a mediator of IGF effects on RRM2. To quantify the role of JUN, we transfected MCF7 cells with RRM2 luciferase promoter reporter, finding that promoter activity was reduced by IGF1R or JUN depletion, and was not further reduced by depleting both factors (Fig. 3K). These analyses suggest an epistatic relationship between IGF1R and JUN, consistent with the function of JUN as a transcriptional effector of MEK-ERK (22), and support a model in which IGFs regulate RRM2 expression at least in part via MEK-ERK-JUN.

We used two models to investigate the role of RRM2 in the replication stress phenotype induced by IGF1R depletion. First, we utilized RRM2 stably overexpressing U2OS cells (23), confirming constitutive RRM2 overexpression that was resistant to IGF1R knockdown (Fig. 4A). In control (empty vector) transfectants, xentuzumab induced phosphorylation of ATR targets CHK1 and RPA, consistent with results in MCF7 (Supplementary Fig. S1D–S1F), while RRM2 overexpression suppressed CHK1 and RPA phosphorylation (Fig. 4B). We quantified 53BP1 bodies as surrogate markers of under-replicated DNA, finding increased 53BP1 bodies in IGF1R-depleted U2OS control cells, also consistent with MCF7 (Fig. 1D), and this was completely prevented by RRM2 overexpression (Fig. 4C). We generated a second model, stably transfecting MCF7 cells with control vector or RRM2 cDNA, and used these cells to perform DNA fiber assays, cell-cycle analysis and 53BP1 immunofluorescence. In control transfectants, IGF1R depletion caused significant shortening of DNA tracts and modest but significant S-phase accumulation (Fig. 4D and E; Supplementary Fig. S3H) comparable to effects in parental MCF7 (Fig. 1E and F; Fig. 2C and D). In MCF7 cells stably overexpressing RRM2, there was partial rescue from DNA tract shortening after IGF1R depletion (Fig. 4D and E) and no increase in S-phase cells (Supplementary Fig. S3H). Consistent with data in the U2OS model (Fig. 4C), RRM2 overexpression also rescued MCF7 cells from xentuzumab-induced increase in 53BP1 bodies, partial rescue likely attributable to reduction in endogenous RRM2 upon IGF1R depletion (Fig. 4F and G). We also tested RNR inhibitor didox, finding no greater viability suppression in MCF7 cells treated with didox plus xentuzumab compared with didox alone (Supplementary Fig. S3I). These data identify RRM2 as a central factor in the replication stress phenotype induced by IGF1R depletion.

Given that IGF axis targeting significantly perturbed replication fork progression, we quantified effects on proliferation, testing xentuzumab in the five luminal breast cancer cell lines in which IGF1R depletion downregulated RRM2 (Fig. 3G). Using xentuzumab at 1 μmol/L, approximating to the steady-state plasma concentration at the dose chosen for phase II evaluation (24), xentuzumab caused relatively little inhibition of cell viability (Fig. 5A). This minor effect in the face of significant replication stress suggested a backup pathway that alleviates toxic effects of replication intermediates, allowing a threshold of tolerable replication stress. To identify components of such a pathway, we performed a targeted compound screen using kinase, cell cycle, DNA replication and repair inhibitors in the five ER⁺ breast cancer cell lines. We discuss MCF7 data here, and will report other outcomes separately. Assessment of the 12 top-ranked compounds achieving greatest viability inhibition with xentuzumab revealed partial overlap at the three screened compound concentrations (Fig. 5B). Included here were inhibitors of PARP, EGFR, and MEK (Supplementary Table S2), already identified as showing potential for combination with IGF1R blockade (1). Of five compounds ranked highly at all concentrations (Supplementary Fig. S4A), only one, ATM inhibitor KU-55933, had not previously been identified as an attractive partner for IGF coinhibition. We validated KU-55933 as a screen hit in MCF7 and ZR-75–1 cells, finding that KU-55933 had little effect in control-treated cells, consistent with the known lack of toxicity of ATM inhibition in the absence of exogenous DNA damage (25), while addition of xentuzumab significantly inhibited viability at KU-55933 concentrations ≥300 nmol/L (Fig. 5C; Supplementary Fig. S4B). We confirmed that at these concentrations, KU-55933 inhibited irradiation-induced phosphorylation of ATM target KAP1, both in the absence and presence of xentuzumab (Fig. 5C and D; Supplementary Fig. S4C). Xentuzumab also caused synergistic inhibition of the viability of MCF7, MiaPaca-2, and BXPC-3 cells when combined with a second ATM inhibitor, AZD0156 (Fig. 5E; Supplementary Fig. S4D and S4E) at clinically achieved submicromolar concentrations (25). Furthermore, MCF7 cell death was significantly increased in cells exposed to xentuzumab with KU-55933 or AZD0156 (Fig. 5F and G; Supplementary Fig. S5A). Given evidence of ATR activation upon IGF1R targeting (Fig. 3A; Supplementary Fig. S1D–S1F), we tested for synthetic lethality with ATR inhibitors AZ20 or VE-821, finding no increased toxicity on addition of xentuzumab (Supplementary Fig. S5B and S5C). These results support ATM-specific effects of the combination with KU-55933 or AZD0156.

ATM mutations and deletions are detectable in a significant proportion of epithelial cancers including colorectal cancer (12%), lung (10%), prostate (7%), pancreatic (4%) and breast (3.5%) cancers (cbioportal.org). ATM protein loss is reported in ≤60% of sporadic breast cancers and ∼13% of pancreatic ductal adenocarcinomas (PDAC; refs. 26, 27), significantly exceeding ATM mutation rates, reflecting complex regulation of ATM expression. Given this evidence of frequent ATM loss, we took five approaches to test IGF dependence in ATM-deficient cells. First, we accessed publicly-available data from 926 cancer cell lines (cancerrxgene.org, kobic.kr), including cell lines harboring homozygous ATM mutations, which lack functional ATM. Homozygous ATM^−/− cell lines were more sensitive to IGF1R inhibitor BMS-754807 (P = 0.0018), with a trend to increased sensitivity to linsitinib (P = 0.08, Fig. 6A; Supplementary Fig. S6A). Second, we took an unbiased approach to assess IGF1R function in ATM^+/+ and null (A-T) fibroblasts, confirming ATM status and using these cells in a kinase inhibitor screen (Fig. 6B; Supplementary Fig. S6B). Consistent with identification of ATM as a hit in the xentuzumab screen (Fig. 5B and C), IGF1R inhibitor GSK1838705 was the top synthetic-lethal interaction in ATM^−/− fibroblasts when testing compounds at 1 μmol/L, and the second-ranked hit at 0.1 μmol/L (Supplementary Fig. S6C). Low-throughput assays confirmed that ATM^−/− fibroblasts were significantly more sensitive to IGF1R kinase inhibitor BI-885578 (28) and xentuzumab (Fig. 5C; Supplementary Fig. S6D). BI-885578 suppressed IGF-induced AKT activation in both ATM-proficient and -deficient cells (Fig. 6C, right), excluding failure of target inhibition as causing resistance of ATM^+/+ cells to IGF1R blockade. Third, in an isogenic ATM-proficient/deficient H322 lung cancer model (29), we found that ATM-null cells were significantly more sensitive to BI-885578 (Supplementary Fig. S6E). H322 cells are nontumorigenic (29), so as a fourth approach, we sought alternative ATM-null models for in vivo testing. Colorectal cancer cell lines LoVo and SK-CO-1 both harbor WT PTEN and β−catenin and mutant KRAS and APC (28) but differ in ATM status, SK-CO-1 being ATM null (30). LoVo cells were resistant to IGF1R inhibition, consistent with (28), despite effective suppression of IGF-induced AKT phosphorylation, while SK-CO-1 cells were significantly more sensitive to xentuzumab and BI-885578 (Fig. 6D; Supplementary Fig. S6F).

We next asked whether ATM status influences response to IGF1R inhibition in 3D models. First, we treated mice bearing SK-CO-1 xenografts with xentuzumab, dose, and schedule as (15). Xentuzumab was well-tolerated (Supplementary Fig. S6G), significantly inhibited growth of ATM-null SK-CO-1 tumors (Fig. 6E and F; Supplementary Fig. S6H), and suppressed phospho-Ser473 AKT in both human tumor and mouse skin (Fig. 6G), as expected, given cross-reactivity with murine IGFs (15). Cleaved caspase-3 signal was increased in xentuzumab-treated tumors, with no evidence of apoptosis in normal mouse skin (Fig. 6G). RRM2 expression is known to increase during S-phase (20), likely explaining high expression in hair follicles, and was significantly downregulated in xentuzumab-treated tumors particularly when considering moderate-strong (2–3+) signal (Fig. 6G; Supplementary Fig. S6I). We assessed BrdU incorporation to measure DNA replication in vivo, finding heterogeneous incorporation with strong signal in tumor cells and hair follicles (Fig. 6G; Supplementary Fig. S6J), and significant reduction in moderate-strong BrdU signal after xentuzumab treatment, consistent with replication defects induced by IGF1R targeting in vitro (Fig. 1E–G). Finally, we obtained isogenic human MiaPaCa2 pancreatic cancer cell lines that are ATM proficient or rendered ATM null by genome editing (31). We confirmed their ATM status (Supplementary Fig. S6K), and used 3D spheroids to assess anchorage-independent growth. ATM-deficient spheroids were highly sensitive to xentuzumab, while ATM-proficient spheroids were significantly growth inhibited by xentuzumab plus ATM inhibitor AZD0165, but not by either drug alone (Fig. 6H and I).

Next, we explored the role of ATM in alleviating toxic effects of replication stress in IGF-inhibited cells. We had speculated that ATM undergoes protective activation in response to DNA lesions induced by IGF-inhibition. However, we earlier found no evidence that xentuzumab induced KAP1 phosphorylation, indicating that ATM was not activated (Fig. 5D; Supplementary Fig. S4C). Similarly, there was no evidence from KAP1 or ATM phosphorylation that ATM was activated by IGF1R depletion (Supplementary Fig. S6L). As a first approach to assessing the contribution of ATM in protecting replication forks, we performed DNA fiber assays on LoVo and SK-CO-1 cells, matched for major mutations and differing in ATM status. Upon IGF1R depletion, we observed significant shortening of DNA tracts in SK-CO-1 cells (16.1 ± 4 μm to 8.7 ± 3.7 μm; Fig. 7A), consistent with effects in MCF7 and KPL1 cells (Fig. 1E and F; Supplementary Fig. S1B). The basal rate of fork progression was slower in LoVo, and there was only a small change upon IGF1R depletion (10 ± 2.9 μm to 9 ± 2.6 μm; Fig. 7A), suggesting a protective role for ATM. Second, we analyzed 53BP1 bodies, finding, as previously (Fig. 1D), that 53BP1 bodies were increased in xentuzumab-treated G1 (cyclin A-negative) cells. However, no 53BP1 bodies were detected after treatment with AZD0156 alone or with xentuzumab (Fig. 7B), consistent with the importance of ATM in phosphorylating the 53BP1 SQ/TQ cluster required for 53BP1 oligomerization into 53BP1 bodies (16). The absence of 53BP1 bodies after dual ATM:IGF inhibition suggests inadequate protection of under-replicated DNA in replication-stressed IGF-inhibited cells.

Finally, we investigated the nature of the DNA damage induced by IGF-inhibition in the presence and absence of functional ATM, hypothesizing that ATM prevents conversion of tolerable DNA lesions to an intolerable form. If so, this could potentially explain relatively minor inhibitory effects of xentuzumab on viability (Fig. 6A) despite significant induction of replication stress (Fig. 1A–D). MCF7 cells were treated with xentuzumab, KU-55933, or the combination, quantifying SSBs and DSBs by alkaline comet assay, and DSBs only by neutral comet assay. Xentuzumab induced DNA fragmentation in alkaline but not neutral assays (Fig. 7C and D), suggesting that the damage was predominantly SSBs, comparable to effects of RNR inhibitor hydroxyurea (32). The xentuzumab:KU-55933 combination significantly increased DNA fragmentation, exceeding the cumulative effects of each drug separately, in both alkaline and neutral assays (Fig. 7C and D), suggesting appearance of DSBs, potentially originating from SSBs that accumulate after IGF1R inhibition. Unrepaired DSBs are highly toxic (12, 33), accounting for the significant increase in cell death in cells exposed to xentuzumab with KU-55933 or AZD0156 (Fig. 5F and G).

RRM2 is the regulatory subunit of RNR, the sole enzyme capable of de novo dNTP synthesis (20). RRM2 is frequently overexpressed in cancers and associates with adverse outcomes (34). Therefore, it is of therapeutic relevance that its expression is regulated by a druggable target, IGF1R. RRM2 expression is already known to be coordinated at the transcriptional, translational and posttranslational levels by BRCA1, CDK4/6, mTORC1, WEE1/SETD2, and SCF^{Cyclin F}/CDKs, respectively (23, 35–37). Here, we identify IGF1R as causing potent RRM2 upregulation, with comparable effects at the level of mRNA, protein and promoter activity, supporting transcriptional rather than post-transcriptional/translational regulation. Our data also implicate AKT, MEK-ERK, and JUN in mediating RRM2 expression. RRM2 downregulation by AKT inhibition is consistent with a role for mTORC1, although both RRM1 and RRM2 were downregulated by mTORC inhibition (36) unlike the RRM2-specific effect we observed. MEK-ERK involvement is implicated by reports of RRM2 upregulation in KRAS-driven colorectal cancer and PDAC cells (although the mechanism was not explored), and by downregulation of DDR proteins including RRM2 following MEK-ERK inhibition in KRAS-mutant PDAC (10, 38).

IGFs are well-known to promote cell-cycle progression, attributed previously to regulation of G₁–S and G₂-M checkpoints (1). We show that by regulating RRM2 and replication fork progression, IGFs also influence passage though S-phase. Thus, IGF axis blockade leads to RRM2 downregulation, shortened DNA tracts, RPA and CHK1 phosphorylation, and intra-S checkpoint activation. These effects are induced by both pharmacologic IGF axis inhibition and siRNA-mediated receptor depletion. We acknowledge that siRNA transfection could affect DNA integrity but consider that the comparable effects we report on RRM2 expression and fork dynamics support the IGF axis as a major regulator of RRM2 expression and DNA replication. There is a potential discrepancy in our finding of shortened replication tracts and reduced origin firing after IGF1R depletion, because these changes should suppress BrdU incorporation and therefore reduce the S-phase fraction. In contrast, we found an increase in IGF1R depleted cells in S-phase, accentuated by Nocodazole trap/release synchronization (Fig. 3C and D). The reduced rate of replication we identify, shown by delayed incorporation of BrdU analogues IdU and CIdU (Fig. 1E and F), presumably results in delayed transit though S-phase. These findings suggest that the S-phase accumulation induced by IGF1R depletion is explained by longer S-phase transit, resolving apparent discrepancy. IGFs have not previously been implicated in regulating fork progression in the absence of exogenous damage. However, replication fork delay induced by alkylating agent was reportedly exacerbated in IGF1R–null fibroblasts, implying that IGF1R promotes DNA damage tolerance (39), and nuclear IGF1R:PCNA interaction is reported to rescue forking stalling after hydroxyurea treatment in HeLa cells (40). The implications for endogenous replication stress were not discussed in that report, but could relate to our finding that IGF1 regulates RRM2 and dNTP supply.

The phenotype we identify upon IGF axis blockade represents the first example of replication stress induced by RTK inhibition, and contrasts with replication stress associated with positive oncogenic functions of MYC, CYCLIN E, and RAS. Specific molecular mechanisms implicated here include increased origin firing and refiring leading to rereplication, replication/transcription conflicts that generate R-loops, G-quadruplexes, and other secondary structures, or through ROS generation or perturbation of dNTP metabolism (14, 41, 42), leading ultimately to exhaustion of replication substrates including dNTPs (33). This suggests that reduced supply of replication substrates represents a final common pathway leading to replication stress induced by both oncogene activation and IGF-blockade. Replication stress occurring during oncogene-induced senescence has been shown to be due to RRM2 downregulation (43), providing further parallels between molecular events triggered by oncogene activation and IGF inhibition.

Although IGF blockade induced marked replication stress with significant delay in replication fork progression, accumulation of ssDNA lesions and ATR-CHK1 activation, we found only minor reduction in cell viability in xentuzumab-treated breast cancer cells. This is reminiscent of minor cell survival inhibition induced only by relatively high (∼100 μmol/L) concentrations of hydroxyurea that targets dNTP synthesis by direct RNR inhibition (44). Given evidence of ATR activation in response to under-replicated DNA, we predicted that ATR activation was protective, and ATR inhibition would be toxic to IGF-inhibited cells. However, this was not so (Supplementary Fig. S3F and S3G). Rather, the compound screens highlighted the toxicity of IGF:ATM coinhibition, leading us to identify a role for ATM in protecting replication forks by preventing conversion of replication stress-associated DNA lesions into DSBs. ATM plays a pyramidal role in regulating the cellular response to DSBs, which can arise after mishandling of SSBs, in which case DSB repair via HR or NHEJ is critical because unrepaired DSBs lead to premature arrest of transcription and replication, chromosome breakage, and cell death (12). Previous reports linking IGF signaling with ATM include the finding that ATM-mutant fibroblasts downregulate IGF1R (45), and our report of reduced ATM protein stability and kinase activity in IGF1R-downregulated B16 melanoma cells (5), in retrospect likely reflecting apoptosis-mediated ATM protein cleavage (46). Recently, ATM mutation was identified as enhancing sensitivity to IGF1R inhibition in bladder cancer cells (47); together with our data in breast cancer, colorectal cancer and PDAC models, this supports testing of ATM loss as a predictive biomarker for response to IGF axis inhibition.

Two novel roles have been described for ATM, distinct from its role in coordinating the DSB response, and which may be relevant here. First, ATM can be activated by unrepaired SSBs, triggering G₁ arrest so that DNA repair can be completed; this avoids replicating damaged DNA, which would result in DSB formation (48). Second, ATM is also reportedly activated by RRM2 knockdown, while loss of ATM function restores dNTP supply in RRM2-depleted cells via metabolic reprogramming to increase substrates for dNTP synthesis, suppressing hallmarks of replication stress and increasing proliferation (49). However, we found no evidence of ATM activation in the context of RRM2 downregulation and replication stress induced by IGF inhibition, although it is possible that SSBs were induced in IGF-inhibited cells below a threshold reportedly required to activate ATM (48). Furthermore, we found that loss of ATM function in the context of IGF1R depletion or IGF inhibition was not protective, but rather caused synergistic inhibition of cell viability and cell death. The mechanism(s) by which ATM protects from replication stress induced by IGF blockade are not the main focus here, but we speculate that synthetic lethality of IGF:ATM coinhibition relates to failure to form protective 53BP1 bodies around incompletely replicated DNA (Fig. 7B) and/or a role in replication fork protection, consistent with more marked replication fork delay upon IGF1R depletion in ATM-deficient colorectal cancer cells than cells expressing functional ATM (Fig. 7A). Such a role has not previously been ascribed to ATM but there is precedent for repair proteins such as BRCA2 contributing to fork protection in the absence of activating phosphorylation (50), suggesting that ATM phosphorylation may not be not a prerequisite for a protective role in the context of replication stress.

In summary, we show here that IGF axis blockade downregulates RRM2 and delays replication fork progression, leading to accumulation of ssDNA lesions that are converted to DSBs by ATM coinhibition. This transition from managed replication stress to DSB accumulation is new and has not been shown for inhibition of other RTKs. This newly-identified function of IGF1R suggests that patients whose tumors lack ATM may be responsive to IGF blockade, and sheds new light on the regulation of global DNA replication.

U. Weyer-Czernilofsky was a full-time employee at Boehringer Ingelheim RCV. V.M. Macaulay reports personal fees from Boehringer Ingelheim consultancy board outside the submitted work. No disclosures were reported by the other authors.

G. Rieunier: Conceptualization, formal analysis, funding acquisition, validation, investigation, methodology, writing–original draft. X. Wu: Formal analysis, validation, methodology. L.E. Harris: Formal analysis, validation, methodology. J.V. Mills: Formal analysis, validation, methodology. A. Nandakumar: Formal analysis, validation, methodology. L. Colling: Formal analysis, validation, methodology. E. Seraia: Formal analysis. S.B. Hatch: Formal analysis. D.V. Ebner: Methodology. L.K. Folkes: Formal analysis. U. Weyer-Czernilofsky: Conceptualization, validation, methodology. T. Bogenrieder: Conceptualization, methodology. A.J. Ryan: Conceptualization, methodology. V.M. Macaulay: Conceptualization, supervision, funding acquisition, validation, investigation, methodology, writing–original draft.

The authors acknowledge the assistance and expertise provided by Graham Brown from the Microscopy core facility, Department of Oncology, University of Oxford. The authors thank Hui Li from Huazhong University of Science and Technology, China for the generous gift of the pGL3-RRM2-firefly construct, Vincenzo D'Angiolella for the kind gift of U2OS cells overexpressing RRM2, Michael Sanderson (ex-Boehringer Ingelheim, now Merck) for the generous supply of BI 885578, Walter F. Bodmer and Jenny Wilding for the gift of HCT15 cells, Sebastian M.B. Nijman for H322 ATM CRISPR cells, Duncan Jodrell and Frances Richards for MiaPaCa2 WT and ATM CRISPR cells, Eric O'Neil for the gift of BXPC-3 and Walter F. Bodmer, Amato J. Giaccia, and Tim Humphrey for comments on the manuscript. This study was supported by Breast Cancer Now (2014NovPR364), Cancer Research UK (C476/A27060), The Rosetrees Trust and John Black Charitable Foundation (M330-F1), and support to V.M. Macaulay from the NIHR Oxford Biomedical Research Centre and the Harrington Discovery Institute.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.