Prospective Analysis of Parametric Response Map–Derived MRI Biomarkers: Identification of Early and Distinct Glioma Response Patterns Not Predicted by Standard Radiographic Assessment

The standard of care for patients with high-grade gliomas is concomitant and adjuvant temozolomide and radiotherapy based on recent clinical results showing improved median and 5-year survival for patients treated with this chemoradiation regimen (1). Current clinical assessment of treatment response is determined 10 weeks posttreatment initiation on the basis of Macdonald criteria (2, 3). Limitations in accurate assessment of response has led to the development of a new standardized response criteria that accounts for transient changes in tumor volume posttherapy by the Response Assessment in Neuro-Oncology Working Group (RANO; ref. 4). There exists a critical need for the development of individualized approaches to the treatment of this disease, which has a 1-year survival rate of roughly 50%. In this regard, MGMT (O⁶-methylguanine methyltransferase) promoter methylation status is a potential predictive biomarker for selection of brain tumor patients who may derive the most benefit from treatment with temozolomide and radiotherapy (5). This initial success provides motivation for the development of noninvasive imaging biomarkers of response capable of predicting sensitivity to novel therapeutic interventions, as not all patients are able to undergo surgical resection and provide sufficient tissue for MGMT analysis. The potential for using quantitative imaging modalities as early surrogate markers of response is compelling due to the ability to noninvasively detect treatment-associated functional and metabolic changes (6).

In this study, we investigated a multimodality, image-based response biomarker for high-grade gliomas to more accurately predict treatment outcome early. Identification and validation of an image biomarker could provide the opportunity to select patients that may benefit from alternative treatment strategies. Quantitative MRI can provide structural and functional information about tumor cellularity and perfusion characteristics that compliment anatomic MRI. The value of this approach is that the microenvironment and physiology of the tumor can be monitored noninvasively, allowing for early assessment of tumor therapy response prior to treatment completion. Two techniques investigated for their diagnostic and prognostic value in brain tumor patients are diffusion weighted (DW)-MRI and dynamic susceptibility contrast (DSC)-MRI (7, 8). DW-MRI, capable of measuring the random thermal (Brownian) motion of water, samples the tumor microenvironment on a subcellular level. A drop in tumor cellularity as a result of significant cell kill following effective therapy has been associated with an increase in water diffusivity as expressed as the apparent diffusion coefficient (ADC; ref. 9). This trend has been observed in preclinical and clinical studies supporting the notion of DW-MRI as a surrogate imaging biomarker for treatment response assessment in oncology (10–12). DSC-MRI, capable of quantifying changes in tumor hemodynamics, provides information on neovascularity and angiogenesis (13–17). At present, the relative cerebral blood volume (rCBV), as measured by DSC-MRI, has shown promise in determining tumor grade as well as prediction and early assessment of patient clinical response (18–21).

Imaging biomarkers (i.e., ADC or rCBV) of tumor response have historically been evaluated using the mean change in the whole tumor volume as the quantitative metric for detection of therapeutic efficacy. However, tumor heterogeneity is an inherent property of glioblastoma multiforme (GBM) and therefore standard methods of analysis using average values throughout an entire tumor can significantly underestimate regional changes following therapy. In 2005, Moffat and colleagues proposed a new approach for evaluating tumor response using serial ADC maps termed the “functional diffusion map” (fDM; ref. 22). For the first time, changes in ADC were evaluated on a voxel-wise basis by spatially aligning (via image registration) serial ADC maps acquired pre- and midtherapy. The tumor volume fraction that exhibited a significant increase in midtreatment ADC from baseline values was found to be predictive of 10-week radiologic response in a cohort of 20 glioma patients as early as 3 weeks into therapy. This technique has now been established in multiple clinical sites including breast cancer, prostate cancer, metastatic cancer to the bone, and head and neck cancer (23–29). For the most part, this technique has only been applied to ADC maps. In 2009, the voxel-based approach was extended to rCBV maps acquired from DSC-MRI of 44 glioma patients. Now referred to more generally as the “parametric response map” (PRM), this postprocessing approach when applied to rCBV (PRM_rCBV) was predictive of survival as early as 1 week posttreatment initiation as well as distinguishing progression from pseudoprogression (30, 31).

We report here the outcome of a prospective single-center trial of WHO grade 3 and 4 glioma patients that was designed to evaluate the predictive accuracy of PRM_ADC and PRM_rCBV imaging biomarkers as single and composite metrics of early treatment response as assessed 3 week posttreatment initiation. We compared parametric models using PRM imaging metrics from all patients (n = 45) to assess correlation with 1-year and overall survival.

Patients newly diagnosed with histologic confirmed grade 3 and 4 gliomas were enrolled on a protocol of intratreatment MRI. This cohort of patients has been used previously in published work investigating the prognostic value of PRM applied to the individual parameters ADC and rCBV (30, 32). Surgical resection was conducted on all patients and the extent was evaluated by postoperative MRI obtained within 72 hours of resection. Radiotherapy was delivered using three dimensional conformal therapy or intensity-modulated radiation therapy (IMRT). Standard techniques were utilized with a 2.0 to 2.5 cm margin on either the enhancing region on gadolinium-enhanced scans or the abnormal signal on T2-weighted scans to 46 to 50 Gy with the central gross tumor treated to a final median dose of 70 Gy in 6 to 7 weeks (33). Twenty-three of these individuals were maintained on a phase 2 protocol of high-dose (>60 Gy) radiation therapy concurrent with temozolomide 75 mg/m² daily, followed by adjuvant temozolomide 150 to 200 mg/m² daily for 5 days of a 28 day cycle. Remaining patients received radiation alone (n = 22). Chemotherapy was commonly administered in both groups depending upon clinical circumstances (Table 1). None of these patients, accrued from 2000 to 2006, received any antiangiogenic therapy during their primary treatment, although some patients did receive Avastin (bevacizumab; Genentech) at time of progression. Informed consent was obtained, and the use of images and medical record was approved by the University of Michigan Institutional Review Board.

Patients were evaluated pretherapy, 3, and 10 weeks after the start of radiation with follow-up scans every 2 to 3 months. DW- and DSC-MRI as well as standard MRI (fluid attenuation inversion recovery, T2-weighted, and gadolinium-enhanced T1-weighted MRI) were conducted on either a 1.5 T Signa (General Electric Medical Systems; n = 31 subjects) or 3 T Achieva (Philips Medical Systems; n = 14 subjects) system. Radiologic response at 10 weeks was based on changes in tumor volume on T1-weighted contrast-enhanced MRI and steroid doses and was classified as complete response (CR), partial response (PR), stable disease (SD), and progressive disease (PD; ref. 2). Steroid doses were recorded before each scan, weekly during radiotherapy, and at each follow-up (34).

DW-MRI utilized a single-shot, spin-echo, diffusion-sensitized, echo-planar imaging (EPI) acquisition sequence. On the 1.5 T system, twenty-four 6-mm axial-oblique sections were acquired using a 22-cm field of view (FOV) and 128 matrix with “b-factor” = 0 and 1,000 s/mm² along 3 orthogonal directions [repetition time (TR) = 1,000 ms; echo time (TE) = 71–100 ms, and number of averages (NS) = 1]. On the 3 T system, at least twenty-eight 4-mm axial-oblique sections were acquired through the brain using a 24-cm FOV and 128 matrix (TR = 2,636 ms; TE = 46 ms; NS = 1 for b = 0, and NS = 2 for b = 1,000 s/mm²). Parallel imaging (sensitivity encoding factor = 3) was used on 3 T to reduce spatial distortion. The diffusion-weighted images for the 3 orthogonal directions were used to calculate an ADC map (32).

For DSC-MRI, 14 to 20 slices were acquired by a gradient-echo EPI pulse sequence (TR = 1.5–2 s, TE = 50–60 ms, FOV = 220 × 220 mm², matrix = 128 × 128, flip angle = 60 degrees, and 4–6 mm thickness and 0 mm gap). Gadolinium-diethylenetriamine pentaacetic acid (Gd-DTPA; Bayer HealthCare Pharmaceuticals) was injected intravenously with a dose of 0.05 to 0.1 mL/kg as a bolus using a power injector at a rate of 2 mL/s, followed immediately by 15 cc of saline flush at the same rate. Subsequently, a Gd-enhanced T1-weighted imaging was acquired. CBV map was generated from DSC images as described previously (35). To assess differences in tumor blood volume during chemoradiation therapy (cRT) and between subjects, CBV maps were normalized to values within white matter regions contralateral to the tumor to generate rCBV maps. For simplicity in notation, relative blood volumes for both brain and tumor have been denoted by “rCBV” throughout this article. The following guidelines were used to define the white matter regions of interest for normalization: (i) contralateral to tumor, (ii) received < 30 Gy accumulated dose, and (iii) as large as possible but avoiding regions with susceptibility artifacts and partial volume averaging.

Pre- and 3-week midtreatment ADC and rCBV maps were coregistered to Gd-enhanced T1-weighted images acquired before cRT using mutual information as an objective function and simplex as an optimizer (36). The coregistration of different and similar weighted MRI scans for the same subject is totally automatic and assumes a rigid body geometry relationship (i.e., rotation and translation) between head scans. Following coregistration, brain tumors, manually contoured by a neuroradiologist, were defined within the enhancing regions of the tumor on the Gd-enhanced T1-weighted images. Only patients with contrast-enhancing tumors of 4 cm³ or greater were included in this study. For simplicity, tumor will refer to only the contrast-enhancing volume on a T1-weighted postcontrast image. Shrinkage or growth of the tumor during the time between scans may have occurred; therefore, only voxels that are contrast enhancing on both the pre-RT and mid-RT tumor volumes were included. Subsequent to coregistration, the resultant maps consist of spatially aligned voxels with ADC and rCBV values before and 3 weeks following treatment initiation.

The PRM of any parameter (PRM_X, where X denotes any parametric map such as ADC and rCBV) was determined by first calculating the difference between X (ΔX = mid X − baseline X) for each voxel within the tumor pretherapy and at week 3 posttreatment initiation. Voxels yielding ΔX greater than a predetermined threshold (set to 1.2 and 55 for rCBV and ADC, respectively; details described previously; refs. 22, 30, 32) were designated red (i.e., ΔrCBV > 1.2; ΔADC > 55). Blue voxels represent volumes whose parameter value decreased by more than the threshold (i.e., ΔrCBV < −1.2; ΔADC < −55) and the green voxels represent voxels within the tumor that were unchanged (i.e., |ΔrCBV| ≤ 1.2; |ΔADC| ≤ 5.5). The volume fractions within the tumor determined from PRM_X were PRM_X+ for increasing X, PRM_X− for decreasing X, and PRM_X0 for unchanged X.

The patient population was stratified on the basis of previously reported cutoff values of 4.7% and 6.8% for PRM_ADC and PRM_rCBV, respectively (30, 32). The composite model of these 2 dichotomized metrics was determined by categorizing patients as responders, nonresponders, or intermediate. Patients were designated nonresponding when both single metrics predict nonresponding (PRM_ADC+ < 4.7% and PRM_rCBV− > 6.8%), whereas patients were designated responders when both single metrics predict responder (PRM_ADC+ ≥ 4.7% and PRM_rCBV− ≤ 6.8%). For the cases wherein single metrics were in disagreement, patients were designated intermediate (PRM_ADC+ ≥ 4.7% and PRM_rCBV− > 6.8% or PRM_ADC+ < 4.7% and PRM_rCBV− ≤ 6.8%). A univariate and multivariate logistic regression with forward entry was used to assess the predictive value of each metric for 6 month progression-free survival (PFS) and 1-year survival. Assessment of overall survival for the metrics was determined using a Kaplan–Meier analysis with log-rank test and a multivariate Cox regression with forward entry. Secondary analysis was conducted to assess the predictive value of patient characteristics to 1-year survival using either an unpaired Student's t test or likelihood ratio test. All statistical computations were conducted with a statistical software package (SPSS Software Products), and results were declared statistically significant at the 2-sided 5% comparison-wise significance level (P < 0.05).

A total of 45 subjects with high-grade glioma were included in this prospective study (Table 1). The median survival for the population, as determined by Kaplan–Meier analysis was 14.9 months (95% CI: 8.6–21.2 months) with 56% of the patient population realizing 1-year survival. Patients expiring prior to 1 year had a median survival of 6.6 months (95% CI: 4.4–8.8 months), whereas those whose survival was beyond 1 year had a median survival of 35.1 months (95% CI: 13.7–56.5 months). Patient age, Radiation Therapy Oncology Group Recursive Partitioning Analysis (RTOG RPA; ref. 37), and radiologic response at 10 weeks were found to be significant predictors of 1-year survival (P = 0.003, P = 0.025, and P = 0.001, respectively; Table 1).

Individually, PRM_ADC and PRM_rCBV have been reported to be predictive biomarkers of clinical outcome (30, 32). The underlying mechanisms driving therapeutic-induced alterations in ADC and rCBV differ and therefore the predictive value of the outcome measures of PRM (PRM_X+ and PRM_X−) vary based on the quantitative parameter used for analysis. A large volume fraction of the tumor with significantly increasing ADC (PRM_ADC+) correlated with overall survival, unlike PRM_ADC− which had no predictive value. In stark contrast, a small volume fraction of tumor with decreasing rCBV (PRM_rCBV−) correlated with a positive outcome, where no predictive value was observed for PRM_rCBV+. Color overlays of PRM_ADC and PRM_rCBV at week 3 posttreatment initiation of a representative responder (overall survival = 17 months; PRM_ADC+ = 12%; PRM_rCBV− = 4%) and nonresponder (overall survival = 2 months; PRM_ADC+ = 1%; PRM_rCBV− = 40%) show spatial variation in the response of these tumors with respect to water diffusivity and blood volume (Fig. 1). Both lesions had high vascular irregularities as can be presumed from the immense contrast-enhancement on the T1-weighted images (Fig. 1A and D). The complex changes in tumor cellularity and vasculature following a therapeutic intervention appear from the PRM overlays to be somewhat dissociated from each other. For example, the responding tumor clearly showed regions of increased ADC, denoted in red (Fig. 1B), which is associated with a loss in tumor cellularity. These same regions had statistically no change in rCBV (green area in Fig. 1E). In contrast, a significant volume of the nonresponding tumor showed no change in ADC, Figure 1C, suggesting no real loss of tumor cellularity. Unlike ADC, Figure 1F shows that a substantial fraction of tumor underwent a significant drop in rCBV posttreatment initiation, perhaps suggesting increased hypoxia in the tumor conferring radioresistance. Therefore, the combination of diffusion and perfusion imaging may provide complementary information that serves to improve sensitivity and specificity in predicting patient outcome. To evaluate a combined metric related to these physiologic changes to treatment, we analyzed these parameters as a unified predictive metric of survival.

We first assessed the sensitivity of the individual parameter models (PRM_ADC+ and PRM_rCBV−) and composite model (PRM_ADC+/PRM_rCBV−), as well as radiologic response for comparison, to predict 6-month PFS and 1-year survival in our cohort of patients using a logistic regression. Univariate analysis found a strong correlation of parametric models with both 6-month PFS and 1-year survival (Table 2). Inclusion of all four imaging metrics (PRM_ADC+, PRM_rCBV−, PRM_ADC+/PRM_rCBV−, and radiologic response) into a single multivariate logistic regression showed that PRM_ADC+/PRM_rCBV− and radiologic response provided the strongest statistical power for assessing 6-month PFS (P = 0.012 and P = 0.005, respectively) but only PRM_ADC+/PRM_rCBV− for assessing 1-year survival (P < 0.0001). To determine the statistical power of age, dichotomized on the basis of responders of 50 years or above and nonresponders of less than 50 years, and RPA RTOG against the imaging models, a multivariate analysis was conducted with age, RTOG RPA, PRM_ADC+, and PRM_rCBV− for assessing 1-year survival. Both age and RTOG RPA classification were not found to contribute to the statistical model.

Stratification of the patient population by each of the 4 models used, which includes radiologic response, are presented as Kaplan–Meier plots in Figure 2. The median survivals for nonresponders and responders as determined by PRM_ADC+ were 8.3 months (95% CI: 6.2–10.4; n = 25) and not reached (n = 20), respectively. Using PRM_rCBV− to categorize nonresponders and responders, their median survival was found to be 7.1 months (95% CI: 5.1–9.1; n = 22) and 20.4 months (95% CI: 1.7–39.1; n = 23), respectively. Median survivals for nonresponders and responders as determined by radiologic response were 8.1 months (95% CI: 6.0–10.2; n = 19) and not reached (n = 26), respectively. For the composite PRM model with the 3 discrete response groups, median survivals for nonresponders, intermediate, and responders were 5.5 months (95% CI: 4.4–6.6 months; n = 14), 16 months (95% CI: 8.6–23.4 months; n = 19), and not reached (n = 12), respectively. Using a log-rank test to assess significance, all 4 models were highly predictive of overall survival (Table 2). Conducting this same statistical test pairwise between response groups defined by the composite model found that nonresponders had a significantly shorter survival than the intermediate and responder subpopulations (P < 0.0001). Although not found to be significant based on our statistical criteria (P < 0.05), the survival of responders when compared with intermediate generated a P value of 0.06. This is partly attributed to the small numbers analyzed in this study. Assessment of overall survival of all models in a unified multivariate test found PRM_ADC+/PRM_rCBV− assessed at week 3 midtreatment to have the strongest statistical power even over radiologic response.

In the cohort of patients studied, 8 were diagnosed with grade 3 gliomas. Additional analysis was conducted to determine whether inclusion of these patients adversely affected the results presented above. As observed in Table 1, 7 of these patients had survived longer than 1 year with only a single patient surviving less than a year (overall survival = 5.1 months). This lone patient had a tumor that exhibited PRM values of ADC and rCBV indicative of a tumor nonresponsive to therapy. On the basis of these results, the composite PRM categorized this patient as a nonresponder. In fact, of the 8 patients with grade 3 glioma, the composite model of PRM only identified the referenced patient as a nonresponder, 3 were intermediate [mean overall survival (SD) = 22 (6.7) months] and 4 were responders [mean overall survival (SD) was 35 (19) months]. Finally, assessment of the composite PRM model when limited to grade 4 gliomas found the composite PRM to be significant for overall survival (P < 0.0001) with median survivals of 6 months (n = 13) for nonresponders, 12.8 months (n = 16) for intermediate response, and median survival not reached at greater than 30 months in the 8 patients scored as responsive. Multivariate analyses, analogous to those conducted in Table 2, of patients with grade 4 gliomas (n = 37) identified a strong correlation of PRM_ADC+/PRM_rCBV− and radiologic response to 6 month PFS (P = 0.016 and P = 0.02, respectively) and only PRM_ADC+/PRM_rCBV− to both 1-year (P = 0.002) and overall survival (P < 0.0001).

Personalized medicine in cancer care is slowly becoming a reality from the “one-size fits all” approach. Advances in molecular and histologic biomarkers are now providing detailed information on tumor biology that is being used to individualize therapy (38). Pretreatment assessment using molecular and histologic biomarkers is increasingly being used to dictate initial choice of therapy. There is also a critical need for the deployment of surrogate biomarkers of response to provide real-time information on the state of the tumor midtreatment. With respect to assessment of treatment in GBM patients, modern advancements in neuroimaging techniques, such as MRI, positron emission tomography (PET), single photon emission computed tomography (SPECT), and computed tomography (CT), are providing detailed functional and quantitative information on tumor heterogeneity. More specifically, techniques utilizing MRI, with its high tissue contrast and resolution, are capable of imaging with high sensitivity changes in tumor microenvironment and hemodynamics at the cellular level allowing for early serial assessment of a tumor during treatment.

Standard management of malignant high-grade gliomas continues to be concurrent temozolomide and radiation of 60 Gy followed by adjuvant temozolomide. Currently, randomized phase III studies are being conducted to determine the additional benefit of novel targeted therapies, including various signal transduction inhibitors encompassing monoclonal antibodies (e.g., Avastin), as well as small molecules. However, these therapies are unlikely to benefit all patients and are extremely costly and associated with a significant increase in the risk of toxicity. Reliable predictive imaging biomarkers of response are important in selecting those patients most likely to benefit from alternative treatment strategies and selecting those patients as early as possible.

We and others have found quantitative neuroimaging methods are highly sensitive to analyzing treatment-induced cellularity and hemodynamic alterations within the tumor. Difficulties in analyzing these parametric maps ensue from the high spatial heterogeneity typical of gliomas. By focusing on regions of increasing or decreasing parameter value, the functional behavior of the tumor can be spatially dissected and interpreted for its prognostic value. Our group has shown the improved prognostic capabilities of quantitative imaging techniques (DW- and DSC-MRI) when analyzed by PRM, which has been compared with standard histogram-based methods and radiologic response (30, 32). Expanding further, we have shown in the present study that a composite of ADC and rCBV, as analyzed by PRM, enhances the sensitivity of the biomarker for predicting those patients resistant to chemoradiation at 3 weeks posttreatment initiation.

It is important to note that administration of steroid would have no confounding effect on the results reported in this study. As indicated in the article, steroid doses were closely followed prior to treatment, weekly during radiation therapy, and at the time of each MRI (pretreatment and 3 weeks during treatment). Given the close timing between the pretreatment scan (within 5 days prior to the start of treatment) and the 3 week scan (on average < 21 days from the start of treatment), there were minimal to no changes in steroid dosing over this short time period. It has been our standard clinical practice not to taper steroids during the first several weeks of radiation therapy due to concerns for possible increased edema during the early course of therapy. Therefore, steroid doses were typically not adjusted until weeks 3 to 6 of radiation therapy. Furthermore, the benefit seen by composite PRM was not simply a measure of generalized steroid effect, as previous results have shown that the percentage change in either ADC and rCBV would also have been impacted by steroid dosing and were not prognostic (30, 32).

Varying MRI field strengths as well as Gd-DTPA doses used in DSC-MRI for calculating CBV were also not suspected of affecting the current results. It is important to note that ADC is independent of field strength. Nevertheless, from the total population studied (n = 45), 31 were imaged at 1.5 T and the remaining 14 were imaged at 3 T. To avoid variability between scanners, the same scanner was used for a given patient throughout the study. An additional analysis was conducted to determine if there were any systematic differences in mean ADC and rCBV values pretherapy and at week 3 midtreatment as well as differences in PRM_ADC+ and PRM_rCBV− as a result of field strength using an unpaired Student's t test. No significant differences in imaging parameters were observed as a result of field strength. As for varying dosing of contrast agent, our algorithm for calculating CBV mitigates the effects of variable Gd-DTPA dose by use of the arterial input function measured on each patient individually, which implicitly scales with the same dose factor. All our CBV analyses use values normalized by white matter CBV which further reduces dose dependency. To mitigate leakage bias as a result of T1 effects, we employed a preinjection of contrast agent prior to a second bolus given during dynamic T2* imaging. In addition, we used a relatively long TR of 1.5 to 2 seconds to further reduce T1 weighting.

This is the first study that describes a composite model of PRM_ADC+ and PRM_rCBV− obtained at week 3 in assessing tumor response. Our data show that the composite model is a much stronger predictive factor of overall survival compared with either metric alone. This composite metric provides a novel imaging biomarker that has significant implications for early stratification in clinical trials in high-grade glioma as well as potentially other clinical sites. When included in a single multivariate statistical analysis, the composite of PRM_ADC+ and PRM_rCBV− obtained at week 3 was found to be a much stronger predictor of overall survival than either metric alone, as well as 10 week radiologic response. Use of the composite imaging biomarker not only provides a more early and sensitive method for identifying populations of responders and nonresponders but also allowed for the identification of a third patient subgroup that exhibited an intermediate response pattern not previously shown by conventional response criteria. Ongoing studies are elucidating the molecular determinants of response that differentiate tumors from each of these 3 response groups. Such information could ultimately facilitate the design of improved therapies for GBM patients identified in this manner.

A biomarker that allows for early identification of patients resistant to standard of care also provides an opportunity to reassess treatment options in a timelier manner compared with current clinical practice (39–41). For example, determination of responsiveness at week 3 into temozolomide with concomitant radiotherapy could allow patients resistant to standard of care to be given adjuvant treatment or be moved more quickly to an aggressive salvage regimen. To obtain further support for this approach, we recently initiated a clinical trial utilizing the composite imaging biomarker at week 3 following treatment initiation to select nonresponding patients for further treatment intensification with bevacizumab and dose painting IMRT. Although a multicenter trial to validate these findings is warranted, the results presented herein currently support a future paradigm shift in the current clinical management of GBM to a more individualized approach. In GBM, this would also provide an opportunity to develop new trials designed to improve patient outcome by selecting patients likely to benefit from more aggressive or experimental treatment regimens.

B.D. Ross has ownership interest (including patents) from ImBio, LLC. B.D. Ross, A. Rehemtulla, and T. L. Chenevert have a financial interest in the underlying DW-MRI technology. The other authors disclosed no potential conflicts of interest.

This work was supported by the U.S. NIH grants P01CA85878, P50CA93990, and P01CA087634.

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.