FANCD2 Is a Potential Therapeutic Target and Biomarker in Alveolar Rhabdomyosarcoma Harboring the PAX3–FOXO1 Fusion Gene

There are approximately 350 new cases of pediatric rhabdomyosarcoma (RMS) diagnosed per year in the United States (1). Despite progress made in this disease, control of the primary site is still a major source of treatment failure, as two out of three of the relapses observed in Intergroup RMS Study-IV (n = 888) were local-regional (2), suggesting an inherent resistance to radiotherapy in this childhood malignancy. Alveolar RMS (ARMS) is the most biologically aggressive subtype of this disease, which confers a 5-year overall survival of less than 50% (10%–30% for metastatic disease) compared with the more favorable, embryonal subtype (ERMS), which confers a 73% 5-year overall survival. Driving the poor prognosis in ARMS is the presence of the t(2;13)(q35;q14) translocation observed in 60% to 70% of ARMS, which gives rise to the chimeric transcription factor, PAX3–FOXO1, a constitutively active oncogene that significantly alters gene expression and confers an 8% 4-year overall survival to patients with this disease (3). The impact of PAX3–FOXO1 fusion gene is so pervasive that it seems to segregate RMS into two distinct molecular and clinical groups: those harboring this translocation (t-ARMS) and all other subtypes, including ERMS, nontranslocated ARMS, and ARMS harboring the PAX7/FOXO1 fusion gene (though its clinical virulence is less clear; ref. 4). Clinical outcome improvements in the treatment of t-ARMS will have the greatest impact upon the pediatric RMS population as a whole. Furthermore, lowering radiation doses required to eradicate gross and microscopic disease reduces the likelihood and severity of acute and long-term toxicity, as total dose and treated volume are directly related to their development, especially in children (5, 6).

Recently, Guo and colleagues demonstrated that the conditional knockout of mTOR resulted in the loss of the Fanconi anemia group D2 (FANCD2) mRNA and protein expression in hemopoietic stem and progenitor cells (7). This resulted in a significant gain of DNA double-strand breaks (DSB) and cell sensitivity when exposed to DNA cross-linking chemotherapeutic agents. This result suggested that mTOR inhibition leads to loss of FANCD2 expression, leading to more accumulated DNA damage and hence cell kill. Interestingly, the activation of the mTOR/Akt pathway seems to be prognostic in pediatric RMS, with high levels of phospho-Akt and phospho-P70S6K (and other intermediates) having significant associations to poorer overall and progression-free survival in patient samples; findings which prompted preliminary in vitro testing with mTOR inhibition (via CCI-779) that resulted in significant reductions in cell growth (8).

New inhibitors have been developed to inhibit both TORC1 and TORC2 complexes (9, 10) and may more closely approximate the effects of mTOR knockdown than the first generation of mTOR inhibitors, which primarily inhibit TORC1. AZD8055 (AstraZeneca), a novel ATP-competitive inhibitor of mTOR kinase activity, acts as a dual TORC1 and TORC2 inhibitor and was used in our study. AZD8055 was recently tested within the Pediatric Preclinical Testing Program, causing only modest growth inhibition against some of the RMS lines (11), with otherwise nontoxic effects on mice. Further in vivo testing of AZD8055 in this setting revealed that indeed there was a drug-induced downregulation of FANCD2 through mTOR inhibition. Specifically, FANCD2 gene transcripts were reduced via the blockade of the mTORC1-S6K1 pathway (12). With these data taken collectively, we postulated that a complete mTOR blockade via pharmacologic means could potentially enhance radiation sensitivity in pediatric RMS.

The FANCD2 protein in humans is encoded by the FANCD2 gene and belongs to Fanconi anemia complementation group (FANC) and is a crucial mediator of this DNA repair pathway (13). Mutations in any one of the genes comprising this complementation group can lead to the development of Fanconi's anemia, often resulting in elevated cancer risks, congenital abnormalities, and bone marrow failure. Mutations within FANCD2 result in most severe phenotype (14). Clinically, patients with FA also display hypersensitivity toward agents that create DNA interstrand cross-links, including many chemotherapeutic drugs as well as ionizing radiation (15, 16). There is also growing experimental evidence suggesting that the FA pathway plays a role in the repair of spontaneously generated DNA DSBs as well (17).

In this report, we tested the efficacy of concurrent mTOR kinase inhibition (in the hope to suppress FANCD2 expression via AZD8055 administration) and radiotherapy using an in vivo model that robustly recapitulates tumor biology and clinical radiation therapy delivery. We further analyzed the pharmacologic impact of AZD8055 on mTOR signaling during the response to radiotherapy as well as the expression and functional dynamics of FANCD2 in xenograft specimens collected in parallel to this preclinical screen. We then validated the importance of FANCD2 in radiation resistance in RMS using siRNA-based knockdown clonogenic survival studies. Finally, we explored the clinical significance of FANCD2 levels as a biomarker in pediatric RMS using two independent, clinically annotated patient sets.

The xenograft lines used for the preclinical testing experiments (obtained from the Pediatric Preclinical Testing Program, Nationwide Children's Hospital, Columbus, OH) were two RMS lines: Rh18 (embryonal) and Rh30 (alveolar) which were propagated and grown in mouse flanks as previously described (18). These two lines express the two major RMS genotypes: Rh30 (t-ARMS-Rh30) expresses the fusion transcription factor PAX3–FOXO1, whereas Rh18 (ERMS-Rh18) does not. Furthermore, these RMS xenograft lines were selected for the current study because they were not exposed to cytotoxic agents before their harvesting from patients and direct engraftment into mice (19). CB17SC scid^−/− female mice (Taconic Farms) were used to propagate subcutaneous tumors.

Mice bearing each xenograft line were dosed orally with 1 mg/mL AZD8055 at 5 to 20 mg/kg of body weight for 21 consecutive days beginning on study day 1, as previously described (11). Thereafter, body weights were measured weekly for dose calculation. Mice received fractionated flank-irradiation treatments in clinically relevant 2 Gy daily doses as previously described using our proprietary mouse-flank irradiation device (18, 20). For mice receiving concurrent therapy, radiation was given 2 to 4 hours after drug administration on days 8 to 12; irradiation treatment resumed on days 15 to 19 after a 2-day break (reflecting the weekends off of typical treatment courses). The treatment schedule is summarized in Supplementary Fig. S1A. Mice receiving concurrent treatment were treated with AZD8055 for 1 week before radiation, as at least 4 days were required for the suppression of FANCD2 (12), due to the relatively long half-life of the protein. Furthermore, mice were monitored daily for skin reaction starting on day 15 from the first day of irradiation treatments through day 35. A numeric grade is given based on a rating system according to the severity of the reaction detailed in Supplementary Fig. S1B. Dose density, complete response rates, recurrence rates, 12-week failure rates, event-free survival, and treatment-related events (as previously described; ref. 18) were recorded. To quantify radiosensitization impact, an enhancement ratio, [(% failure-free rate with XRT+drug)/(dose density with XRT+drug)]/[(%failure-free rate with XRT-alone)/(dose density with XRT-alone)] was used. A ratio of 1.0 signifies no enhancement.

Tumors collected from mouse xenografts and flash frozen in liquid nitrogen were homogenized via mortar and pestle. The prepared tissue was then transferred to a cell lysis solution containing 1% protease inhibitor cocktail (Sigma #P8340), 1% phosphatase inhibitor cocktail 2 (Sigma #P5726), 1% phosphatase inhibitor cocktail 3 (Sigma #P0044), and 1× phenylmethylsulfonylfluoride in T-Per Tissue Protein Extraction Reagent (Thermo Scientific #78510). The extracted protein concentrations from the resulting samples were then assessed using a BCA protein estimation kit (Pierce). From lysis samples representing tumors from mice that had undergone the same treatment, equivalent amounts of proteins were pooled into a combined sample for use in Western blot analysis. Protein from cells maintained in vitro without antibiotics were also extracted using the aforementioned cell lysis solution.

To quantify differences in protein levels, immunoblot analyses were imaged using Bio-Rad's VersaDoc MP 4000 System and software (Bio-Rad). Relative density values (intensity/mm^2) of blots for FANCD2 and β-actin were determined for the 0 Gy, 2 Gy, 4 Gy, and 6 Gy XRT Alone and XRT+AZD8055 samples using the Volume Rect Tool and Volume Analysis Report provided by the software. For each sample, a normalization value was determined by dividing the FANCD2 density by the β-actin density. This was performed for two separate immunoblot experiments. SD values were determined for each sample's normalized values. Each normalized value was then renormalized relative to the 0 Gy XRT Alone control and plotted.

The FA triple staining immunofluorescence method using formalin-fixed paraffin-embedded (FFPE) tissues has been previously described (21). FFPE tumor tissue was cut at 4 microns, placed on positively charged slides, and stained with hematoxylin and eosin. Additional sections for immunofluorescence staining were placed in a 60°C oven for one hour, cooled, deparaffinized, and rehydrated through xylenes and graded ethanol solutions to water in standard fashion. Antigen retrieval was performed by placing slides in Dako's TRS antigen retrieval solution (Dako) in a calibrated vegetable steamer (Black and Decker) at 94°C for 25 minutes. Slides then were placed on a Dako Autostainer for automated staining. The tissue sections were incubated with a primary antibody cocktail of rabbit polyclonal FANC-D2 antibody (Novus Biologicals) at a dilution of 1:1,000 and a monoclonal anti-Ki67 mouse antibody (Dako) at a dilution of 1:150, for one hour at room temperature. Sections then were coincubated with a secondary antibody (FITC conjugated to anti-rabbit IgG and Alexa fluor 594 donkey anti-mouse IgG, Invitrogen) at 1:1,000 for one hour at room temperature. All rinses were performed on the autostainer with TBST. The sections were mounted on glass slides using a 4′ 6-diamidino-2-phenylindole (DAPI)-containing embedding medium (Vysis Dapi 1, Abbott Laboratories). FANCD2 foci-negative cells (PD220 or PD20) and foci-positive cells (MCF-7 or FA corrected PD220 or FA corrected PD20) were used as controls on the sample slide during the procedure. The slides were analyzed under a Nikon E-400 fluorescence microscope. Cells with three or more bright green foci in the nuclei were scored as FANCD2 positive, and three similar fields (at ×100 magnification) of cells were enumerated for each time point.

In vitro transfections were carried out via INVITROGEN's Lipofectamine RNAiMAX reverse transfection procedure. The true transfection group was transfected using equal parts of two FANCD2 siRNAs provided by SIGMA-ALDRICH: SASI_Hs01_00137854 and SASI_Hs02_00307032. The negative control group was transfected using SIGMA-ALDRICH's MISSION siRNA Universal Negative Control #1. All transfections took place in T-25 tissue culture flasks. Cells were initially plated at approximately 40% to 50% confluence per flask. The flasks were incubated for a total of 48 hours posttransfection at 37°C, 5% CO₂ before receiving irradiation and immediate plating for clonogenic survival analysis.

Rh30 and Rh18 in vitro cells lines (also obtained from the PPTP) were both separated into three groups: control, negative (scrambled) siRNA, and FANCD2 siRNA. The plated cells were allowed to grow into colonies for 14 days before being stained with 0.05% crystal violet and counted. The 0 Gy control plates for each group were used to calculate a plating efficiency as defined in Franken and colleagues 2006 Nature Protocols.

The initial set consisted of 101 publicly available clinically annotated pediatric rhabdomysarcoma specimens with Affymetrix U133_Plus 2.0 Array. CEL files from a previously published report (4). The second set consists of 124 clinically annotated pediatric RMS patient core samples that comprise the tissue microarray slides available from the Children's Oncology Group banking protocol for soft tissue sarcomas (COG D9902). The PAX/FOXO1 fusion status was known for the majority of cases.

The log-rank test was used in univariate comparisons of survival times (event free in mice and overall survival in patient cases), and Cox regression analysis was used in the multivariate analysis (backward step-wise Wald). Actuarial outcome rates and factor distributions between groups were determined using the Fisher exact test. Spearman rank was used for determining the correlation between factors. Statistical analysis was performed using SPSS software.

Of note, 101 RMS samples arrayed by Affymetrix U133 Plus 2.0 platform were downloaded from Array Express (4) website. The chips were processed by RMA with background adjustment and then quantile normalized via using Matlab Statistics toolbox. All analyses described herein were performed at a gene level. The gene level data were created using the Gene Set Enrichment Analysis (GSEA) application. An individual gene expression level was derived from maximum of all probes that mapped to a particular gene. For the molecular similarity analysis of RMS: Non-negative matrix factorization, a technique to elucidate global molecular patterns (22), was applied to the 3,000 most variable genes as determined by coefficient of variation. GSEA settings: 1,000 phenotype permutations were performed to assess the level of significance for gene set member rank distribution in FANCD2 defined signal-to-noise profile. Other algorithm settings were set to default. The reported GSEA statistics, in addition to P values, were false discovery rate (FDR), family-wise error rate (FWER), and normalized enrichment score rank (NES rank; ref. 20).

Paraffin-embedded sections (5-μm thick) were deparaffinized in xylene and rehydrated through an ethanol series. Slides were placed in a pressure cooker with Tris-EDTA buffer (pH 9.0) and heated for 10 minutes at 97°C to unmask antigen epitopes. Slides were then incubated in distilled water for 5 minutes at room temperature. Endogenous peroxidase activity was quenched by treatment with 3% hydrogen peroxide in absolute methanol for 30 minutes. Slides were washed in PBS containing 0.025% Tween 20 and incubated in normal blocking serum (Vectastain Elite ABC kit, Vector Laboratories) for 60 minutes. After washing in PBS, slides were incubated with the primary antibodies overnight. Tissue sections were washed in PBS with 0.025% Tween 20, and immunostaining was performed using a species-specific biotinylated secondary antibody, avidin/biotinylated horseradish peroxidase complex, and substrate-chromogen mixture (Vectastain Elite ABC Kit). Immunoreactivity was visualized using Vector DAB (brown color) peroxidase substrate as a chromogen. After washing with distilled water, the sections were dehydrated in a graded series of ethanol followed by xylene and mounted with VectaMount permanent mounting media (Vector Laboratories). The antibodies used for labeling were FANCD2, p-AKT (Ser493), and p-P70S6K (Thr389; Cell Signaling Technology) at 1:300, 1:100, and 1:100 dilutions, respectively. Scoring the staining intensity was based on a two-tiered system: with 2+ signifying intense nuclear staining throughout the tumor specimen; while a score of 1+ indicated less-intense or patchy staining throughout the tumor specimen. Samples with no staining received a score of 0. The scores were assessed while blinded from all clinical data. When there were samples with multiple cores that were assigned different scores, the most common score was the assigned score for the patient case.

We previously reported the results of our pilot study, in which the t-ARMS-Rh30 demonstrated an increased resistance to radiotherapy alone, that the dose density of 59 to 60 Gy/cc resulted in 12-week treatment failure rates of 78% versus 25% with Rh30 and ERMS-Rh18, respectively (18). In a dose-escalation study (using 5, 10, and 20 mg of AZD8055 at radiation doses of 10 and 20 Gy), we observed that the addition of AZD8055 to XRT (after a one-week pretreatment with daily AZD8055 alone; see Supplementary Table S1 for treatment schedule) resulted in a selective sensitization of the Rh30 line. With a mean XRT dose density of 27 Gy/cm³, the 12-week failure rate of Rh30-bearing mice improved to four of 15 (27%). For the Rh18 group, this rate was eight of 15 (53%) at a mean dose density of 44 Gy/cm³. Thus, the radiosensitivity enhancement factor of AZD8055 toward Rh30 was 7.3, whereas with Rh18 this ratio was 0.83 when comparing with the previous results of radiotherapy alone (P < 0.001, Fisher exact; a ratio of 1.0 signifies no enhancement; Supplemental Table S2).

On the basis of the results of the dose-escalation study, groups of 10 mice were subjected to either no treatment, AZD8055 alone for 21 consecutive days (10 mg/kg via gavage), XRT alone (20 Gy in 10 fractions in 2 weeks), or combined treatment (AZD8055 10 mg/kg + 20 Gy in 10 fractions with concurrent treatment lasting for 2 weeks after one week of AZD8055 alone). The addition of AZD8055 to radiotherapy resulted in an improvement in event-free survival rate in the Rh30-bearing mice (median EFS of 67 days in XRT-alone vs. >91 days in the combined arm, P = 0.009, log-rank), but did not affect the outcome in Rh18. (Fig. 1A and B). Rh18 responded favorably to 20 Gy alone, with little change in outcome when the drug was added. Mice bearing the Rh30 line and receiving 20 Gy alone had a propensity to become cachectic and develop distant metastases, resulting in early removal from study and suggesting that these xenografts were under treated and remained viable to metastasize. The use of drug alone inhibited xenograft growth for about 2 to 3 weeks, but then it resumed as previously observed (11). There were no treatment-related deaths and the skin reaction severity of 20 and 30 Gy with and without AZD8055 had similar average maximum intensities and recovery times (Supplementary Fig. S1A and S1B).

In parallel with the previously mentioned preclinical murine screening, three to five flank xenografts (per time point) were harvested 30 to 60 minutes following daily 2-Gy fractions, at cumulative doses of 2, 4, 6, and 12 Gy. Xenografts that displayed recurrence in mice receiving cumulative doses 20 or 30 Gy in daily 2 Gy fractions were harvested as well. To account for tumor regional heterogeneity and necrosis, samples from mice receiving equivalent treatment were combined in equal protein proportions and assessed via Western blotting. Figure 2A and B reveals a more pronounced drug action in t-ARMS-Rh30. Without AZD8055, there is a brisk upregulation of FANCD2 at 2 and 4 Gy, followed by a gradual decline at subsequent time points. In the presence of the drug, the upregulation of FANCD2 after 2 Gy is significantly blunted and the protein is entirely absent in later time points. Closer inspection of the immunoblot analysis at a cumulative dose of ≥4 Gy reveals not only blockade of the downstream TORC1/TORC2 phosphorylated/activated species, but also loss of total Akt, P70S6K, and mTOR (Supplementary Fig. S2), suggesting a comprehensive shutdown of these prosurvival pathways. Within the ERMS-Rh18 line, the effect of AZD8055 is less pronounced. FANCD2 protein expression is neither as briskly upregulated in the early fractions of radiotherapy nor are these levels as strikingly abrogated in the presence of this drug. The protein and phospho-protein levels of TORC1/TORC2 signaling cascade remain more stable throughout the XRT treatment course with or without coadministration of AZD8055 as well.

To determine whether the FANCD2 function of the two xenograft lines was affected differently by AZD8055 in these treatment settings, we performed a FANCD2 functional foci staining assay on FFPE thin sections of the xenografts harvested at the parallel timepoints of our immunoblot analysis (Fig. 2C and D). Consistent with the preclinical screen outcomes and immunoblot analysis, the addition of AZD8055 seems to preferentially alter/diminish the function of FANCD2 in t-ARMS-Rh30. At all cumulative timepoints, the presence of AZD8055 significantly reduces the FANCD2 activity within this xenograft line. Furthermore, as observed at the protein expression level, the greatest difference in activity between the Rh30 xenografts treated with radiation alone and radiation plus drug occurs at the 4 Gy timepoint, where AZD8055 seems to lower the percentage of positively staining nuclei from 25% to 8%. In the ERMS-Rh18 line, there is a modest gain in activity of FANCD2 in the 2 and 4 Gy samples, but the impact of AZD8055 on this activity seems to have a random/variable effect from timepoint to timepoint, with a lower magnitude of differences between treatment regimens.

To verify the importance of FANCD2 in the radiation responsiveness of these two xenograft lines and to rule out any potential “off-target” effects of AZD8055, we performed clonogenic survival assays of Rh18 and Rh30 cells in vitro, with and without direct, transient siRNA inhibition of FANCD2 expression (Supplementary Fig. S3). Compared with untreated and nontargeted controls, the knockdown of FANCD2 in both Rh18 and Rh30 resulted in increased sensitivity to radiation. Rh30 demonstrated a broader shoulder than Rh18 and AZD8055 had a greater impact at the higher dose-per-fraction ranges.

The patient characteristics and prognostic factors of this set are summarized in Supplementary Table S3 (Gene Expression Set). These data reflected an adequate representative RMS population. Furthermore, based on the clinical stage and grouping as well as histology, it can be estimated that 95 out of the 101 patients would have received radiotherapy as all or part of their local therapy component. The mRNA levels were analyzed as a continuous variable as well as a quartile rank. Interestingly, in a univariate analysis, the mRNA expression levels were prognostic with the upper quartile having the worst overall survival time at a mean of 5.1 years, whereas the lowest quartile had a mean survival of 8.7 years and the mid-quartiles yielded two subsets, ranging between 5.5 years and 6.5 years (Fig. 3A, P = 0.043, log-rank). Quartile FANCD2 expression directly correlated with stage IV disease (CC: 0.170, P = 0.046, Spearman), and had a trend in association with the PAX3–FOXO1 fusion gene (CC: 0.122, P = 0.111, Spearman). Because of the observed differences in FANCD2 expression and activity after radiotherapy in t-ARMS-Rh30 and ERMS-Rh18, we sought to determine whether FANCD2 had a differential prognostic impact on these respective clinical subtypes. When stratified by histology, the top quartile rank of FANCD2 transcript expression was found to be adversely prognostic in the ARMS group, but not the ERMS group (Fig. 3B). When stratified by the PAX3–FOXO1 fusion gene status, the top quartile rank of FANCD2 transcript expression was found to be adversely prognostic in the fusion-positive group, but not the fusion-negative group (Fig. 3C). Cases with both high FANCD2 and the PAX3–FOXO1 fusion gene had the worst outcome (mean 2.9 years) compared with the dually negative subgroup which had the most favorable outcome (mean 9.9 years; P = 0.045, pairwise log-rank). The group with one positive marker had an intermediate outcome (mean 6.6 years). To test whether dual presence of PAX3–FOXO1 fusion gene and high FANCD2 mRNA expression was an independent prognostic variable, a Cox multivariate analysis was performed on this set (Table 1). In addition to stage IV (metastatic disease) and presence of the PAX3–FOXO1 fusion gene, the coexpression of this oncogene and high FANCD2 was independently and adversely prognostic [HR: 3.2; 95% confidence interval (CI), 1.5–7.5, P = 0.033], after accounting for histology, nodal stage, primary tumor site, tumor size, FANCD2 expression, and other fusion statuses.

We hypothesized that mTOR interacting genes are more likely to be associated with FANCD2 mRNA in the setting of PAX3–FOXO1 fusion gene versus those RMS without this translocation, based on the selective efficacy of AZD8055 in Rh30 and observations of FANCD2 activity and expression after several 2 Gy daily fractions of radiotherapy. The Ingenuity Pathway Analysis Interactive Pathway Builder (Ingenuity Systems; www.ingenuity.com) was used to discover a panel of genes with experimentally observed direct interaction, either up- or downstream of mTOR, and is summarized in Supplementary Table S4. These mTOR-associated genes were then analyzed using a supervised k-means clustering with respect to FANCD2 mRNA within both the PAX3–FOXO1 fusion-positive and -negative subsets using Matlab Statistics toolbox. Three distinct mRNA populations arose from clustering within both groups characterized primarily by high, low, and intermediate FANCD2 transcript levels. Fusion-positive samples with “high” FANCD2 (n = 18) had 4.3-fold higher FANCD2 mRNA levels than those with “low” status (n = 3); similarly, RMS without the PAX3–FOXO1 fusion gene demonstrated a 4.8 fold-change between “high” (n = 18) and “low” (n = 7) groups (the results of the k-means clustering are visualized in Fig. 4A and B). The groups with intermediate levels were not analyzed. To test our mTOR-FANCD2 hypothesis, we applied gene set enrichment analysis (23) by using a convenient desktop application maintained by the Broad Institute (Cambridge, MA; ref. 24). We found by sample permutation that PAX3–FOXO1 fusion-positive RMS with “low” FANCD2 mRNA demonstrated direct enrichment of mTOR-associated genes, whereas in samples with “high” FANCD2 mRNA, there was an inverse enrichment of mTOR-associated genes. Although weakly significant (P = 0.07, FDR = 0.68, FWER = 1, NES rank = 410), this clearly displayed a nonrandom enrichment and the plot can be visualized in Fig. 4C. Interestingly, those RMS with no recorded translocation demonstrated a random rank distribution of mTOR interacting genes (Fig. 4D) with respect to the FANCD2 mRNA signal-to-noise profile and was not statistically significant (P = 0.54, FDR = 0.61, FWER = 1, NES rank = 3,760). Although the data are correlative, those demonstrate that there is a potential mechanistic relationship between mTOR and FANCD2 that may be specific to the molecular framework that the PAX3–FOXO1 fusion gene creates.

There were 121 cases in which at least one core was adequate for immunohistochemical staining of FANCD2, p-P70S6K (Thr 389), and p-Akt (Ser 473). Most cases had two to three individual cores within the array. There were 71 (59%) ARMS cases, 47 ERMS (39%) cases, and one case each of mixed histology, undifferentiated, and RMS not otherwise specified. Of the 121, only 94 cases had complete survival data. For complete demographics and clinical prognostic factors in these 94 patients, please refer to Supplementary Table S3 (TMA Set). All but two cases would have received radiotherapy as part of the multimodality treatment course. There were no cases in which the FANCD2 score was 0, but there were three distinct staining patterns observed in both p-P70S6K and p-Akt. A Spearman rank screen of molecular and clinical markers across all cases demonstrated that FANCD2 was significantly correlated with p-P70S6K and the presence of the PAX3–FOXO1 fusion gene (Table 2; all cases), but no other clinical or molecular factors. Although there was a trend in the FANCD2/p-Akt correlation, pAkt and p-P70S6K were strongly correlated, suggesting that concordant TORC1/TORC2 activation frequently occurred in these samples. Within the cases harboring the PAX3–FOXO1 fusion gene (n = 30), there was a strong correlation between FANCD2 and p-P70S6K (correlation coefficient: 0.60; P < 0.001; Table 2; fusion-positive cases). However, FANCD2 is not significantly correlated with p-Akt expression, consistent with our previous observation that FANCD2 is regulated primarily via TORC1 signaling in the t-ARMS-Rh30 cell line (12). In the cases which do not harbor the fusion gene (n = 65), FANCD2 was neither correlated with p-P70S6K nor p-Akt (Table 2; fusion-negative cases). The scoring breakdown, stratified by fusion gene status, is presented in Supplementary Table S5.

Assessment of FANCD2's survival associations in the PAX3–FOXO1 fusion gene-positive cases proved to be difficult in this patient set. There were only six of 22 cases of fusion-positive cases with complete survival data that had low (1+) FANCD2 scores. Furthermore, with 64% of the cases censored, within this set, identification of factors which can have subtle impacts on survival is difficult to identify as well.

Our preclinical and clinical data suggest that FANCD2 has a significant role in resistance to radiotherapy and potentially overall survival in t-ARMS. Furthermore, our data demonstrate that FANCD2 expression and activity may be modulated by mTOR signaling (specifically by TORC1 signaling) within this disease type. Thus, mTOR inhibition during radiotherapy may represent an effective radiosensitization strategy in the clinic setting. This finding is significant in that it is the first time a specific potential mediator of treatment resistance has been identified in the t-ARMS subtype. Improving the efficacy of radiotherapy in t-ARMS, which has a high rate of local failure and mortality, could have a profound impact in pediatric RMS, as this is the most common and lethal form of this disease. The significance of FANCD2 in ERMS and those RMS without the PAX3 fusion gene is less clear. Although the knockdown of FANCD2 in vitro resulted in a sensitization of the ERMS-Rh18, this line was already significantly more sensitive to radiotherapy than the t-ARMS-Rh30 (as is seen clinically), suggesting that greater vulnerability to ionizing radiation that is present within this xenograft line and supersedes the impact of FANCD2.

On the basis of the expression and activity dynamics of FANCD2 within our preclinical screen, it seems that this is a radiation response protein, especially in the early phases of fractionated treatment. It was surprising, then, that the presence of elevated expression (at the transcript level) in patient specimens was prognostic within the t-ARMS subgroup. It is worthwhile to note that the majority of patients with RMS undergo induction chemotherapy before their local therapy. Thus, at the time of tissue collection, there are already active compensatory mechanisms to address DNA damage brought upon by systemic agents with cross-linking induction capabilities (namely, cyclophosphamide). It is well established that DNA damage and replicative stress induce the expression and activity of FANCD2 (25). It may very well be that determining FANCD2 expression status at the time of surgery provides insight into the tumor's ability to resist all DNA-directed cytotoxic therapy in the treatment course of RMS (including radiation) and thus may explain this impact on overall survival. It is further tempting to speculate that if this resistance mechanism in t-ARMS is significantly driven by mTOR activity and is unique to these tumor cells, there could be a rationale for mTOR kinase inhibition before or during XRT in patients with these tumors and subsequently improving the therapeutic ratio. Because of the consistency of the preclinical testing and correlative clinical findings and the dichotomous nature of the PAX3–FOXO1 oncogene, which segregates RMS into two distinct disease subtypes, clinically and molecularly, we feel these data may be applicable to the RMS population as a whole, despite only testing two xenograft lines. These suppositions, of course, warrant further investigation.

Finally, the results of these experiments demonstrate the utility and feasibility of our approach in testing novel agents and other potential radiosensitizers in vivo while identifying a possible mechanism of resistance which has clinical relevance. This represents the first time that a systemic agent has been given with radiotherapy within the Pediatric Preclinical Testing Program model, as well. We are able to use large numbers of mice, to generate sufficient samples for statistically powered outcomes studies and numerous biospecimens for molecular studies. It is the anticipation of our group to continue these preclinical screens, to find efficacious sensitizers, and elucidate the molecular mechanisms behind pediatric tumor radiobiology in parallel. This approach is especially important in childhood cancer research, given the relative rarity of the many solid tumors which arise in this patient population.

No potential conflicts of interest were disclosed.

Conception and design: M. Singh, R. Kaplon, P.J. Houghton, A. Chakravarti

Development of methodology: J.M. Leasure, C. Chronowski, K. Bondra, W. Duan, N. Li, C. Shen, R. Kaplon, S. Hammond

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): J.M. Leasure, C. Chronowski, K. Bondra, L.A. Hensley, M. Villalona-Calero, A.M. Vergis, C. Shen, N. Sebastian, R. Kaplon, S. Hammond, P.J. Houghton

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): M. Singh, J.M. Leasure, C. Chronowski, B. Geier, W. Duan, L.A. Hensley, M. Villalona-Calero, A.M. Vergis, R.T. Kurmasheva, D. Fabian, P.J. Houghton, K. Palanichamy, A. Chakravarti

Writing, review, and/or revision of the manuscript: M. Singh, J.M. Leasure, C. Chronowski, W. Duan, M. Villalona-Calero, A.M. Vergis, R.T. Kurmasheva, G. Woods, D. Fabian, R. Kaplon, S. Hammond, P.J. Houghton, K. Palanichamy, A. Chakravarti

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): J.M. Leasure, C. Chronowski, K. Bondra, R. Kaplon, P.J. Houghton, A. Chakravarti

Study supervision: M. Singh, K. Bondra, P.J. Houghton, K. Palanichamy, A. Chakravarti

The authors thank Christopher Pelloski for his contributions to this work, Irene Snyder for her assistance in preparing this manuscript for publication, Natalie Beeler for her assistance in obtaining the COG D9902 tissue microarray and corresponding data, and AstraZeneca for supplying the AZD8055 for the experiments.

This work was supported through: 1P50CA127001 (C.E. Pelloski); NO1-CM-42216 (to P.J. Houghton), CA77776 (to P.J. Houghton), RC2CA148190 (to A. Chakravarti), R01CA108633 (to A. Chakravarti), from the National Cancer Institute; a Hope On Wheels Programmatic Grant from the Hyundai Corporation of North America; and a grant from the Veterans of Foreign Wars of Ohio, Sarcoma Seed Funding Program, and The Ohio State University Comprehensive Cancer Center (to A. Chakravarti).

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.