Background
The insulin-like growth factor 1 (IGF1) pathway plays a critical role in cell growth, cell survival, and protection from apoptosis. It is therefore not surprising that epigenetic and transcriptional changes in IGF1 signaling can induce cancer development and progression [
1,
2].
The administration of exogenous insulin is a common therapy for types I and II diabetes. New insulin-like molecules have small modifications of the insulin molecular structure to improve pharmacokinetic parameters of the original molecule and thereby increase stability and temporal bioavailability. These molecular changes possibly affect the binding affinity towards receptors including the insulin receptor (IR) or the IGF1 receptor (IGF1R). Consequently, insulin analogues with an increased IGF1R affinity likely promote mitogenesis [
3]. Insulin X10 is an insulin analogue which possesses a well-known increased affinity towards the IGF1R and which consequently could promote tumorigenesis (for this reason insulin X10 has never entered the pharmaceutical market). Insulin glargine, currently the most commonly prescribed insulin therapy worldwide [
4], also has an increased binding affinity towards the IGF1R. In vivo glargine is rapidly degraded into compounds with a lower affinity towards the IGF1R. Even today there are some concerns about the carcinogenic risk that the use of glargine might induce [
5,
6].
To study the role of IGF1R signaling in breast cancer, we recently evaluated the tumor promoting capacity of chronic insulin analogue treatment in a human relevant breast cancer (p53
R270H/+WAPCre) mouse model [
7]. In this model, the WAPCre system ensures mammary gland (MG)-specific expression of the heterozygous p53 mutation, which corresponds to a mutational hotspot often found in patients with the
Li Fraumeni cancer syndrome [
8]. Eventually, all mice spontaneously developed these human relevant MG tumors within approximately 1 year. Chronic treatment with compounds that possess a high affinity towards the IGF1R, IGF1 and the insulin analogue X10, significantly decreased the tumor latency time. Frequent injections with insulin glargine, a compound that only mildly activates the IGF1R in vivo [
5], showed a similar trend but the observed tumor latency time decrease was not significant compared to regular insulin [
7]. Systematic signaling pathway mapping of all tumors revealed that the MAPK/ERK signaling cascade was especially strongly activated in IGF1- and X10-induced tumors. Although this provides some insight in to the alternative signaling wiring in insulin analogue-related tumors, a systematic evaluation of the genetic modifications and consequently alterations in cellular pathways and network biological differences of insulin analogue-related tumors is still obscure.
In this study, to gain more insight in to the modulation of tumor development and progression by chronic IGF1R activation, we used a systematic in-depth next-generation sequencing (NGS) approach. RNAseq analysis was performed on 50 insulin analogue-induced MG tumors (control, insulin, IGF1, X10, and glargine treatment). Overall genetic modifications were determined at a tumor level. NGS transcriptome analysis did shed light on the specific tumor development and progression in relation to chronic IGF1R activation. For this, we specifically evaluated the alternative modulation of the hallmarks of cancer [
9] to detect treatment-specific tumor features.
Methods
Chronic in vivo insulin analogue treatment
Previously, we have reported the effect of insulin analogues on tumor development [
7]. Here, for the chronic exposure experiment, 200 (40 mice per treatment), 8-week-old female p53
R270H/+WAPCre mice were obtained from an in-house breeding project. The point mutation in the tumor suppressor p53 gene corresponds to the
Li Fraumeni cancer syndrome mutational hotspot (R273H) in humans. Every other day these mice have been injected (subcutaneously) with either vehicle, insulin, glargine, X10, or IGF1 until tumor development. Once the tumors reached a size of 1 cm
3 and 24 h after their last injection the mice were sacrificed. Tumors and other tissues were isolated. For this study, ¼ of the tumor was stored in RNALater (Ambion, Austin, Texas) at 4 °C for RNA isolation. A miRNA isolation kit (Macherey Nagel, Germany) was used to isolate and purify small and large RNA molecules in one fraction. Tumor latency time is defined as the time (in weeks) for the tumor to form, from the start of the experiment to the first time the tumor was palpated.
Single insulin analogue treatment: an animal experiment
To determine the short-term effects of insulin analogue treatment on mammary gland gene expression, a single insulin analogue exposure experiment was performed with 40 (4 mice per treatment/time point) female, 8-week-old inbred FVB/NRj mice (obtained from Janvier, rodent research models, France). This specific mouse strain was used as it is the closest relation to the p53
R270H/+WAPCre mouse strain. Mice received a single subcutaneous injection with either vehicle, insulin, glargine, X10, or IGF1. The mice were sacrificed 1 or 6 h after the injection, blood was collected (mini collect, Greiner/Omnilabo), and MGs were stored in RNALater (Ambion, USA) at 4 °C for RNA isolation. For this, a Nucleospin RNA isolation kit (Macherey Nagel, Germany) was used. For further technical details, please refer to our previous publication [
7].
In vitro stimulation experiments
Next, an in vitro experiment was performed to reveal the transcriptomic effects of insulin analogue exposure on a human breast cancer cell line. MCF7 human breast cancer cells with an overexpression of the IGF1R and a stable knockdown of the IR (MCF7 IGF1R) were seeded at 60% confluence and starved for 2 days in 5% charcoal/dextran-treated FBS (CDFBS, Hyclone, USA) containing RPMI 1640 (Gibco, USA) medium. Cells were stimulated (with 10 nM compound) for 1 or 6 h after which RNA was isolated using NucleoSpin® miRNA isolation kit (Machery Nagel, Düren, Germany). Stimulations included: insulin NPH (Insuman Basal, Sanofi Aventis), insulin glargine (Lantus, Sanofi Aventis), insulin X10 (AspB10, Novo Nordisk), and IGF1 (Increlex, Ipsen). For more technical details about cell line generation and characterization or growth factor stimulation, please refer to our previous publication [
7].
Next-generation sequencing and gene expression analysis
For the chronic insulin analogue exposure experiment, the quality and integrity of the RNA samples were analyzed using the bioanalyzer with an RNA nanochip. The Ion Total RNA-Seq kit was used to process the samples. Samples were Poly-A selected prior to library preparation. This library preparation included the cDNA synthesis and purification steps with the Ion Total RNA-Seq kit v2 (Life Technologies, UK) according to the manufacturer’s instructions. The Ion PI Template OT2 200 Kit v3 and Ion Sequencing 200 kit v3 (both Life Technologies) were used according to the manufacturer’s instructions for sequencing libraries on the PI chip. Sequence runs were performed on the Ion Proton Sequencer (at ServiceXS, Leiden). PI chip analysis, base calling, and quality checks were performed using the Torrent Server Suite. On average, 40 million reads per sample were sequenced with an average read length of 100 base pairs. No additional trimming or filtering of reads was performed before processing. Reads were aligned to mouse genome build GRCm38-Ensembl using Tophat2 (Version 2.0.10). Reads which could not be aligned using Tophat2 were aligned in an additional step, using Bowtie2 (Version 2-2.10) in the local, very sensitive mode. Tophat2 and Bowtie2 aligned reads were merged into a single .bam file for each sample before further analysis.
Gene expression was quantified using HTSeq-Count (Version 0.6.1) using the default options. Differential gene expression was analyzed for compound versus vehicle treatment and was performed using DESeq2 (Version1.2.10). For this analysis, genes with a read count of <50 reads across samples (average of <1 read per sample) were filtered out before the analysis. For the estimation of individual exon expression analysis, a RPKM table was generated with the read counts normalized to library size and gene length (using DEXSeq version 1.8.0). For the mutational profiling, the reads (unfiltered and untrimmed) were aligned to mouse genome build GRCm38.73-DNA primary assembly using TMAP within the Torrent Suite version 4.0.2. Variant calling was performed using the Torrent Suite Variant Caller version 4.0-r72612 with the settings tuned for the detection of somatic mutations at a low stringency level. The reference genome used was the same as that used for read alignment—GRCm38.73. SnpEff version 3.6c was used to filter and annotate the mutations. The list of mutations was filtered to include only exonic mutations with a quality score higher than 250. Several mutations are found in the exact same position in all tumor samples, probably strain-specific single nucleotide polymorphisms (SNPs). Known SNPs in coding regions for the mouse strain FVB/NJ (the strain most closely related to the
p53R270H+/–WAPCre) were downloaded from the Mouse Genome Informatics database and these mutations were discarded from the list. Mouse homologs of the list of human tumor driver genes [
10] were used to define the most clinically relevant mutations.
Phenotypic prediction based on transcriptomic data
To predict the phenotypic characteristics of the treatments of different insulin analogues using their transcriptomics, we constructed a support vector machine (SVM) classifier. We followed a procedure similar to that used in [
11] to identify genes whose expression is significantly associated with cancer cell migration and proliferation across 52 breast cancer cell line data. We then applied them to the orthologs of the mouse transcriptomics data to predict the migratory and proliferative potential of 50 mouse mammary tumor samples using LIBSVM [
11].
We estimated the metabolic fluxes that are most consistent with the transcriptomics data using a computational framework called iMAT [
12]. iMAT integrates the transcriptomics, as ‘soft’ constraints, by ternary partitioning the expression to lowly (–1), mediocrely (0), and highly (1) expressed genes. iMAT then attempts to collect the metabolic states that best correspond to these cues, which constructs a mixed integer linear programming (MILP) problem. We applied iMAT to the human genome-scale metabolic model Recon1 [
13] with a standard medium condition (RPMI) in a condition-specific manner for the five different insulin analogues treatments. The iMAT predicts (i) the biomass production rate and (ii) the metabolic flux rates of individual metabolic reactions. The biomass production rate is the rate at which the biomass precursors are generated with appropriate proportions, and it is incorporated in the metabolic network model as a putative reaction that takes cell and energetic requirements needed to produce biomass as input [
14]. With the metabolic fluxes predicted by iMAT, we performed a pathway enrichment analysis of differentially activated metabolic reactions in X10/IGF1-treated cells to the remainder conditions (insulin/glargine/vehicle). We selected the metabolic pathways whose reactions are significantly enriched in the up-/downregulated group using a hypergeometric test followed by multiple hypotheses correction with the false discovery rate (FDR) 0.05. The predicted biomass production rate does not involve standard deviation because we focused on the metabolic states where the biomass production rate is optimized (thus single-valued).
Statistical analysis
Graphpad Prism version 5.01 software was used for the statistical analysis. All standard error bars in the graphs represent standard deviations. Unpaired two tailed t tests were performed to calculate significance. Multi-experiment viewer (MeV version 4.8.1) was used for the hierarchical clustering analysis.
Discussion
In this study, we used a next-generation sequencing-based transcriptome analysis to characterize mammary gland (MG) tumors from the p53R270H/+WAPCre mouse model that were chronically exposed to insulin-like molecules. We found indications that tumors of mice that received chronic treatment of X10 or IGF1 (compounds with a high affinity towards the IGF1R) show a transcriptomic profile that can be linked to a phenotype with an increased growth potential, enhanced migratory capabilities, and a higher Warburg potential. Moreover, the candidate cancer driver mutations in Ezh2 and Hras were highly enriched in X10 and IGF1 tumors.
This is the first study in which a human relevant breast cancer mouse model has been used to study the tumorigenic effects of chronic insulin analogue treatment. Eventually, all mice from this model developed spontaneous MG tumors with a high human relevance. In this way it was possible to compare tumors induced by insulin analogue treatment with tumors induced by chronic insulin or vehicle treatment. This is in contrast to other studies using wild-type mice where only a few tumors with an origin less relevant to the human situation could be evaluated [
18,
19].
Tumors often have an increased IRA:IRB ratio [
20,
21] that can possibly influence the in vivo transformational effects of insulin analogues [
22]. We found that IRA gene expression levels are upregulated (twofold) in pre-neoplastic MGs compared to expression levels in normal MGs, but IRA gene expression levels in tumors were similar to that of normal MG tissue. This might indicate that the IRA plays a stimulatory role in the transformation of normal to neoplastic MG tissue, but once the MG tumor is established the A isoform of the insulin receptor does not play a key role in proliferative signaling anymore. Surprisingly, IRB gene expression levels were strongly downregulated (over 10-fold) in MG tumors. This suggests that the IRA:IRB ratio is indeed increased in MG tumors, but this effect is mainly caused by downregulation of IRB expression levels rather than an upregulation of IRA.
We anticipated that chronic stimulation with insulin-like molecules would decrease the Warburg potency, as insulin deprivation in human fibroblasts led to an induction of anaerobic glycolysis [
23]. Indeed, we saw that 90% of the insulin-induced tumors showed an increased oxidative phosphorylation response compared to 60% for the spontaneous vehicle-induced tumors. Interestingly, we found that compounds that induced an equi-glycemic response but with an increasing mitogenic potential at the doses used in this study (insulin, glargine, X10, and IGF1) [
7] also showed an increasing percentage of tumors depending on anaerobic glycolysis (10%, 30%, 50%, and 60%, respectively). This might suggest that proliferative signaling is indirectly or directly coupled to the Warburg effect. Tumor samples have deliberately been taken 24 h after the last injection, since we were interested in the long-term rather than the short-term in vivo effects. During this time the exogenous compounds are fully degraded by enzymes and therefore no short-term signaling effect of the compounds can be observed [
24,
25]. However, we cannot fully exclude the effect of direct insulin analogue treatment on tumor cell metabolism, since treatment of MCF7-IGF1R cells with the various insulin analogues also affected the glycolytic metabolic program.
Using the SVM model we could detect a sustained proliferative signaling in the chronic X10/IGF1-treated tumors which might suggest that chronic growth factor treatment can transform tumors in such a way that an autocrine growth factor signaling pathway is induced. A likely explanation would be a differential mutational pattern in cancer driver genes that underlie such a differential proliferative pathway. Chronic insulin, glargine, X10, or IGF1 treatment did not result in more mutations, and no correlation could be detected with the number of mutations or tumor latency time. Interestingly several X10/IGF1-enriched mutations were observed, including Ezh2, Hras, and Traf7, of which Ezh2 and Hras are prominent modulators of human breast cancer. It is possible that these specific mutations contribute (in)directly to the X10/IGF1 phenotypes and enhanced tumor development and progression. Since we have only performed RNAseq we can, of course, not exclude other genomic mutations that were missed in our analysis and that could contribute to MG tumor development and explain the IGF1R-driven enhancement of tumor development/progression in our models.
Conclusions
Altogether our data suggest that the observed decreased tumor latency time in the p53R270H/+WAPCre mouse model after chronic X10/IGF1 treatment is a result of an enhanced tumor biomass production rate. Furthermore, these treatments might facilitate tissue invasion and metastasis and deregulate the cellular metabolism in the tumor. All these factors contribute to an enhanced tumor development, thus decreasing the MG tumor latency time in this model. We did not find any evidence that chronic glargine treatment induced a more aggressive tumor phenotype or increased the biomass production rate, but a slight increased Warburg potential was observed compared to tumors induced by insulin treatment.
Acknowledgements
We thank Dr. Virginie Boulifard (Ipsen, France) for providing the Increlex, IGF1. We are grateful to Dr. Norbert Tennagels and Dr. Ulrich Werner (Sanofi-Aventis, Germany) for providing us with insulin glargine, insulin Basal, and vehicle solutions for the exposure experiments. We thank Dr. Bo Falck Hansen (Novo Nordisk, Denmark) for providing the X10. ServiceXS is thanked for their help during the analysis of the NGS data.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (
http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (
http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.