Molecular predictors of response to a humanized anti–insulin-like growth factor-I receptor monoclonal antibody in breast and colorectal cancer

The insulin-like growth factor (IGF) signaling pathway is a major regulator of cellular proliferation, stress response, apoptosis, and transformation in mammalian cells, which is dysregulated and activated in a wide range of human cancers. The central components of this signaling module are the IGF-I receptor (IGF-IR), a homodimeric receptor tyrosine kinase, and its ligands IGF-I and IGF-II. Numerous studies have shown that ligand-mediated stimulation of IGF-IR results in receptor clustering and autophosphorylation followed by transphosphorylation of the β subunits (1). These phosphorylation events create multiple docking sites for the substrate adaptor proteins IRS1, IRS2, and SHC, which are essential transducers and amplifiers of IGF-IR signaling that recruit signaling complexes to the membrane and result in proliferative and antiapoptotic cellular responses (2). Mechanistic studies have shown that phosphorylation of IRS1 triggers activation of the phosphatidylinositol 3-kinase/Akt pathway and ultimately leads to sequestration and inhibition of the proapoptotic protein BAD as well as activation of the cell cycle initiator cyclin D (3), suggesting that inhibition of IGF-IR signaling may have both proapoptotic and antiproliferative consequences.

Alterations of key components of IGF-IR signaling have been shown to be associated with increased risk of cancer as well as neoplastic transformation. Specifically, high levels of circulating IGF-I have been shown to be associated with increased risk of developing breast, prostate, and colorectal cancer (4), whereas epigenetic loss of imprinting at the IGF-II locus has been shown to be common in colorectal cancer and to constitute a potential biomarker of colorectal cancer risk (5). In addition, genetic studies have shown that overexpression of IGF-I leads to neoplastic transformation in prostate epithelium (6), whereas overexpression of IGF-II in transgenic mice results in metastasizing mammary carcinomas, suggesting that these ligands can be key drivers of tumorigenesis when dysregulated and overexpressed (7). Several studies have suggested that IGF-IR expression is absolutely required for the acquisition and maintenance of a transformed phenotype in diverse genetic backgrounds and multiple cell types in vivo and in vitro (8–10). Taken together, the role of IGF ligands in driving neoplastic transformation and the requirement of receptor activity for maintaining the transformed phenotype have implicated the IGF axis as an attractive candidate pathway for therapeutic intervention.

Indeed, by one recent estimate, >25 molecules aimed at targeting IGF-IR as an anticancer therapy are currently in different stages of clinical and preclinical development at various pharmaceutical and biotechnology companies (11). The two predominant strategies to target IGF-IR are specific kinase inhibitors or monoclonal antibodies raised against IGF-IR that can block receptor function. We have described previously the development of a humanized, affinity-matured anti-human IGF-IR monoclonal antibody, h10H5, and shown that this agent has antitumor activity in mouse xenograft models and potently decreases Akt signaling as well as glucose uptake in preclinical models (12). The mechanism of action of h10H5 is similar to other blocking antibodies and involves blockade of ligand binding, cell surface down-regulation of receptor levels, and down-regulation of intracellular signaling mediated by Akt (12).

Although h10H5 is effective in inhibiting in vitro proliferation of many types of tumor cells, it lacks activity in others. Therefore, an important outstanding question in the clinical development of agents such as h10H5 is whether predictive diagnostic tests can be developed to identify appropriate patient populations, allowing specific treatment of patients whose tumors show addiction to this pathway for continued survival and proliferation. Previous studies have suggested that IGF-IR levels are strongly associated with preclinical response to a humanized anti-IGF-IR antibody in rhabdomyosarcoma cells and thus that levels of the target itself may constitute a predictive biomarker for response to IGF-IR-targeting antibodies in this indication (13). Others have looked at the predictive value of phosphorylation of IGF-IR itself or of the substrate IRS1 as markers of pathway activation that may predict response or at gene expression signature predictors of response to the small-molecule inhibitor BMS-536924 (11). These studies have provided interesting hypotheses that await clinical validation, but as yet studies looking broadly at response in other tumor types where IGF-IR may play an important role, such as breast and colorectal cancer, have not been reported. To address this question, we have identified putative predictive biomarkers of response to h10H5 in a large panel of breast and colorectal cancer cell lines with detailed accompanying molecular genetic characterization.

Cell lines described in this study were obtained from commercial sources and have been described previously (14, 15), with the exception of BT474EEI, a derivative of BT474 obtained by subculturing BT474 tumors grown in vivo in the absence of estrogen pellet supplementation (exogenous estrogen independent) as described previously (16). All breast cancer cell lines were plated out at 3,000 cells per well. Colorectal lines were plated out between 1,000 and 3,000 cells per well (depending on growth properties) in 10% fetal bovine serum (FBS)/DMEM and allowed to settle and recover overnight. The following day, the cells were washed in 0% FBS/phenol red–free DMEM and then serum-starved for 5 h in 0% FBS/phenol red–free medium. After serum starvation, medium containing 2.5% FBS was added back to the plates and the cells were dosed with IGF-IR antibody (h10H5) starting at a highest concentration of 10 μg/mL with 1:3 serial dilutions across the plate. Cells were incubated at 37°C for 72 h and then assayed in an ATP-based viability assay (CellTiter Glo Assay; Promega). Sigmoidal dose-response curves were fitted and EC₅₀ values were calculated using Prism software (GraphPad).

Western blotting experiments were conducted using standard protocols. The blotting antibodies used were IRS1, pIRS1, pAkt, Akt, mitogen-activated protein kinase, phospho–mitogen-activated protein kinase, cyclin D1, pS6, p27 (BD Bioscience), p4E-BP1, pIGF-IR, IGF-IR, and β-actin; antibodies were from Cell Signaling Technology unless otherwise stated. Quantitation of immunoblot bands was accomplished using NIH ImageJ software. Signal intensity was compared between lanes by normalization to total Akt protein levels.

MCF-7, SKNAS, and HCT15 cells were seeded into growth medium (50:50 DMEM/Ham's F-12 supplemented with 10% FBS, 2 mmol/L l-glutamine, 1× penicillin-streptomycin) in 24-well tissue culture plates (BD Falcon) at 100,000 to 200,000 per well and incubated overnight at 37°C in 5% CO₂. The h10H5 antibody was iodinated using the Iodogen method, and the radiolabeled antibody was purified from free [¹²⁵I]Na by gel filtration using a PD-10 column. The purified [¹²⁵I]h10H5 antibody had a specific activity of 13.3 μCi/ug. Cells were washed three times with binding buffer [50:50 DMEM/Ham's F-12 supplemented with 2% FBS and 50 mmol/L HEPES (pH 7.2)]. The washed cells were incubated for 4 h on ice with a mixture of a fixed concentration of ¹²⁵I recombinant human monoclonal antibody IGF receptor (∼139 pmol/L = 100,000 cpm per 0.2 mL) and decreasing concentration of the unlabeled antibody, which was serially diluted from 0.5 μmol/L in binding buffer for 13 concentrations assayed in triplicate. The cells were washed three times with binding buffer and lysed with lysis buffer [1% SDS, 8 mol/L urea, 100 mmol/L glycine (pH 3.0)]. The cell lysates were counted on a Wallac Wizard 1470 gamma counter (Perkin-Elmer Life and Analytical Sciences). The binding data were evaluated using Genentech's program New Ligand, which uses the fitting algorithm of Munson and Rodbard (17) to determine the binding affinity of the antibody and the total binding sites per well. Binding sites per cell was determined by averaging the cell counts of six wells from the assay plates and dividing the total binding sites by the averaged number of cells per well, and the assay was done in triplicate for each cell line.

All small interfering RNA (siRNA) experiments were done in phenol red free medium containing 10% FBS. OnTARGET Plus siRNA specific to human IGF-IR (Dharmacon), ESR1 (Dharmacon) or a control siRNA that does not target any sequence in the human genome (non-target control, NTC; Dharmacon) were used in transient transfection experiments. For IGF-IR and ESR1 knockdown the following siRNAs were used Human IGF-IR; ON-TARGETplus Set of 4 LQ-003012-00-0010, Human IGF-IR; ON-TARGETplus SMARTpool L-003401-00-0010, Human ESR1; ON-TARGETplus Set of 4LQ-003401-00-0010, Human ESR1. Optimal siRNA duplex and lipid concentrations were determined for each cell line. For the adherent cell line MCF-7 cells were plated at 8000 cells per well in a 96 well plate with 0.25 μL of Lipofectamine RNAiMAX (Invitrogen) and 25 nmol/L siRNA per well. Cells were incubated for 3 days in siRNA-containing medium and then h10H5 (IGF-IR antibody) was added for 3 days, followed by addition of Cell Titer Glo. To allow quantitation of knockdown, a duplicate plate was made for each cell line where no drug was added and RNA was collected using Qiagen TurboCapture 96 mRNA Kit. mRNA was directly converted to cDNA using the ABI cDNA archive kit (ABI). For quantitative reverse transcription-PCR analysis cDNA was diluted 1:10 and mixed with TaqMan Universal PCR Master Mix (ABI) and one of the following 20× primer probes: PPIA Hs99999904_m1 (housekeeping gene), UBC Hs00824723_m1 (housekeeping gene), ESR1 Hs01046818_m1, and IGF-IR Hs00609566_m1. Analysis was done using the ΔΔCT method, normalizing first to the average of the housekeeping genes and then to control siRNA-treated cells.

Breast and colorectal cancer cell lines were profiled on Affymetrix HGU133P Plus 2.0 microarrays as described previously (14, 15) and data have been deposited in the Gene Expression Omnibus database under accession numbers GSE12777 (breast data) and GSE8332 (colorectal data). Microarray data were analyzed and visualized using Spotfire and Cluster/Treeview software. Genes differentially expressed between sensitive and resistant colorectal cell lines were identified using the Cyber T algorithm, a modified t test that uses a Bayesian estimate of variance (18), and false discovery rates were estimated by the q-value method of Storey and Tibshirani (19). Cell lines were binned into sensitive and resistant classes using a cutoff of 20% growth inhibition (cell lines showing >20% inhibition in response to 1 μg/mL h10H5 were classified as sensitive and those showing <20% inhibition as resistant). Comparison with published cancer gene expression signatures was done in the Oncomine database (20).

Female nu/nu or irradiated BALB/c nude mice were inoculated s.c. with Colo-205 or MCF-7 tumor cells, respectively, or female nu/nu mice were inoculated with CXF-280 explant tumor fragments. Once tumors reached a mean volume of 130 to 260 mm³, mice were then randomized into groups of 8 to 10 mice and treated with vehicle or h10H5 at 1, 5, 15, or 20 mg/kg through i.p. injections. Tamoxifen was given as 5 or 10 mg 60-day slow-release drug pellets that were embedded s.c. Tumor volumes were measured in two dimensions (length and width) using UltraCal-IV calipers (Fred V. Fowler Company). The following formula was used with Excel version 11.2 to calculate tumor volume: tumor volume (mm³) = (length × width²) × 0.5.

The formalin-fixed, paraffin-embedded (FFPE) specimens were sectioned at 5 μm onto slides. After deparaffinization and rehydration, sections were processed for IGF-IR immunohistochemical analysis. Antigen retrieval was done using preheated Trilogy buffer (Cell Marque) at 99°C for 30 min. Endogenous peroxidase activity was quenched with KPL Blocking Solution at room temperature for 4 min. Endogenous avidin/biotin was blocked with Vector Avidin Biotin Blocking Kit (Vector Laboratories). Subsequently, sections were incubated with 2.5 μg/mL mouse anti-IGF-IR (clone 5E3; Genentech) monoclonal antibody in blocking serum for 60 min at room temperature and followed by incubation with biotinylated secondary horse anti-mouse antibody for 30 min. Streptavidin-conjugated horseradish peroxidase was applied for 30 min and signals were further enhanced by tyramide amplification. Metal-Enhanced DAB (Pierce Biotechnology) was used to develop the slides.

Forty-one breast cancer cell lines were assayed for in vitro sensitivity to the humanized recombinant anti-IGF-IR antibody h10H5 as measured in a 3-day ATP-based cell viability assay. Seven of the 41 cell lines were found to be sensitive, with EC₅₀ values below 1 μg/mL (Fig. 1A). Lack of sensitivity was associated with low expression of IGF-IR itself, because only 1 of 21 cell lines with expression below the median level for the panel was sensitive, whereas 6 of 20 cell lines with IGF-IR expression above the median were sensitive to h10H5 (P = 0.05, Fisher's exact test). Thus, whereas the positive predictive value of IGF-IR above median is relatively low at 28%, the negative predictive value of IGF-IR expression below median is 95%. This is consistent with a hypothesis wherein a minimal level of expression of IGF-IR is required for sensitivity to a biotherapeutic targeting this receptor but where expression alone is not sufficient to confer sensitivity. To investigate the role of the IGF signaling axis further in these cell lines, we also determined an IGF-I stimulation index, defined as the percent increase in cell growth of cells cultured in 1 ng/mL IGF-I compared with cells grown in serum-free medium, for a subset of the breast cancer cell lines. We found that IGF-I was most potent at stimulating cell growth in cells that show in vitro response to h10H5, whereas most nonresponsive cell lines had little or no proliferative response to IGF-I stimulation (Supplementary Fig. S1). This suggests a model where only a subset of breast cancer cells has a functional IGF-I/IGF-IR signaling axis that is linked to the cell cycle machinery and can respond to ligand-driven cellular proliferation and where cellular response to anti-IGF-IR-targeting therapies is only effective in the context of an active signaling pathway.

We also sought to identify additional molecular predictors of response to h10H5 and pathway activation in breast cancer using gene expression microarray data, because receptor expression alone could account for only approximately one-third of the sensitivity but were unable to identify any additional genes whose expression was associated with sensitivity in a statistically significant manner based on a false discovery rate below 10%. Because pathway components such as IRS1 and IRS2 have been implicated in response to other IGF-IR-targeting agents (21), we also looked in a directed manner at the relationship between expression of these genes and h10H5 response across the cell line panel (Fig. 1B). Through this analysis, we determined that, with the exception of SW527 (which expresses high levels of the ligand IGF-II; Fig. 1B), all of the sensitive cell lines expressed moderate to high levels of IGF-IR as well as high levels of either IRS1 or IRS2 (Fig. 1B). In addition, we found that a derivative of the BT474 cell line derived by in vivo passaging, BT474EEI (16), showed marked sensitivity to h10H5 that is not seen in the parental line (Supplementary Fig. S2A). Supervised analysis of gene expression differences between these two lines identified IRS1 overexpression as one of the most dramatic differences between these cell lines (Supplementary Fig. S2B), which otherwise are quite similar, again consistent with the hypothesis that high levels of IRS effector function are essential to enable cellular responsiveness to h10H5. This model predicts that high levels of IRS1 and IRS2 are important determinants of whether the IGF/IGF-IR signaling pathway is coupled to extracellular signaling and thus whether the pathway is active in a given cell line, in which case the function of these genes should be required for cellular proliferation in response to IGF-I. To test this, we used siRNA-mediated knockdown of IRS1 and found significant decreases in cell viability under IGF-I-driven growth conditions in the high IRS1-expressing cell lines, MCF-7 and BT474EEI (Supplementary Fig. S2C and D), showing that this adaptor plays an important role in proliferation in response to extracellular signals in these cells. These observations are consistent with a body of literature suggesting that IRS1 and IRS2 are required to mediate cellular responses to IGF-I/IGF-IR signaling, although somewhat surprising because IRS1 is generally regarded as the main adaptor linking IGF-IR signaling to proliferative responses (reviewed in ref. 22). Our results suggest it may be prudent to clinically evaluate expression of both IRS1 and IRS2 in conjunction with IGF-IR and IGF-II levels to ultimately determine utility of these putative biomarkers in patient selection.

The analyses described above have all focused on the identification of predictive biomarkers that could be used to select patients for therapy based on analyses of archival tumor tissue or pretreatment biopsies, but we were also interested in identifying pharmacodynamic biomarkers that might potentially allow assessment of drug activity by comparison of pretreatment and post-treatment biopsies. To address this, we treated both sensitive MCF-7 cells and resistant MDA-MB-231 cells with h10H5 for 24 h and then examined levels of key signaling proteins and their phosphorylated isoforms by Western blotting. In resistant MDA-MB-231 cells, we detected low levels of IGF-IR and observed that h10H5 treatment caused down-regulation of total receptor as well as decreases in pAkt (S473), although levels of these analytes were low and accurate quantitation was difficult (Fig. 1C; Supplementary Fig. S3). In addition, we observed minimal effects on distal markers such as pS6 or p4E-BP1 in MDA-MB-231 cells. Similar analyses in sensitive MCF-7 cells treated with h10H5 also showed down-regulation of total and phosphorylated receptor as well as decreases in pAkt (S473), suggesting that these proteins and phosphoproteins might have utility as biomarkers of target modulation. In contrast to the MDA-MB-231 results, we observed that h10H5 treatment in MCF-7 cells resulted in a 50% increase of the negative cell cycle regulator p27 and a 50% decrease in levels of p4EB-P1 (S65; Fig. 1C; Supplementary Fig. S3), suggesting that distal outputs of the phosphatidylinositol 3-kinase/Akt pathway on cell cycle and translational components may correlate with efficacy in response to h10H5 treatment. Assays for such analytes might thus be used to monitor patient response to anti-IGF-IR therapies, potentially providing an early indication of therapeutic benefit and also giving information on optimal biological doses for such therapies.

Because our studies suggest that mRNA levels of the target itself may have predictive value in determining efficacy in response to an anti-IGF-IR-targeting antibody, we developed an immunohistochemical assay to show that IGF-IR protein levels were also associated with response to h10H5. Such an assay might also be useful for patient stratification in clinical trials. Initial validation was done on a tissue microarray constructed from FFPE cell pellets derived from 42 breast cancer cell lines for which accompanying gene expression microarray data were available. This allowed comparison of IGF-IR mRNA levels in each cell line with protein staining intensity determined by immunohistochemistry (Fig. 2A) and showed overall excellent agreement between these two different methods of determining target levels, suggesting that the immunohistochemical assay is faithfully reading out IGF-IR levels.

Previous studies have shown that IGF-IR number plays an important role in both mitogenic response and transformation in mouse embryonic fibroblasts and that 15,000 receptors per cell were sufficient to render mouse embryo cells competent to grow in serum-free medium supplemented solely with IGF-I (23). To gain a better understanding of the number of receptors that might be required for a cell line to respond to h10H5 in our in vitro assays and also how our immunohistochemistry score reflects receptor density, we performed radioligand cell binding assays and Scatchard analysis to determine IGF-IR density in three cell lines. We found that h10H5-sensitive MCF-7 cells (immunohistochemistry 3+) had on average 23,622 receptors per cell, h10H5-sensitive SK-N-AS cells (immunohistochemistry 2+) had on average 9,494 receptors per cell, and h10H5-resistant HCT15 cells (immunohistochemistry 1+) had on average 1,382 receptors per cell (Supplementary Fig. S4). Thus, it seems that somewhere between 1,382 and 9,494 IGF-IR molecules per cell are likely to be required for an in vitro antiproliferative response to this antibody. These numbers generally agree with a previous study suggesting minimal anti-IGF-IR-mediated growth inhibition of rhabdomyosarcoma cell lines with <3,000 receptors per cell and maximal inhibition of cells with >30,000 receptors per cell (13), although our results do suggest that in some contexts cells with lower receptor numbers (∼10,000 per cell) are sensitive to an anti-IGF-IR-targeting antibody.

We also used our immunohistochemical assay on a series of breast and colorectal tumor samples and showed that in both tissues a wide range of IGF-IR expression is detectable by this assay, with 60% of colorectal samples and 54% of breast cancer samples exhibiting strong staining (immunohistochemistry 2+ or 3+; Fig. 2B and C). Thus, this immunohistochemical assay may be a valuable tool for evaluating IGF-IR levels as a patient stratification biomarker in clinical samples. Because our studies also implicate components of IGF-IR signaling such as IGF-II and the adaptors IRS1 and IRS2, we developed a multiplex quantitative reverse transcription-PCR assay that may be used to assess levels of all of these biomarkers in FFPE tumor specimens. The multiplex assay was validated using control FFPE cell pellet RNA and comparison with microarray data from matched samples (data not shown). We applied the assay to RNA prepared from FFPE tumor material and showed a wide range of expression of these potential biomarkers 4(Fig. 2D), suggesting that such an assay could be used to clinically test the hypotheses that high expression of IGF-IR and IRS1 or high expression of IGF-II might identify responsive patients.

Because breast cancer molecular subtypes are relatively well understood and provide a framework for other targeted therapies [e.g., tamoxifen or aromatase inhibitors in estrogen receptor (ER)–positive breast cancer], we sought to determine whether the IGF-IR pathway was associated with particular breast cancer subtypes and whether this might provide a contextual basis for developing anti-IGF-IR therapies in breast cancer. In particular, we noted that high IGF-IR expression was associated with ER status, because 13 of 21 cell lines with IGF-IR expression above the median were ER positive, but only 3 of 20 cell lines with below median IGF-IR expression were ER positive (P = 0.003, Fisher's exact test). To confirm that this association was not a cell line–specific phenomenon, we analyze microarray data from 111 human breast tumors for expression of IGF-IR, IGF-I, and ER (encoded by the ESR1 gene) and found that high IGF-IR expression was significantly associated with ESR1 transcript levels in this data set (P < 0.001, Mann Whitney U test; Fig. 3A). We also note that IGF-IR is a member of the “intrinsic set” of breast cancer subtype classifier genes and is associated strongly with the luminal, hormone receptor–positive subtype (24).

We next sought to address the functional relationship between ER and IGF-IR expression in breast cancer as well as the consequences of dual blockade of these pathways on cell viability. First, we performed siRNA-mediated knockdown of both ESR1 and IGF-IR in ER-positive MCF-7 cells using both siRNA pools and two individual siRNA duplexes. Quantitative reverse transcription-PCR analysis of lysates prepared from these cells showed that the siRNAs targeting each gene efficiently knocked down their respective targets (Fig. 3B). In addition, we found that each of the ESR1 siRNAs resulted in a 30% to 40% reduction in IGF-IR levels and each of the IGF-IR siRNAs resulted in 40% to 50% reduction in ESR1 levels. These results suggest that IGF-IR transcript levels are positively regulated either directly or indirectly by the ER, and ESR1 levels are likewise regulated by IGF-IR signaling and are consistent with previous reports suggesting extensive crosstalk between these pathways (25, 26). One implication of this finding is that therapeutic agents such as fulvestrant or tamoxifen that target ER (27) might enhance the effects of anti-IGF-IR antibodies on cell viability. To test this, we performed in vitro combination studies with h10H5 and fulvestrant under both normal FBS and medium conditions as well as in phenol red–free medium with charcoal-stripped FBS, because previous studies have suggested that phenol red can act as an estrogen mimetic and FBS may contain traces amounts of estrogens (28). Consistent with this, we noted that growth of MCF-7 cells is substantially more inhibited in the phenol red–free charcoal-stripped FBS than in normal medium, suggestive of the presence of estrogens obscuring response to h10H5 in normal medium (Supplementary Fig. S5). In addition, we found that the addition of fulvestrant to h10H5 resulted in substantially greater inhibition of cell growth than either single agent alone in normal medium (Fig. 3C).

We next sought in vivo confirmation of the synergistic interaction between h10H5 and antiestrogen-targeting therapeutics in nude mice harboring s.c. implanted MCF-7 xenograft tumors (Fig. 3D). In this experiment, we found that once weekly h10H5 had no detectable tumor growth inhibition at the dose and schedule examined, perhaps reflective of the fact that in vivo propagation of these tumors requires estrogen pellets, and consistent with our in vitro studies showing that estrogen signaling up-regulates IGF-IR and may mask the effects of an IGF-IR-targeting antibody. However, we did observe significantly greater tumor growth inhibition when tamoxifen was combined with h10H5 (P < 0.001) compared with tamoxifen alone, suggesting that dual targeting of these pathways results in greater antitumor effects than either single agent alone in this xenografted model (Fig. 3D). Thus, a diagnostic strategy suggested by our results in breast cancer would be enrichment for patients with high IGF-IR-expressing tumors by focusing clinical development on ER-positive cancers based on the observation that high IGF-IR expression occurs predominantly in this subset of breast cancer. Focusing on a disease subtype might be a surrogate approach to screening directly for receptor levels.

We next determined the responsiveness of a panel of 27 colorectal cell lines to h10H5 in an effort to identify molecular correlates of response in this tumor type (Fig. 4A). We found that overall 9 of the 27 cell lines were sensitive and had EC₅₀ values of <1 μg/mL, suggesting relatively strong dependence on IGF-IR signaling in this tumor type. IGF-IR expression itself showed a trend toward higher levels in sensitive models, because 7 of 13 cell lines with IGF-IR expression above the median for the panel were sensitive compared with only two cell lines with expression below the median. The negative predictive value was not as strong as seen in breast cancer and the trend did not reach statistical significance. We also noted that overall expression levels of IGF-IR were correlated with percent inhibition in response to h10H5 (R² = 0.33; Fig. 4B), again suggesting possible diagnostic utility of receptor levels and consistent with previous reports that levels of IGF-IR are correlated with mitogenicity, transformation, and adhesion phenotypes (23, 29). We also looked at pharmacodynamic response to h10H5 in sensitive and resistant colorectal models and observed similar results to those in breast cancer; that is, we observed substantial h10H5-mediated down-regulation of pAkt (S473) in both sensitive HT-29 cells and resistant HCT-116 cells (Fig. 4C) but saw more pronounced effects on distal markers such as p27, pS6, and p4E-BP1 specifically in the sensitive cell line (Fig. 4C).

Because IGF-IR levels alone do not explain all of the sensitivity and resistance seen in colorectal cell lines, we sought to identify a molecular signature of anti-IGF-IR response by supervised analysis of gene expression microarray data. This analysis identified 75 probes corresponding to 60 genes that are differentially expressed between sensitive and resistant lines with a false discovery rate of <10% (Fig. 5A; Supplementary Table S1). Reassuringly, IGF-IR itself was identified through this unbiased analysis as one of the top genes predicting sensitivity. In addition, pathway analysis implicated components of Wnt signaling such as Wnt-11 and β-catenin as negative predictive factors in response, suggesting that activation of parallel signaling pathways may render cells less sensitive to the inhibitory effects of anti-IGF-IR antibodies. This analysis also identified factors that regulate ubiquitination (e.g., Trim36) and trafficking such as Rab family members, as well as negative regulators of the cell cycle such as Tob1, as additional candidate biomarkers of response. Finally, the P-selectin ligand CD24 also showed significant positive association with h10H5 sensitivity (Fig. 5A and B). Expression of CD24 has been shown to be a poor prognostic marker in colorectal cancer (30) and to be associated with a cancer stem cell phenotype (31), suggesting a possible role for IGF-IR targeting in a clinically important subpopulation of colorectal cancer. Based on this, it is intriguing to note that a recent report showed that colorectal cancer models selected for resistance to 5-fluorouracil or oxaliplatin manifest a stem-cell like phenotype and enhanced sensitivity to an anti-IGF-IR-targeting antibody (32). A model for how some key components of the signature may relate to IGF-IR signaling is shown in Fig. 5C. To assess the relationship of this colorectal response signature to other published gene expression signatures, we queried the Oncomine database, a compendium of 18,000 cancer-related gene expression microarrays (20, 33). This analysis assesses overlap between the query signature and signatures in the database by generating 2 × 2 contingency tables and then performing a Fisher's exact test to assess statistical significance between the datasets. Intriguingly, querying this database with the colorectal cancer h10H5 response signature revealed a highly significant relationship (P = 7.12 × 10^-5) to a published data set from MCF-7 breast cancer cells treated with IGF-I (34). Components of the signature such as TOB1, CD24, MAP2K6, and SMAD6 were all found to be down-regulated on IGF-I treatment (Supplementary Fig. S6). Thus, expression of these putative markers not only correlates with anti-IGF-IR activity but also is functionally affected by signaling through the pathway, strengthening the rationale for evaluation of this signature a potential predictor of patient response to anti-IGF-IR-targeting therapies.

We next sought in vivo confirmation of h10H5 activity in both high IGF-IR-expressing and high IGF-II-expressing models by selecting select representative xenograftable cell lines or tumor explants to test each hypothesis. We first tested h10H5 activity in nude mice harboring s.c. implanted Colo-205 xenograft tumors, because this model expresses high levels of IGF-IR (Fig. 6A) and is sensitive to the effects of h10H5 in vitro. We observed substantial tumor growth inhibition at an h10H5 dose of 20 mg/kg in this model (Fig. 6B), providing in vivo proof-of-concept that anti-IGF-IR antibodies may show benefit in colorectal cancers expressing high receptor levels. Colorectal cancers also frequently express high levels of IGF-II ligand (Fig. 6A), so we also sought to determine whether h10H5 had antitumor activity in primary tumor explant model CXF-280, which expresses high levels of IGF-II but low levels of IGF-IR. Such models are derived from patient tumors that have been transplanted s.c. directly into nude mice. They are reported to have maintained their typical tumor histology, including a stromal component and vasculature (35), and hence may be somewhat more representative of actual patient tumors than xenografted cell lines. Notably, we found that h10H5 at doses of 5 or 15 mg/kg substantially reduced tumor growth compared with vehicle or a control antibody in CXF-280 explants (Fig. 6C) and also significantly delayed time to tumor progression for both doses of h10H5 compared with control antibody-treated animals (data not shown; log-rank P = 0.03 for 15 mg/kg group and P = 0.02 for 5 mg/kg group). In addition, antitumor activity of h10H5 has been shown previously in tumor xenograft models of the breast tumor cell line SW527 and the neuroblastoma cell line SK-N-AS (12) and we note that both of these models express high levels of IGF-II (Supplementary Fig. S7), again suggesting a potential role for receptor targeting in situations where tumor growth may be driven by autocrine growth loops involving IGF-II production by the tumor. Thus, our in vivo data are consistent with anti-IGF-IR-directed biotherapeutics having activity in tumors that express components of the signaling pathway and support clinical evaluation of pathway-focused diagnostic tests.

The major aim of this study was to identify predictive diagnostic biomarkers to help inform patient stratification efforts during clinical development of an anti-IGF-IR antibody in solid tumor malignancies, particularly breast and colorectal cancer. A key challenge to diagnostic development that must be met if a diagnostic is to be codeveloped with a drug is the identification of clinically useful biomarkers in the absence of clinical data. To address this challenge, we have used preclinical studies in well-characterized panels of cell lines and tumors to evaluate putative predictive biomarkers based on close connection to the pathway biology of IGF-IR signaling and also to identify novel biomarkers using unbiased pharmacogenomic analysis. These studies have yielded insights into the potential diagnostic utility of the target itself (IGF-IR), as well as key ligands and associated molecules (IGF-II, IRS1, and IRS2), and in addition have identified a gene expression signature associated with response in colorectal cancer.

Our data suggest that in breast cancer, and to some extent in colorectal cancer, expression of IGF-IR is necessary but not sufficient for antitumor activity. Thus, stratification of patients based on tumor IGF-IR levels may have utility in identifying patients unlikely to respond due to weak pathway activity. Previous studies have examined the role of role of IGF-IR number in IGF-I-mediated mitogenesis and transformation of mouse embryo fibroblasts and found that transfected cells with <15,000 IGF-I receptors per cell were not competent to grow in serum-free medium supplemented solely with IGF-I (23). Our studies suggest that somewhere between 1,300 and 10,000 receptors per cell are necessary for in vitro response to an anti-IGF-IR antibody and that models with low expression of IGF-IR are unlikely to exhibit dependence on this pathway. Our results are also consistent with previous detailed studies in rhabdomyosarcoma that showed an association between IGF-IR levels and antiproliferative activity of an anti-IGF-IR antibody (13). Together, these results support clinical evaluation of immunohistochemistry-based assays to predict response to anti-IGF-IR-targeting agents. The use of immunohistochemistry as a companion diagnostic test for patient selection has met with mixed success in the context of receptor-targeting biological agents for cancer treatment. In one case, immunohistochemical screening for HER2 status has been shown in both prospective and retrospective analysis to be useful in selecting responsive patients for trastuzumab (36, 37). However, it should be noted that routine testing in community laboratories has been proven challenging and efforts are still ongoing to increase accuracy of testing many years after the initial approval of trastuzumab (38). In a second case, the epidermal growth factor receptor–targeting antibody cetuximab was specifically approved in colorectal cancer patients determined to have epidermal growth factor receptor expression through an immunohistochemical test. Initial clinical studies evaluated cetuximab activity specifically in patients selected to have positive immunohistochemical staining, but subsequent studies have shown similar response rates in patients who do not express epidermal growth factor receptor (39). Thus, evaluation of both diagnostic-positive and diagnostic-negative patients will be important to validate the clinical utility of an IGF-IR immunohistochemical test.

The in vivo responsiveness we observe in high IGF-II-expressing xenograft models suggests that ligand expression may also have predictive value in identifying tumors with addiction to IGF-IR pathway activity mediated by autocrine growth factor stimulatory loops. We note that an IGF-IR-specific inhibitor has been shown to have excellent preclinical antitumor activity in Ewing's sarcoma (40), and early reports at meetings have suggested that patients with this disease are highly sensitive to anti-IGF-IR-targeting agents (discussed in ref. 22). A possible explanation is the described existence of an autocrine growth-stimulatory loop involving IGF-I/IGF-IR that is present in both Ewing's sarcoma cell lines and clinical samples and can be interrupted by treatment with an IGF-IR-blocking antibody (41). Addiction to IGF-I/IGF-IR signaling may be intimately tied to the genetics of Ewing's sarcoma, because EWS-FLI1 translocation that is hallmark of Ewing's sarcoma has been shown to up-regulate IGF-I and to require the expression of IGF-IR (reviewed in refs. 22, 42). Another putative autocrine loop, this time involving tumor overexpression of IGF-II, has also been implicated as a predictive factor of anti-IGF-IR response in synovial sarcoma (43), suggesting possible parallels to the in vivo activity we see in xenograft models that overexpress IGF-II.

An additional important factor in response to anti-IGF-IR-targeting agents in breast cancer seems to be expression of high levels of either of the substrate molecules IRS1 and IRS2, which is somewhat surprising and clearly requires further validation, because IRS1 is more directly coupled to proliferative responses to IGF-I than IRS2 in previous studies. Specifically, it has been shown that, without IRS1 expression, IGF-I actually induces differentiation of hematopoietic cells and that restoration of IRS1 expression allows cellular proliferation in response to IGF-I (44). Moreover, studies in breast cancer cells normally lacking IRS1 and IRS2 expression have shown that expression of IRS1 is required to mediate the proliferative response to IGF-I and expression of IRS2 to mediate a motility response to IGF-I (21). Finally, expression of IRS2 in IRS1-null mouse embryonic fibroblasts has been shown to reconstitute IGF-I activation of phosphatidylinositol 3-kinase and immediate-early gene expression to the same degree as expression of IRS1 but to have only a minor effect on IGF-I stimulation of cell cycle progression, suggesting that IRS1 and IRS2 are not functionally interchangeable adaptors for stimulation of mitogenesis (45). It will thus be interesting to see if expression of IRS1, IRS2, or both will have clinical utility in predicting patient response to anti-IGF-IR-targeting agents.

It has recently been proposed that the concept of “oncogene addiction” be modified to encompass “oncogene and/or growth factor addiction” in the case of the IGF-I/IGF-IR axis (22). We suggest that the pathway-focused biomarkers we identify may have clinical utility in identifying patients with tumors displaying the hallmarks of IGF-IR pathway addiction and that a prudent approach may be to stratify patients based on levels of expression of key pathway components and thus determine through clinical studies whether these putative biomarkers will be useful in enriching for patients more likely to respond to anti-IGF-IR-targeting therapies.

All authors are either current or former employees of Genentech. No other potential conflicts of interest were disclosed.

We thank Sedita Lakic for excellent administrative assistance, Jennifer Batson for assistance with in vivo studies, and Rob Akita for help with receptor density studies.