Alternative signaling network activation through different insulin receptor family members caused by pro-mitogenic antidiabetic insulin analogues in human mammary epithelial cells

Introduction Insulin analogues are designed to have improved pharmacokinetic parameters compared to regular human insulin. This provides a sustained control of blood glucose levels in diabetic patients. All novel insulin analogues are tested for their mitogenic side effects, however these assays do not take into account the molecular mode of action of different insulin analogues. Insulin analogues can bind the insulin receptor and the insulin-like growth factor 1 receptor with different affinities and consequently will activate different downstream signaling pathways. Methods Here we used a panel of MCF7 human breast cancer cell lines that selectively express either one of the isoforms of the INSR or the IGF1R. We applied a transcriptomics approach to assess the differential transcriptional programs activated in these cells by either insulin, IGF1 or X10 treatment. Results Based on the differentially expressed genes between insulin versus IGF1 and X10 treatment, we retrieved a mitogenic classifier gene set. Validation by RT-qPCR confirmed the robustness of this gene set. The translational potential of these mitogenic classifier genes was examined in primary human mammary cells and in mammary gland tissue of mice in an in vivo model. The predictive power of the classifier genes was evaluated by testing all commercial insulin analogues in the in vitro model and defined X10 and glargine as the most potent mitogenic insulin analogues. Conclusions We propose that these mitogenic classifier genes can be used to test the mitogenic potential of novel insulin analogues as well as other alternative molecules with an anticipated affinity for the IGF1R. Electronic supplementary material The online version of this article (doi:10.1186/s13058-015-0600-5) contains supplementary material, which is available to authorized users.


IN THE PICTURE
Immunofluorescent microscopy set up. This type of microscopes is often used to detect fluorescent light emitted by your protein of interest. In this way a high resolution and magnified picture can be created so that specific cell features can be studied. By taking pictures over time, the migratory behaviour of certain cells under specific conditions can be studied.

IN BEELD
immunofluorescentie microscoop opstelling. Dit type microscoop wordt vaak gebruikt om fluorescerend licht, uitgestoten door een te onderzoeken eiwit, te detecteren. Op deze manier kunnen hoge resolutie en vergrote plaatjes worden gecreëerd. Door op verschillende tijdpunten foto's te nemen, kan het migrerende gedrag van cellen onder bepaalde condities worden bestudeerd. o IGF1 and AspB10 induce a very strong transcriptomic response, especially 1hr after stimulation o A mitogenic gene expression signature could be defined that was validated with different models and showed a high predictive potential as compared to functional assays o AspB10 and glargine are the only insulin analogues with an increased mitogenic potential compared to regular insulin as determined with this transcriptomic analysis

Introduction
Diabetes mellitus is the most common endocrine disease with over 380 million patients in 2013, worldwide [142]. A common treatment for both type-1 and type-2 diabetics is the use of insulin analogues, which are insulin-like molecules with altered pharmacokinetic parameters so that they are either absorbed more rapidly or slower compared to regular insulin after injection. A combinational treatment with these short and long-acting insulin analogues provides the patient with normal blood glucose levels. These insulin analogues have been used for several decades, but recently some epidemiological studies found a correlation between the use of some of these increase the risk of cancer [147]. Firstly, the changes to the molecular structure of insulin affect the binding properties towards different receptors (e.g. insulin receptor-A (IRA) [134] or insulinlike growth factor-1 receptor (IGF1R) [148]). As a consequence these insulin analogues have an increased mitogenic potential. In this scenario the insulin analogues could act either as a tumor initiator by transforming benign or (pre-) neoplastic cells which often express increased levels of IRA and IGF1R [136], or as a tumor promoter by stimulating the increased growth potential of these cells. Secondly, insulin analogues might induce mutagenic action either directly or indirectly as a statistical consequence of the increased mitogenic potential. However, evidence for an indirect enhanced mutagenic effect due to insulin analogue treatment has never been observed and therefore the first hypothesis is the most plausible scenario. As indicated before some insulin analogues have an increased binding potential towards the IGF1R [114] and/or a  . We used transcriptomics to define gene sets involved in insulin analogue induced mitogenic signaling. These genes are candidate mitogenic classifiers to predict the mitogenic potential of newly developed insulin analogues or growth factors in general that act on the IGF1R.

Primary cell isolation, Cell line generation and cell culturing
Cell lines based on the human breast cancer MCF7 cell lines, which predominantly express the IRA, IRB or IGF1R have been described previously [46]. All MCF7 derivatives were cultured in RPMI 1640 medium (Gibco, Invitrogen, Carlsbad, CA, USA) supplemented with 10% fetal bovine serum (FBS) and 100 U/mL penicillin-streptomycin (Invitrogen).
Primary human mammary cells have been isolated from cryopreserved biopsies of two individuals as described previously [152]. The two biopsies were obtained from two female patients who had undergone breast cancer-related surgery at the Leiden University Medical Center (LUMC). Procedures were followed according to the Dutch Medical Treatment Act and the study protocol was compliant with "the Code of proper secondary use of human tissue in the Netherlands" issued by the Dutch Federation of Medical Scientific Societies and approved by the Medical Ethical committee of the LUMC (P10.226). Specimens were coded anonymously in a way that they were not traceable back to the patient by laboratory workers. As much as possible fat tissue was removed from the human mammary biopsies, thereafter they were cut into 8 mm 3 pieces, which were then dried and attached to the culture flask for 30 minutes. 20% FBS containing DMEM-F12 medium (GIBCO/Invitrogen, Breda, the Netherlands) was gently added and refreshed every 5 days. Around the edges of the tissue, cells (mainly fibroblasts) started growing and after 3 weeks the culture flask was confluent with cells. The fraction of epithelial cells was enriched by multiple short trypsinisation steps in which part of the fibroblasts were removed. For 2 more passages the cells were cultured in HuMEC Ready Medium (GIBCO/Invitrogen). After this step the primary mammary cells were cultured in DMEM-F12 medium supplemented with 10% FBS and 100 U/mL penicillin-streptomycin (Invitrogen).

Insulin, insulin analogues and IGF1 in vitro stimulation
Prior to compound stimulation the cells were starved in 5% charcoal/dextran-stripped fetal For this 10,000 MCF7 cells per well were seeded in 96-well plates in complete growth medium.
24 hours after seeding, 50 nM smartpool siRNA mix was delivered to the cells using a standard transfection method with DharmaFECT 4 transfection reagent (Dharmacon Technologies) according to the company's instructions. 24 hours after transfection, the small interfering RNA (siRNA) transfection mixture was replaced with 5% CDFBS starvation medium for drug treatment and sulphorhodamine B (SRB) proliferation assay.

Sulforhodamine B colorimetric assay determining cell proliferation
A SRB assay was used to measure the total amount of protein as a measure for cell proliferation.
Transfected and drug-treated cells in 96-wells plates were fixed with 30 μl 50% trichloroacetic acid directly added to 100 μl of assay medium per well for 1 hour at 4°C on a shaker, washed five times with distilled water and air-dried. Fixed cells were stained with 60 μl of 0.4% SRB (dissolved in 1% acetic acid) at room temperature on a shaker for 30 minutes. After the SRB protein binding, the plates were washed five times with 1% acetic acid to remove unbound dye and air-dried between the washing steps. Next, the protein-bound SRB in each well was solubilized in 200 μl 10 mM unbuffered Tris solution (pH>10) for 10 minutes on a plate shaker. Absorbance was measured at 530 nm with a FLUOstar OPTIMA plate reader (BMG LABTECH, Offenburg, Germany).

Western blotting
Western blotting was used to determine the knockdown efficiency of the siRNA transfection. To prepare cell lysates for Western blot analysis, cells were washed two times with ice-cold PBS and lysed with 1x SPB with 1:20 β-mercaptoethanol. Samples were boiled at 95°C for 5 minutes and stored at -20°C. Before loading, samples were denatured at 95°C for 5 minutes. 20 μl (about 30 ug) protein solution per lane was separated by SDS-polyacrlyamide gel electrophoresis on a 7.5% acrylamide gel and electrophoretically transferred to a polyvinylidene fluoride membrane (Millipore, Billerica, MA, USA). Prior to primary antibody probe, membranes were blocked for 1 hour at room temperature with 5% bovine serum albumin (BSA) or 5% milk in Tris-bufferd saline Tween 20 (TBST) buffer (100mM Tris, pH 7.4, 500mM NaCl, 0.05% Tween 20). ERK, AKT, PTEN and tubulin antibodies were probed in 1% BSA-TBST buffer, whereas IGF1Rβ antibodies were probed in 1% milk-TBST buffer. HRP-conjugated secondary antibody incubation was performed in 1% BSA-TBST or 1% milk-TBST buffer, corresponding to the primary antibodies used. Protein bands were visualized by using the ECL (Amersham) method, after which the membrane was scanned by using a Typhoon 9400 imager (GE Healthcare). Anti-phospho-Akt (Ser473) and antiphospho-Erk (Thr202, Tyr204) have been purchased from Cell signaling Technology, MA, USA).
For a detailed description of the methods and origin of the antibodies we refer to our prior publications [46] [153].

Microarray studies
For the microarray the cells were seeded at a confluence of 60% in 6 cm plates, starved for 2 days in 5% CDFBS containing medium, followed by 1 hr or 6 hrs compound stimulation (10 nM) in serum free medium. Small and large RNA was isolated and purified using NucleoSpin® miRNA isolation kit (Machery Nagel, Düren, Germany) according to manufacturers' instructions. RNA quality and integrity were assessed by using the Agilent 2100 Bioanalyzer System (Agilent Technologies, Santa Clara, CA, USA). The Affymetrix 3' IVT Express Kit (Affymetrix, Santa Clara, CA, USA) was used to synthesize biotin-labeled cRNA, and this was hybridized to an Affymetrix HG-U133plus PM Array plate reader. Probe annotation was performed using the hgu133plus2.db package and probe mapping was performed with hgu133plus2cdf package installed using

RT-Q-PCR
For the qPCR analysis, messenger RNA from MCF7 cells (80% confluent 6-well) or mammary glands (30 µg tissue) was isolated/purified using NucleoSpin® miRNA isolation kit (Machery Nagel, Düren, Germany). cDNA was made using the universal cDNA synthesis kit (Exiqon). qPCR was performed in triplicate using SYBR Green PCR (Applied Biosystems) on a 7900HT Fast Real- and water were provided ad libitum. Animals received a single subcutaneous injection of 100 μl compound/vehicle solution. The doses were chosen so that the glucose drop was constant among the different compounds (Supplemental Figure 3A) (glargine and insulin 100 nmol/kg, X10 1200 nmol/kg and IGF1 12.5 mg/kg). One or six hours after the injection the mice were sacrificed, blood was collected (mini collect, Greiner/Omnilabo), blood glucose levels were measured (Freestyle light, 70812-70, Abbott), the 3 rd and 4 th mammary glands were isolated and used for Western blot protein quantification and quantitative PCR respectively (see Ter Braak, 2014 and2015) [46] [153]. For each condition (treatment/time point) 4 mice were included.

Statistical analysis
For the statistical analysis of the microarray data, R (version 3.1) software was used. Rest of the analysis was performed with Graphpad Prism version 4.00. Student's t-tests were used to determine significance between conditions. P-values lower than 0.05 were considered to be significant. In all graphs the error bars represent standard deviations.

Mitogenic signaling is regulated via highly similar signaling cascades in the INSR and IGF1R signaling pathway
To better understand the involvement of the IRA, IRB and IGF1R pathways ( Figure 1A) in the context of mitogenic signaling of insulin analogues, we used our previously described human MCF7 breast cancer cell lines that express either IRA, IRB or IGF1R [46]. Exposure of these individual cells to the pro-mitogenic insulin analogue X10 that activates both the INSR and the IGF1R, resulted in intact downstream signaling cascades in all three cell lines, indicating functionality of the receptors ( Figure 1B). As a next step we wanted to ensure that the IRA, IRB and IGF1R are not entirely different regarding their key intracellular proliferative signalling pathways. For this purpose, we tested the proliferative potential of the cells after knockdown of several key signaling molecules in these pathways ( Figure 1A). As a first step, we optimized the knockdown efficiency using IGF1R and ERK1/2 as controls. The knockdown efficiency of IGF1R was almost 100% and constant over five days of culturing; the knockdown efficiency of ERK1 and ERK2 was around 50% after day 1 till 95% at day 5 ( Figure 1C), we assume that the knockdown efficiency is constant for other targets but obviously it was practically not feasible to test them all in this manuscript. To assess the proliferative and anti-apoptotic effects of these knockdowns, we used the SRB proliferation assay. After knockdown of ERK1/2 and IGF1R, the amount of cells after five days of culturing was significantly decreased with 25% ( Figure 1C), indicating that the SRB proliferation assay is a sensitive assay to pick up any anti-proliferative effects.
Next, we determined the effect on proliferation of ten individual signaling molecules that are key It could be argued that the effects described above are not (solely) due to pro-mitogenic effects, since the INSR/IGF1R signalling pathway can also induce anti-apoptotic effects (See figure 1A).
These anti-apoptotic effects could also lead to more cells and thus a higher SRB assay read out.
To investigate this we determined the apoptotic fraction with a FACS analysis upon stimulation with the different growth factors (insulin, glargine, X10, IGF1) at 0, 10 and 100 nM. As expected, we found a slightly, but dose dependent, higher fraction of apoptotic cells in the growth factor stimulated cells (~6%) versus the unstimulated (~4%) (data not shown). Since this is such a small difference we assume that the anti-apoptotic effects play a minor role compared to the promitogenic effects in the growth factor stimulation experiments.
In conclusion, these combined data indicates that the core signaling pathways involved in cell  Insulin analogues trigger different transcriptomes in the different cell lines.
To detect the differences in gene expression levels between the different cell lines, we next performed a microarray experiment using the same cell line panel ( Figure 1B). We determined the significantly differentially expressed genes (DEGs) per condition (1 hour stimulation, Figure 2B; 6 hours stimulation, Figure 2C)

Differential pathway activation by the various insulin analogues.
To further understand the biological pathways up regulated by these different compounds we performed an Ingenuity Pathway Analysis (IPA), focussing on the MCF7 IGF1R cell line using both time points (Venn diagram Figure 3A). A mitogenic cluster was defined that included all DEGs of IGF1 treatment only, or IGF treatment in combination with X10 and/or glargine treatment. We included glargine treatment in this cluster as glargine, like X10, is highly mitogenic in the absence of serum [46]. In a similar way a metabolic cluster was defined, including all DEGs of insulin treatment only, or insulin treatment in combination with glargine and/or X10 since all these compounds are known to have a strong metabolic effect in vivo. Using IPA we found 'ERK/MAPK' and 'p70S6K' signaling pathways significantly enriched in the mitogenic cluster, while the 'PI3K' and 'Cell cycle control' pathways were not enriched. For the metabolic cluster the IPA results were the other way around. We also performed IPA analysis on the individual treatment DEG lists. 'Cell cycle control of chromosomal replication' was highly enriched after treatment with compounds with a high affinity for the INSR (insulin, glargine and X10). Other metabolic processes like glycogen degradation and D-myo-inositol-5-phosphate metabolism were also enriched in the DEGs of these insulin molecules. On the other hand PI3K/AKT, IGF1, p53 and ERK/MAPK signaling were more enriched for the insulin-like molecules that also have a strong affinity for the IGR1R. signaling. For insulin signaling these pathways were also enriched but less significant.

A classifier gene set predictive for the pro-mitogen action of insulin analogues.
To evaluate which genes drive the strong mitogenic responses of IGF1R signaling we performed a variance test with selected genes showing a strong up or down regulation after strong activation of the IGF1R. For this we selected IGF1 and X10 exposures and contrasted this with the weak mitogenic response inducer insulin. We excluded glargine for the selection. In total we selected the top 10 hits in both the 1 hr and 6 hrs hit lists ( Figure 4A). Interestingly, many of these genes are known to play a role in mitogenic processes, including the early growth response (EGR) genes (all four EGR genes are in the top 20 gene list). Most of these genes have not directly been linked to the INSR or IGF1R signaling pathway so far. Next we validated these candidate genes using RT-Q-PCR with a separate independent set of samples. For 18 of these mitogenic classifier genes the RT-Q-PCR validation was successful and showed a highly similar trend for insulin, X10 and IGF1 conditions ( Figure 4B). For ZIC4 and ZMYND8 the expression was probably too low since no amplicon was detected even after 40 cycles. Finally, we evaluated the effect of glargine on the expression of these classifier genes. Importantly, the overall expression of the classifier genes after glargine treatment was more similar to X10 than insulin treatment. Expression is indicated as fold changes relative to unstimulated MCF7 IGF1R.

Validation of mitogenic classifiers through testing of commercially available insulin analogues.
We hypothesized that the expression of the mitogenic classifier genes could predict the mitogenic outcome of other insulin analogues. We performed an exposure experiment with MCF7 IGF1R cells including all commercially available insulin analogues (glargine, aspart, lispro, glulisine, determir). Since glargine showed expression of the predictive genes ( Figure 4B), and since glargine is rapidly metabolized into two metabolites (M1 and M2) in serum, we also included M1 and M2 in our study. A hierarchical clustering of the expression of all the tested classifier genes after stimulation with the different insulin analogues was performed ( Figure 5).
This resulted in the clustering of glargine with IGF1 and X10, while the glargine metabolites M1 and M2 clustered with other relatively non-mitogenic insulin analogues. We calculated a 'relative mitogenic potential' which was determined as the sum of the absolute values of log 2 fold changes of the expression of mitogenic classifiers of one compound treatment. As expected the 'relative mitogenic potential' was highest for IGF1 (69), followed by glargine (40) and X10 (31).

Validation of mitogenic classifier genes in vitro in primary human mammary gland cells and in vivo in mouse mammary glands.
To further validate the insulin analogue mitogenic classifier genes we tested additional in vitro and in vivo models. We first determined the robustness of the insulin analogue mitogenic classifier genes in primary cultured cells isolated from human mammary glands. These cells were anticipated to be the main target for increased mitogenic signaling of insulin analogues in diabetic patients. Primary cells were isolated from two independent individuals and exposed to the different insulin-like molecules. The activation of the INSR/IGF1R pathway was validated by Western blotting (Supplemental Figure 2) and a clear activation of the INSR/IGF1R as well as the PI3K/AKT signaling pathway was observed. Next the gene expression levels of three mitogenic classifier genes that were up regulated (EGR4 and TNFRSF11B) or down regulated (SLC1A2) in MCF7 IGF1R cells were measured ( Figure 6A). Although the fold change expression of these three genes in these primary human mammary cells was not as profound compared to the MCF7 IGF1R cells, in general the same direction of expression was observed. In addition, we investigated these three classifier genes in vivo in the mammary glands of mice treated with the different insulin analogues. In this experiment 40 wild type FVB mice received a subcutaneous injection of vehicle, insulin, glargine, X10 or IGF1. A very clear and constant drop in the glucose levels was observed 1 hour after the injections of insulin, glargine, X10 and IGF1, indicating that these compounds did induce the expected pharmacological response (Supplemental Figure 3A). The glucose levels returned to their normal levels (5 mmol/L), 6 hours after the injection. We then investigated the activation of the INSR and IGF1R (Supplemental Figure 3B and 3C). 1 hour after the insulin analogue injections a clear up regulation of p-AKT was observed, while after 6 hours the p-p70S6K levels were significantly (p=0.0022) up regulated. Also the insulin analogue mitogenic classifier genes showed a very clear modulation after treatment ( Figure 6B). Thus, EGR4 was even induced up to 18 times after IGF1 treatment and X10 also showed a clear up regulation of this candidate gene compared to no stimulation. Similarly after 6 hours treatment IGF1 induced TNFRSF11B and down regulated SLC1A2 levels. For these latter gene changes none of the other insulin-like molecules caused a significant change in gene expression. Gene expression in glargine conditions showed a similar trend as regular insulin, suggesting that glargine is rapidly metabolized into M1 and M2 in vivo, which are known to be compounds with a low pro-mitogenic signaling potential [46].

Discussion
It is well established that insulin has strong metabolic effects and in addition mild pro-mitogenic characteristics [148]. Small changes in the structure of insulin have improved the pharmacokinetic parameters so that the use of the insulin analogue is more convenient for diabetic patients. Yet, these small structural changes might also increase the binding affinity of insulin analogues towards the IGF1R and, consequently, increase the mitogenic potency of these molecules compared to regular insulin. Current in vitro systems that are used to determine mitogenic potential of insulin analogues are largely based on the proliferation capacity and do not take into account the molecular mechanisms of receptor signaling (Bronsveld 2015 manuscript submitted). In this study we used a transcriptomics approach to assess the preferential activation of pro-mitogenic signaling pathways by insulin analogues. We identified a subset of classifier genes that can be used to define the primary mode of action of insulin analogues. Moreover, we demonstrated that these classifier genes can be translated to primary human mammary cells as well as mouse mammary glands in vivo. These mechanism-based novel predictive genes are likely a more reliable method to classify the proliferative potency of insulin analogues that act preferably on the IGF1R.
For the safety profiling of insulin analogues this increased mitogenic potential is critical. Currently there is still a debate on the mechanism of such an increased mitogenic potential: on one hand the high binding affinity towards the IGF1R, while on the other hand a prolonged occupancy time towards the IRA is suggested causative [149]. In our current study we have been in the unique situation to evaluate these mechanisms in our MCF7 cell line panel. We found 40% more differentially expressed genes in the MCF7 IGF1R cell line after X10 and glargine stimulation compared to the MCF7 IRA cell line. These results suggest that the IGF1R is the main receptor that is mediating downstream pro-mitogenic signaling after insulin analogue stimulation. This is in line with our previous study in which we tested the proliferative potency of nine insulin-like molecules using the same MCF7 cell line panel and found that X10 and glargine induce proliferation more profoundly in the MCF7 IGF1R cells than the MCF7 IRA cells [46]. For this reason we based the further mitogenic classifier analysis on the MCF7 IGF1R cell line.
For the mitogenic classifier hit selection a training set was based on microarray gene expression of three compounds, in which insulin served as the reference compound with a low mitogenic potency. X10 and IGF1 served as two insulin-like molecules with a strong mitogenic potential. This resulted in a total of twenty genes either up or down regulated at 1 or 6 hrs after IGF1 and X10 treatment that most strongly differed from the insulin effect. Many of these genes have been Glargine is the most frequently prescribed anti-diabetic insulin analogue. There are conflicting conclusions regarding the intrinsic mitogenic potential of insulin glargine [127] [169]. We purposely excluded our glargine transcriptome analysis from the training set to identify candidate predictive classifier genes for pro-mitogenic signaling by insulin analogues. This allowed us to unravel the potency of glargine as a pro-mitogenic insulin analogue. Interestingly, the mitogenic potential of glargine was even higher compared to insulin X10 ( Figure 5). This is in full agreement with the kinase activation measurement of INSR/IGF1R pathway components in our previous study [46]. In diabetic patients glargine is rapidly processed by enzymes in the serum into two metabolically active compounds, M1 and M2, in which M1 is most prominent metabolite [83]. Therefore we also determined the mitogenic potential of M1 and M2 based on the gene expression profiles of the mitogenic classifier genes and we observed that both metabolites have a mitogenic score that is even lower than insulin. This is in agreement with previous studies in which the mitogenic potential was based on IGF1R binding affinity, kinase activation or proliferation assays [114] [46].
Two other studies also performed a mitogenic assessment of a panel of insulin analogues. These studies included a proliferation assay (Kurtzhals, Schaffer et al. 2000) and an IGF1R affinity evaluation [114]. We systematically compared our mitogenic classifier gene score with these two independent functional read outs (Figure 7). There was a striking correlation between our classifier scoring (based on Figure 5) and both the proliferation and IGF1R affinity. These combined data sets demonstrate that IGF1, X10 and glargine have a higher mitogenic potential compared to insulin, which is associated with a high affinity for IGF1R. Aspart and lispro have a mitogenic potency similar to each other. Determir and the two metabolites of glargine (M1 and M2) have a lower mitogenic index compared to regular insulin, associated with a low affinity for the IGF1R.
Some epidemiological studies suggest a correlation between insulin glargine use and breast cancer occurrence in the diabetic patients [76]. Since glargine might promote proliferation of mammary epithelial cells in vivo, we wanted to test whether expression of some of our classifier genes could be translated from MCF7 cells to primary human mammary cells. We could confirm a similar gene expression trend in primary human mammary cells after stimulation with insulin, glargine, X10 and IGF1 as in MCF7. Yet, the gene expression fold changes in the primary human mammary cells were far lower compared to the MCF7 IGF1R cells. The reason for this can be partly due to the isolation procedure which did not result in a pure population of mammary epithelial cells. Overall, the effect of glargine in the primary human mammary cells was not as profound as for IGF1 and X10. Previously we demonstrated that IGF1 and X10 significantly promote tumorigenesis in a conditional mammary gland tumor mouse model [153]. Glargine did not significantly enhance this tumorigenesis. Therefore we also evaluated the translation of our classifier genes to the in vivo situation and determined the gene expression changes in the mammary glands of mice that received a subcutaneous injection of the insulin-like molecules insulin, IGF1, X10 and glargine. We could validate the in vitro effect of IGF1 in the in vivo situation, indicating that a true IGF1R responses can be observed in this model. X10 showed some correlation with the effect of IGF1.
Yet, in contrast to the in vitro data, the gene expression profiles of glargine were more similar to insulin than to X10/IGF1. This effect is very likely caused by the metabolism of insulin glargine by factors in the serum of the blood of the mice (similar to the glargine conversion in human serum).
We therefore speculate that the observed pro-mitogenic signaling events of glargine in our in vitro breast cancer cell line models are presumably not occurring under in vivo conditions in the mammary gland. Yet, we cannot exclude that IGF1R mediated responses by glargine take place in other tissues in vivo.

Conclusions
In the current study we propose a new robust classifier gene set that allows the quick, robust and quantitative analysis of the pro-mitogenic potential of newly developed insulin analogues. These classifiers can be used within the pharmaceutical industry as well as in a regulatory setting to define the safety profile of insulin analogues as well as other growth factors that might act on the IGF1R.

Competing interests
There is no duality of interest that could be perceived as prejudicing the impartiality of the research reported. None of the authors declares any conflict of interest. shows the number of hits after insulin treatment, the second graph glargine, the third X10 and the last graph shows the number of hits after IGF1 treatment. B) The Venn diagrams of the late regulators (t=6hrs).

Supplemental Figure 2. Protein levels of primary human mammary cells stimulated with insulin like compounds.
A) Primary human mammary gland cells were treated with different insulin-like molecules followed by Western blotting for various INSR and IGF1R signaling pathway components. B) Quantification of Western blot data of the p-IGF1R/p-IR, p-Akt and p-Erk. The y-axis represents the average protein expression level compared to vehicle exposure.