Nuclear IGF1R Interacts with Regulatory Regions of Chromatin to Promote RNA Polymerase II Recruitment and Gene Expression Associated with Advanced Tumor Stage

Growing evidence implicates the insulin-like growth factor (IGF) axis in promoting risk of cancer and propensity for metastasis, therapy resistance and cancer-related death (1–4). IGFs signal via cell surface type I IGF receptors (IGF1R), activating multiple effectors including AKT and ERKs (5). Until recently, the ability of IGF1R to regulate transcription was thought to be explained solely by these canonical signaling networks downstream of cell surface IGF1Rs (5). This view was challenged when our group and Larsson and colleagues showed that following clathrin-dependent endocytosis, activated internalized IGF-1Rs traffic to the nucleus (6, 7). Furthermore, Larsson's group found that IGF1R import through the nuclear pore complex requires IGF1R-β SUMOylation and activities of p150 Glued and importin-β/RanBP2 (6, 8). We previously reported that nuclear IGF1R is a feature of preinvasive lesions and invasive cancers including prostate, renal, and breast cancers, and identified association between nuclear IGF1R and adverse prognosis in renal cancer (7). Subsequent data associate nuclear IGF1R with proliferation, tumorigenicity, resistance to EGFR inhibition, and clinical response to therapeutic anti–IGF1R antibodies (7, 9–15), suggesting that IGF1R nuclear import requires strong IGF axis activation amounting to IGF dependence.

While nuclear IGF1R is known to interact with chromatin (6, 7), genomic-binding sites previously identified by chromatin immunoprecipitation sequencing (ChIP-seq) in melanoma cells were predominantly intergenic (6), hence of uncertain significance. The aims here were to investigate whether nuclear IGF1R is recruited to transcriptionally active regions of genomic DNA and probe the significance of this phenomenon in clinical cancers. We now report that nuclear IGF1R associates with advanced tumor stage and is recruited selectively to regulatory regions of chromatin including JUN and FAM21A promoters. We identify JUN and FAM21A as mediators of cell survival and IGF-induced migration, properties that tumors require to attain advanced stage. Finally, we detect IGF1R on JUN and FAM21A promoters in tumors that contain nuclear IGF1R, and identify association between tumor nuclear IGF1R content and JUN expression.

Formalin-fixed paraffin-embedded (FFPE) radical prostatectomy (RP) sections were used for IHC using IGF1R antibody #9750 (Cell Signaling Technology) as described (see Supplementary Methods; refs. 16, 17). IGF1R was scored blinded by Uro-Pathologist CV for intensity and percentage of tumor stained, generating immunoreactive scores (range, 0–12) for membrane, cytoplasmic and nuclear IGF1R, and also internalized (cytoplasmic/nuclear, 0–24), and total (membrane/cytoplasmic/nuclear, 0–36) IGF1R. We utilized the same method and scoring system for JUN IHC on adjacent sections using antibody ab32137 (Abcam). The study was approved by National Research Ethics Service Committee Oxfordshire Committee C (study 07/H0606/120). All patients provided written informed consent to use of tissue in research.

DU145 prostate cancer (from Cancer Research UK Clare Hall Laboratories) and SK-N-MC Ewing Sarcoma Family Tumor (ESFT) cells (from Professor Nicholas Athanasou, University of Oxford, United Kingdom) were cultured in RPMI 1640 medium with 10% fetal calf serum (FCS). Both were mycoplasma-free when tested with MycoAlert (Lonza Rockland Inc.). Cultures were used within 20 passages of authentication by STR genotyping (Eurofins Medigenomix Forensik GmbH). Xentuzumab (BI 836845) was provided by Boehringer Ingelheim, bafilomycin A1 (BafA1) and long R3-IGF1 purchased from Sigma-Aldrich.

Serum-starved DU145 and SK-N-MC cultures (50 × 10⁶ cells per condition) were treated with 50 nmol/L IGF1 for 30 minutes, fixed, lysed, and subjected to ChIP using antibodies to IGF1R (#3027, Cell Signaling Technology), H3K4me1 (ab8895, Abcam), H3K4me3 (ab8580, Abcam), RNAPol2 (ab5095, Abcam), or IgG (Santa Cruz Biotechnology, negative control) and the ChIP Assay Kit (17-295, Millipore) according to the manufacturer's instructions (see Supplementary Methods). Independent replicate ChIP-DNAs underwent paired-end sequencing (HiSeq, Illumina). ChIP-Seq reads were mapped using Bowtie2 (18) aligned to the human reference genome (hg19) from UCSC. Aligned reads were filtered against IgG DNA and analyzed with MACS2 for peak calling (19). These softwares reported peaks with assigned FDR values and P values that identify DNA regions with statistically significant binding enrichment. ChIP-seq–identified peaks were validated on triplicate independent samples by ChIP-quantitative Polymerase Chain Reaction (qPCR).

RNAs were extracted and reverse transcribed using Pure Link RNA Mini RNA extraction kits (Ambion) and SuperScript III First-Strand Synthesis SuperMix (Invitrogen). ChIP DNAs and cDNAs were amplified using primers shown in Supplementary Table S1 and Sybr Green PCR Mix (Applied Biosystems) on a 7500 Fast RT-PCR System (Applied Biosystems).

ChIP-seq data were used to design 80 bp oligonucleotides, each 5′ biotin end-labelled on the sense strand (Supplementary Table S1). After annealing (95°C for 5 minutes, cooling to 23°C over 2 hours), biotinylated double-stranded (ds) oligonucleotides were used in electrophoretic mobility shift assay (EMSA) with recombinant human IGF1R residues 960-1397 (rhIGF-1R, Thermo Fisher Scientific) using the EMSA assay kit (Active Motif), according to (20) and the manufacturer's protocol with minor modifications. Each reaction used 100 pmol biotinylated oligonucleotide probe with 0.2 μg rhIGF1R in the absence or presence of 500-fold excess unlabeled probe.

Western blotting, immunoprecipitation, and immunofluorescence were performed as previously with minor modifications (see Supplementary Methods; ref. 7).

DU145 genomic DNA was used as a template to amplify nucleotides –982 to +394 of the JUN promoter (see Supplementary Methods; ref. 21). The 1.4 kb PCR product was digested with XhoI and HindIII-HF (New England Biolabs), cloned into similarly digested pNLCol2 vector (Promega), and the sequence confirmed by DNA sequencing (Source Bioscience). DU145 cells were transfected with pNLCol2-JUN or pNLCol2 empty vector (EV) using Lipofectamine 3000 (Invitrogen), selected with 500 μg/mL hygromycin, and stable clones screened for promoter activity in ONE-Glo EX Luciferase assays (Promega) on a POLARstar Omega platereader (BMG Labtech). DU145 clones incorporating EV or JUN promoter plasmid were serum starved overnight, treated with 50 nmol/L IGF1 for 24 hours, and luciferase assays performed as above.

Assays for proliferation, cell survival, motility, and migration were performed as described in (2) and Supplementary Methods.

T tests were used to analyze two groups, one-way or two-way ANOVA for >2 groups, and Wilcoxon matched pairs signed rank test for nonparametric data. We assessed the significance of variation in IGF1R with clinical parameters with χ², Mann–Whitney U tests and correlation analyses using Prism v6 (GraphPad Software) and Stata package release 11.2 (Stata Corporation). All tests were two sided and P value <0.05 was considered significant.

As a first approach to investigate the significance of IGF1R subcellular localization, we used IGF1R IHC to score IGF1R in the membrane, cytoplasm, and nucleus of 137 RPs from British men with prostate cancer recruited to the Prostate Cancer Mechanisms of Progression and Treatment (ProMPT) study (Supplementary Table S2). IGF1R was detected in benign and malignant epithelium of all RPs, with a luminal-basal IGF1R gradient in benign epithelia that was lost in the cancers (Fig. 1A). Total IGF1R in the cancers was greater than in benign areas of the same RPs (Fig. 1B; Supplementary Fig. S1A), supporting our previous report of IGF1R overexpression in primary prostate cancers (22). Malignant epithelium contained significantly more internalized (nuclear/cytoplasmic) IGF1R, while IGF1R was predominantly in the plasma membranes of benign glands (Fig. 1C; Supplementary Fig. S1B). This difference in subcellular localization is novel and may reflect increased IGF1R activation in malignant versus benign epithelium. Importantly, nuclear IGF1R associated with higher pathological tumor stage (pT1–2 vs. 3, P = 0.011; Fig. 1D; Table 1). We also identified borderline association between internalized (nuclear plus cytoplasmic) IGF1R and higher pathological grade (primary Gleason grade 3 vs. 4–5, P = 0.057; Table 1; Supplementary Fig. S1C).

Having identified association between nuclear IGF1R and adverse clinical factors in men with prostate cancer, we next investigated nuclear IGF1R function by ChIP-seq in human DU145 prostate cancer cells. We also performed ChIP-seq for RNAPol2, and H3K4me1 and H3K4me3 that mark active enhancers and promoters respectively (23). Of approximately 7 to 14 × 10⁶ reads per sample, ≥85% were mapped to the human genome (Supplementary Table S4). Peak calling identified 16,239, 19,759, and 21,782 peaks of RNAPol2, H3K4me1, and H3K4me3 enrichment, respectively, consistent with findings in other cell lines, with a pattern of sharp peaks of RNAPol2 and H3K4me1 recruitment, and broader peaks of H3K4me3 (Supplementary Fig. S2A), reportedly associated with increased transcriptional consistency (23, 24). In contrast, we identified 62 regions with a clear increase in IGF1R ChIP fragment depth compared with control (IgG) ChIP (Supplementary Fig. S2B–S2C). To test the robustness of our data, we repeated IGF1R ChIP-seq in a second model, SK-N-MC Ewing sarcoma cells, which like DU145 showed nuclear IGF1R positivity and inhibitory response to IGF-neutralizing antibody xentuzumab (25) (Supplementary Fig. S2D–S2F). The genome of SK-N-MC cells contained 66 IGF1R–binding peaks, of which 25 were shared with DU145 (Supplementary Fig. S2C).

By comparison with peaks called in RNAPol2 and H3K4me1/3 ChIP-seq, we explored the genomic locations of sites of IGF1R recruitment. Predictably, most RNAPol2 and H3K4me1/3 peaks were within 300 kb of the transcription start site (TSS). Unexpectedly, given the intergenic location of the majority of IGF1R–binding sites reported by (6), our analysis showed that IGF1R peaks also clustered near a TSS (Fig. 2A). Supplementary Table S5 lists the coordinates of IGF1R peaks, the distance from the nearest TSS and the identity of the nearest gene. Of the 62 unique regions of IGF1R binding, 59 (95%) were coincident with RNAPol2 peaks, 54 (87%) with H3K4me1 peaks, and 31 (50%) with H3K4me3. We detected only two peaks in common with IGF1R peaks identified by ChIP-seq in melanoma cells, on chromosome 8 (Supplementary Fig. S2G; ref. 6).

We focused on IGF1R–binding sites within the JUN and FAM21A/C genes that coincided with RNAPol2 and H3K4me1 peaks in both DU145 and SK-N-MC cells (Fig. 2B; Supplementary Fig. S3A), suggesting conserved binding to active regulatory regions. In EMSA, IGF1R bound directly to dsDNA probes representing IGF1R binding regions of JUN and FAM21A promoters (lanes 2, 5; Fig. 2C). The specificity of this interaction is supported by its abolition in reactions containing excess unlabeled probe (lanes 3, 6; Fig. 2C). We then performed ChIP-qPCR to validate ChIP-seq–detected IGF1R binding, first confirming that IGF1 activated IGF1R over 30 minutes (Supplementary Fig. S3B). IGF1R recruitment to JUN and FAM21A/C promoters was enhanced by IGF1 and suppressed by xentuzumab (Fig. 2D and E), supporting requirement for IGF1R activation. Similar ChIP-qPCR would be required to validate the additional IGF1R peaks we identified in ChIP-seq. Contrasting with data from breast cancer and melanoma cells (26, 27), we did not detect IGF1R on CCND1 or IGF1R promoters (Supplementary Fig. S3C and S3D).

Identification of nuclear IGF1R at the TSS of the JUN and FAM21A/C promoters (Fig. 3A) suggests regulatory function. As an initial step to explore this hypothesis, we cloned the proximal JUN promoter, representing the peak of IGF1R recruitment (nucleotides –982 to +394), into a luciferase reporter. In DU145 cells stably transfected with JUN promoter reporter, we detected luciferase activity significantly greater than that in empty-vector transfectants, and in serum-starved cells, reporter activity was enhanced by IGF1 (Fig. 3B). We noted that IGF1R–binding regions of the JUN and FAM21 promoters contained GATA-2 binding motifs, and the JUN promoter peak also contained a KU80-binding motif and AP-1 (FOS/JUN)-like site (Fig. 3A; Supplementary Fig. S4A). This prompted us to question whether nuclear IGF1R interacts with these transcriptional effectors. Therefore, after confirming IGF1R detection in DU145 nuclear extract (Fig. 3C), we performed reciprocal coimmunoprecipitation (co-IP) experiments, revealing evidence of interaction between IGF1R and RNAPol2. This appeared to be IGF dependent when detected by IGF1R IP but constitutive (i.e., present in serum-starved cells) in RNAPol2 IPs. We also detected ligand-independent interaction of IGF1R with KU80 and GATA-2 (Fig. 3D and E; Supplementary Fig. S4B). Noting the abundance of IGF1R in cytoplasmic extract (Fig. 3C), we also tested for interaction between RNAPol2 and cytoplasmic IGF-1R. However, RNAPol2 was almost undetectable in the cytoplasm, with no evidence of IGF1R co-IP (Supplementary Fig. S4C).

Identification of IGF-induced interaction between nuclear IGF1R and RNAPol2 led us to speculate that IGF axis activation might influence RNAPol2 recruitment to chromatin. We used three approaches to test the dependence of RNAPol2 recruitment on nuclear IGF1R. Firstly, using RNAPol2 ChIP-qPCR, we found that IGF1 enhanced recruitment of RNAPol2 to JUN and FAM21A/C promoters (Fig. 3F). Secondly, we assessed RNAPol2 recruitment to the TSS of β2-microglobulin and FOS genes that lack IGF1R peaks. We detected RNAPol2 on these promoters but found no enrichment of RNAPol2 binding upon IGF1 treatment (Supplementary Fig. S4D). Thirdly, to differentiate functions of cell surface and nuclear IGF1Rs, we used BafA1, a vacuolar H⁺-ATPase inhibitor that blocks vesicular trafficking (28). Both xentuzumab and BafA1 blocked IGF1R nuclear translocation; xentuzumab did this by inhibiting IGF1R activation, hence also suppressing downstream signaling, while BafA1 prevented IGF1R internalization without preventing ligand-induced activation of cell surface IGF1Rs and their ability to signal via AKT and ERKs (Fig. 4A–C; Supplementary Fig. S4E). We found that IGF-induced enhancement of RNAPol2 recruitment to the JUN and FAM21A promoters was suppressed by both xentuzumab and BafA1 (Fig. 4D). Furthermore, IGF1 upregulated expression of JUN and FAM21A; as with RNAPol2 recruitment, these effects were also inhibited by xentuzumab and BafA1, although only partially in the case of IGF-induced JUN upregulation (Fig. 4E). While BafA1 blocks internalization of many proteins (28), inhibition of IGF-induced recruitment and transcription does associate this effect with IGF1R. We also detected IGF-induced RNAPol2 recruitment to the FAM21C promoter, but this transcript was not upregulated by IGF1 (Fig. 4D–E, right).

To assess the functional significance of JUN and FAM21A upregulation, we tested effects of depleting these proteins (Fig. 5A). Consistent with the known protumorigenic role of JUN (29), cell survival was reduced in JUN-depleted, although not FAM21A-depleted prostate cancer cells (Fig. 5B). Seeking a FAM21A-associated phenotype, we noted that FAM21 is a component of the Wiskott-Aldrich syndrome protein and SCAR homolog (WASH) complex involved in endosomal trafficking (30). WASH-associated proteins are reportedly required for actin polymerization and cell motility (31), leading us to speculate that FAM21A contributes to this process. Indeed, FAM21A depletion caused delay in cell migration that was significant at 12 and 36 hours, while JUN-depleted cells showed a delay only at 12 hours (Fig. 5C and D; Supplementary Fig. S5A and S5B). To assess more specifically whether JUN and FAM21A contribute to IGF-dependent migration, we performed transwell assays in low serum (0.2% FCS), detecting enhancement of migration toward IGF1 (1.23 ± 0.03 fold, P < 0.001) in control transfectants. This effect was suppressed by both JUN and FAM21A depletion (Fig. 5E), supporting the hypothesis that these proteins contribute to promigratory effects of IGF1. In controls, we observed greater enhancement of migration using 10% FCS as stimulus (1.56 ± 0.63 fold, P < 0.001), consistent with the presence in serum of promigratory factors in addition to IGFs, and this effect was partially suppressed in JUN-depleted but not FAM21A-depleted cells (Supplementary Fig. S5C). To assess potentially confounding effects of proliferation, we also performed viability assays on parallel siRNA-transfected cultures, finding no differences in proliferation over the 24-hour time course of migration assays (Supplementary Fig. S5D).

Having identified a transcriptional role for IGF1R in the nucleus of cultured prostate cancer cells, we used two approaches to investigate the significance of nuclear IGF1R in clinical cancers. First, we performed ChIP-qPCR on fresh-frozen primary prostate cancers and were able to detect IGF1R on JUN and FAM21A promoters, with higher signal in tumors with abundant nuclear IGF1R (Fig. 6A), supporting the clinical relevance of IGF1R ChIP-seq findings in cultured cells. Finally, to further probe the relationship between nuclear IGF1R and JUN expression, we performed IHC for JUN in adjacent FFPE sections of the radical prostatectomies in which we had evaluated IGF1R expression and subcellular localization (Fig. 1). After scoring JUN signal in the malignant epithelium, it was apparent that JUN showed significant correlation with nuclear IGF1R (Fig. 6B and C; Supplementary Fig. S6A). This correlation was not seen for total IGF1R (Supplementary Fig. S6B), supporting the importance of nuclear IGF1R in upregulating JUN. Taken together, these data highlight an important noncanonical nuclear role for IGF1R that promotes the properties required to attain advanced tumor stage (Fig. 6D).

The principal findings of our study are that IGF1R binds directly to DNA, interacts with key transcriptional regulators, contributes to RNAPol2 recruitment and gene expression, and associates with advanced tumor stage. We identified IGF1R recruitment to regulatory DNA sequences by performing parallel ChIP-seq for RNAPol2 and histone marks of active enhancers and promoters. This strategy allowed us to locate regulatory regions of the genome and also assess ChIP-seq efficiency. We compared RNAPol2 and H3K4me1/3 peak numbers with those reported in ref. 23, which used ChIP-seq to study transcriptional regulators in LNCaP prostate cancer cells, obtaining 1.36–10.22 × 10⁶ uniquely mapped reads, and reporting 7,028 binding sites for RNAPol2, 25,469 for H3K4me1, 24,921 for H3K4me3. Therefore, we identified more RNAPol2-binding sites and similar numbers of H3K4me1/3 sites, supporting the ability of our ChIP-seq protocol to detect enrichment of regulatory proteins on DNA. We identified far fewer peaks of IGF1R binding, although two factors support the credibility of the identified peaks and their functional importance. Firstly, we identified IGF1R–binding peaks in common between two cancer cell lines, and we validated IGF1R binding by ChIP-qPCR. Secondly, two previous studies had performed ChIP-seq using the same IGF1R antibody, finding relatively few IGF1R–binding sites in genomic DNA. Larsson's group was the first to use this approach, identifying 568 IGF1R–binding sites in melanoma cells, of which 80% were intergenic and 3.4% (∼20 sites) were ≤20 kb of a TSS (6). In immortalized corneal epithelial cells, Wu and colleagues identified nuclear IGF-1R:INSR hybrid receptors, reporting 88 binding peaks for IGF1R and 86 for INSR, assigned to nearest genes involved in proliferation, cell death, differentiation, cell adhesion, signal transduction, metabolism, and cell communication (32). Thus, there is support for the concept that IGF1R binds to a limited subset of sites in the human genome. The location of these sites may be cell-type specific, possibly related to differences in nuclear structure and chromatin organization (33), given that the binding sites we identified appear to cluster selectively around the TSS, unlike the majority of sites identified in ref. 6.

In addition to clustering around a TSS, the majority of sites of IGF1R recruitment we identified were coincident with peaks of RNAPol2 and H3K4me1 enrichment and 50% coincided with H3K4me3 peaks. This identification of nuclear IGF1R–binding sites at regulatory DNA regions is consistent with a model in which interaction of nuclear IGF1R with DNA regulates gene transcription, supporting previous reports identifying IGF1R on the CCDN1 and IGF1R gene promoters (26, 27). The major difference is that we identified specific promoters by ChIP-seq, while these previous reports were guided to the CCDN1 promoter by recognition that nuclear IRS-1 is also present at TCF/LEF sites of this promoter (27), and the IGF1R promoter by interest in regulation of IGF1R gene expression (26). We did not detect IGF1R on either promoter (Supplementary Fig. S3C and S3D), again suggesting cell-type–specific differences.

The concept that IGF1R binds directly to DNA is supported by our EMSA data using probes corresponding to ChIP-seq–identified IGF1R–binding peaks. In these assays, recombinant IGF1R protein was the only component added to reactions in which probe mobility was retarded (Fig. 2C). Such data have been considered to provide evidence for direct protein:DNA interaction in EMSA characterizing DNA binding of other recombinant or highly purified proteins (20, 34–36). Our co-IP data (Fig. 3D and E) indicate that within intact cells, nuclear IGF1R exists in protein complexes and may be recruited in this context to chromatin. Importantly, we report a hitherto-unrecognized interaction between nuclear IGF1R and RNAPol2. The functional implications of this interaction are currently unclear, given the discrepancy between the time course of the interaction, increasing over 30 minutes (Fig. 3C), and more rapid (5 minutes) IGF-induced RNAPol2 recruitment (Fig. 3F), which could suggest that this response is independent of nuclear IGF1R. We considered probing the contribution of nuclear IGF1R by manipulating its localization, by mutating a nuclear localization sequence (NLS) to block trafficking via importins, or expressing SUMO-site mutant IGF1R (6). However, recent work detects SUMO-site mutant IGF1R in the nucleus, possibly via heterodimerization with INSR (37), and furthermore IGF1R lacks an identifiable NLS, and unlike (8) we cannot detect IGF1R interaction with importin-β (7). Therefore, we adopted the strategy of comparing complete pathway blockade by xentuzumab with internalization inhibition using BafA1. This approach generated evidence implicating nuclear IGF1R, by the suppression of IGF-induced RNAPol2 recruitment by BafA1 (Fig. 4D), which blocks receptor internalization but not membrane signaling (Fig. 4A–C). Furthermore, IGF1 did not influence RNAPol2 recruitment to promoters lacking IGF1R–binding sites (Supplementary Fig. S4C). The functional significance of IGF1R:RNAPol2 complex formation could be further explored by identifying and disrupting the IGF1R domain(s) required for RNAPol2 interaction, using ChIP to test recruitment of IGF1R and RNAPol2 to JUN and FAM21A promoters.

While nuclear IGF1R is detectable by immunohistochemistry, immunofluorescence, and subcellular fractionation, these techniques also detect abundant cytoplasmic IGF1R. It is unclear to what extent nuclear IGF1R functions are related to/dependent on cytoplasmic IGF1R. Ligand-induced IGF1R activation promotes IGF1R internalization into the cytoplasm, but we were unable to detect interaction with RNAPol2 in this subcellular compartment. However, internalized IGF1R is known to mediate sustained signaling to AKT, and/or reflects IGF1R that is en route to degradation, recycling to the plasma membrane or trafficking to the nucleus (6, 7, 38, 39). Given these considerations, it is plausible to consider that plasma membrane IGF1R might be inactive, without protumorigenic function. In future, it will be interesting to assess tissue IGF expression, to determine whether nuclear IGF1R positivity associates with, and is potentially a response to, high ambient ligand levels.

It is increasingly recognized that multiple components of the IGF-insulin axis undergo nuclear translocation, including IGF and insulin receptors, docking molecules and IGF-binding proteins (26, 40–43). Indeed, INSR reportedly undergoes insulin-stimulated recruitment to the promoters of genes already known to be insulin induced, contributing to glucose homeostasis (41). Similarly, IGF1 is known to promote tumor growth at least in part by upregulating JUN (44). FAM21 has not previously been linked with IGF signaling, but FAM21A is clearly upregulated here by IGF1, and IGF-induced JUN and FAM21A transcription is comparably inhibited by IGF-neutralizing antibody xentuzumab and by preventing IGF1R internalization (Fig. 4E). The equivalence of these responses suggests that internalized/nuclear IGF1R contributes to IGF-induced transcription, although we acknowledge that canonical signaling, for example, via ERKs may also contribute to this transcriptional effect. IGF1 does not upregulate FAM21C, suggesting either that IGF1R recruitment has no functional effect at this locus, or is consistent with reports that transcription is initiated but not completed in a large fraction of human genes (45).

Together with our earlier report of association with adverse outcome in renal cancer (7), the finding here that nuclear IGF1R associates with advanced stage in prostate cancer supports a link with aggressive tumor behavior. While it is not possible to infer a causative association from these clinical findings, experimental data support functional relevance, linking nuclear IGF1R with increased IGF-induced proliferation, gefitinib resistance, and enhanced tumorigenicity (9, 10, 13). These phenotypes could be mediated at least partly by FAM21, reported to promote chemo-resistance in pancreatic cancer (46), and JUN, which associates with radioresistant prostate cancer in patients and murine models (47, 48). In radiotherapy-treated prostate cancers, we recently reported that IGF1R upregulation associates with high Gleason grade and risk of metastasis, and cytoplasmic and internalized IGF1R with biochemical recurrence, although there were no specific associations with nuclear IGF1R (49). However, nuclear IGF1R was recently found to interact with PCNA to influence the response to DNA damage (50). Finally, we identify JUN and FAM21 as mediators of IGF-induced migration (Fig. 5E). Suppressed migration toward FCS in JUN-depleted but not FAM21A-depleted cells (Supplementary Fig. S5C) suggests that JUN mediates migration induced by additional stimuli present in FCS, while FAM21A may more specifically mediate chemotactic response to IGF1.

Taken together, these data highlight an important noncanonical nuclear role for IGF1R that associates with advanced tumor stage and reveal hitherto-unrecognized molecular pathways through which IGFs promote tumor cell survival and motility. Given the reported association of nuclear IGF1R with clinical response to IGF1R inhibition (14, 15) and our demonstration that IGF-neutralizing antibody antagonizes nuclear IGF1R functions, these findings have implications for clinical evaluation of IGF-inhibitory drugs.

T. Bogenrieder is clinical development leader in Boehringer Ingelheim. No potential conflicts of interest were disclosed by the other authors.

Conception and design: T. Aleksic, G. Rieunier, E. Osher, M. Sanderson, V.M. Macaulay

Development of methodology: T. Aleksic, G. Rieunier, E. Osher, J. Mills, R.J. Bryant, V.M. Macaulay

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T. Aleksic, X. Wu, R.J. Bryant, K. Hutchinson, F.C. Hamdy

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): T. Aleksic, N.E. Gray, X. Wu, G. Rieunier, E. Osher, J. Mills, R.J. Bryant, C. Han, M. Sanderson, S. Taylor, V.M. Macaulay

Writing, review, and/or revision of the manuscript: T. Aleksic, G. Rieunier, E. Osher, R.J. Bryant, U. Weyer-Czernilofsky, M. Sanderson, T. Bogenrieder, V.M. Macaulay

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): T. Aleksic, N.E. Gray, A. Lambert, U. Weyer-Czernilofsky, T. Bogenrieder, S. Taylor

Study supervision: V.M. Macaulay

Other (scoring of IHC and selection of tumor areas): C. Verrill

Other (identified and provided samples of research material from Biobank of prostate cancer material): A. Lambert, R. Kumar

We are very grateful to the prostate cancer patients who gave permission for use of tumor tissue in research. We are also grateful for technical advice and support from Simon Engledow (High-Throughput Genomics Group, Wellcome Trust Centre for Human Genetics Oxford), and Graham Brown (Microscopy Core Facility, Department of Oncology, Oxford), and for comments on the manuscript from Eric O'Neill and Sovan Sarkar. This study was supported by Prostate Cancer UK (G2011/20, G2012/25), Development Fund of Cancer Research UK Oxford Cancer Research Centre (CRUKDF 0715-VMTA), UCARE-Oxford (TA/VM-2016), National Institute for Health Research (NIHR) Research Capacity Funding (grant AC14/037), Breast Cancer Now (2014NovPR364), The Rosetrees Trust and John Black Charitable Foundation (M330-F1), and support to V.M. Macaulay from the NIHR Oxford Biomedical Research Centre. The ProMPT study is supported by the UK NIHR, Cancer Research UK and the MRC, and the Cambridge and Oxford Biomedical Research Centres. The funding source had no role in the design, conduct of the study, collection, management, analysis and interpretation or preparation, review, or approval of the manuscript.

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.