Survival prediction in patients undergoing radionuclide therapy based on intratumoral somatostatin-receptor heterogeneity

The NETTER-1 trial demonstrated significantly improved progression-free survival (PFS) for peptide receptor radionuclide therapy (PRRT) in neuroendocrine tumors (NET) emphasizing the high demand for response prediction in appropriate candidates. In this multicenter study, we aimed to elucidate the prognostic value of tumor heterogeneity as assessed by somatostatin receptor (SSTR)-PET/CT. 141 patients with SSTR-expressing tumors were analyzed obtaining SSTR-PET/CT before PRRT (1-6 cycles, 177Lu somatostatin analog). Using the Interview Fusion Workstation (Mediso), a total of 872 metastases were manually segmented. Conventional PET parameters as well as textural features representing intratumoral heterogeneity were computed. The prognostic ability for PFS and overall survival (OS) were examined. After performing Cox regression, independent parameters were determined by ROC analysis to obtain cut-off values to be used for Kaplan-Meier analysis. Within follow-up (median, 43.1 months), 75 patients showed disease progression (median, 22.2 m) and 54 patients died (median, 27.6 m). Cox analysis identified 8 statistically independent heterogeneity parameters for time-to-progression and time-to-death. Among them, the textural feature Entropy predicted both PFS and OS. Conventional PET parameters failed in response prediction. Imaging-based heterogeneity assessment provides prognostic information in PRRT candidates and outperformed conventional PET parameters. Its implementation in clinical practice can pave the way for individualized patient management.


INTRODUCTION
Over the past decades, a rising incidence of neuroendocrine tumors (NET) has been reported [1]. With a delay in diagnosis of 5-7 years, NET typically present at advanced stages. In clinical routine, physicians often have to rely on single tumor biopsies for treatment decisions and might therefore be prone to sampling bias which ultimately leads to either misdiagnosis or underestimation of therapeutic response [2,3]. Therefore, non-invasive, whole-body assessment of tumor heterogeneity Clinical Research Paper www.impactjournals.com/oncotarget is highly desirable. Due to its ability to visualize functional alterations on a molecular level instead of pure morphological characteristics, positron emission tomography (PET) has proven its prognostic value in risk stratification for several types of cancer [4][5][6].
As a prerequisite, pre-therapeutic somatostatin receptor (SSTR)-PET/computed tomography (CT) is mandatory to confirm adequate receptor density on the tumor cell surface [11,12]. Consequently, the imaging agent 68 Ga-DOTATATE was recently approved by the FDA which will also promote use of PRRT and simultaneously emphasizes the need for reliable response prediction prior to treatment initiation.
In this present multi-center trial, we aimed to elucidate the prognostic capability of intratumoral heterogeneity parameters assessed by baseline SSTR-PET/ CT in patients scheduled for radionuclide therapy.

Correlation of clinical parameters with both PFS and OS
Cox regression analysis of clinical parameters (as given in Table 1) was performed. Regarding PFS and OS, cumulative dose was the only parameter reaching significance (PFS, p < 0.02, OS, p < 0.01). Moreover,

Correlation of PET parameters with both PFS and OS
In Cox hazard analysis, the only parameter that showed significant correlation for both PFS (p = 0.02) and OS (p = 0.0002) was Entropy with Hazard Ratio (HR) of 0.59 and 0.35, respectively. In addition, Skewness showed significant correlation (p = 0.04) with OS with an HR of 0.58.
In Receiver Operating Characteristics (ROC) analysis of the independent parameters Entropy, Correlation, Short Zone Emphasis and Homogeneity demonstrated significant prognostic ability for PFS. Significant prognostic values for OS were found for Entropy, Correlation, Homogeneity, Short Zone Emphasis and Size Variation. Of the investigated conventional PET parameters, Tissue Receptor Expression (TRE) was significant for OS (p = 0.003), whereas all other parameters failed in response prediction. Details can be found in Table 2  Kaplan-Meier analysis using the threshold evaluated by ROC revealed a significant distinction between high-and low-risk patients for both PFS and OS for the following textural parameters: Entropy, Correlation, Short Zone Emphasis and Homogeneity. In addition, TRE showed a significant distinction for OS. Details including the negative and positive hazard ratios can be found in Table 3, respective Kaplan-Meier-plots for PFS are given in Figure 1 and for OS in Figure 2. An overview of the respective AUC values for both standard and heterogeneity PET parameters regarding PFS and OS are given in Table

DISCUSSION
In the present study, we investigated the prognostic value of PET-assessed tumor heterogeneity in patients scheduled for PRRT. Several textural characteristics like Entropy, Skewness, Correlation, Short Zone Emphasis and Homogeneity demonstrated superior diagnostic capability than standard PET parameters such as mean and maximum standardized uptake value (SUV max / SUV mean ).
Several studies have reported the feasibility of texture analysis for heterogeneity assessment and its prognostic implication for individual patient outcome based on pretherapeutic CT and magnetic resonance imaging, e.g. in non-small cell lung carcinoma or prostate cancer [13][14][15]. Due to its ability to visualize whole-body tumor burden on a molecular level, PET-based tumor heterogeneity offers certain advantages for assessing intraindividual heterogeneity patterns in tumor biology. Consequently, prognostic capability of imaging-derived intratumoral heterogeneity using 18 F-FDG but also SSTR-PET/CT has been evaluated in several tumor entities, such as breast, rectal, thyroid or esophageal cancer [4,6,[16][17][18]. In NET, given the complexity of various diagnostic procedures and treatment options, early identification of subjects likely to benefit from PRRT would be of great value for individualized treatment tailoring. As demonstrated in the present study, application of textural parameters could be helpful in differentiating high-risk from low-risk groups during PRRT. Strikingly, the 4 parameters Entropy, Correlation, Short Zone Emphasis and Homogeneity provided a significant distinction between responders from non-responders (Table 3, Figure  1 and 2).
In NET, conventional CT-derived therapy response assessment failed to predict disease-related progression or survival [19,20]. Regarding functional imaging, data is unequivocal: On the one hand, increasing uptake in SSTR scintigraphy (Octreoscan ® ) was reported as a significant predictor of PFS [21,22]. On the other hand, conventional PET parameters like SUV max failed to predict time-to progression in SSTR-expressing tumor entities [23]. In line with this finding, Gabriel and colleagues also demonstrated no benefit from conventional 68 Ga-DOTATOC PET parameters neither for response nor survival prediction in GEP NET patients scheduled for PRRT [24].
By contrast, heterogeneity parameters demonstrated prognostic value in this study, thereby, outperforming most conventional PET parameters. Of note, only the standard feature TRE proved useful in terms of OS prediction. Not surprisingly, the cumulative treatment dose correlated significantly with both PFS and OS. This could be regarded as a kind of intra-observer quality control of our study: with increasing progression-free and overall survival, more treatment cycles are performed and, consequently, the administered dose rises.
Interestingly, well-established histology or serumbased parameters of tumor aggressiveness or burden, such as the proliferation index Ki67 (PFS, p = 0.09) or CgA (OS, p = 0.06), failed to reach prognostic significance. This finding might be influenced by sampling bias: Couvelard and colleagues showed that two randomly taken cores from the same metastasis of pancreatic NETs led to   a change in grading in half of the cases [25]. Therefore, especially in patients with multiple lesions, non-invasive and reliable whole-body assessment of intra-tumor heterogeneity might yield completely different results [29] and contribute to individualised treatment decisions [30].
Diagnostic imaging of NET is the domain of SSTR agonists as well as 18 F-FDG PET/CT in case of dedifferentiation [31]. More recently, specific ligands targeting C-X-C motif chemokine receptor 4 (CXCR4) for diagnostic and therapeutic purposes were also introduced [32-37] and in the future, a combination of these abovementioned radionuclides might also potentially offer additional insight in NET biology and its underlying heterogeneity.
This study has several limitations: As disadvantage of a large multicentric trial, imaging as well as therapeutic protocols might differ from center to center. Especially when dealing with elaborated image analysis as textural features, varieties due to different PET/CT machines and different acquisition protocols may occur. However, using 18 F-FDG, Tixier et al. proved robustness of certain local or regional characterization features (e.g. Entropy), which also reached significance in our analysis [38]. Moreover, we are aware of the multiple testing problem; however, even when we correct for alpha inflation, the majority of tests remain significant.
Though NET is a rare disease with an annual incidence of approximately 5 new cases per 100.000 inhabitants [1], we were able to enroll a cohort of more than 140 patients which underwent PRRT. Therefore, advantages for such a multicentric trial might overcome the disadvantage of different diagnostic or therapeutic protocols. Also the low incidence of this disease complicates prospective reproduction of such analysis, which would be still the next step to achieve more standardized results.
In conclusion, in this multicenter trial enrolling 141 patients, tumor heterogeneity as assessed by baseline SSTR-PET/CT proved prognostic value in PRRT candidates and outperformed common conventional PET parameters. Assessment of intratumoral heterogeneity might significantly contribute to a more individualized

Contrast 2 st
Measures the difference of the grey value when going to the next voxel. It is high when the intensity changes very often between single voxels.

Short Zone Emphasis 3 st
Measures the distribution of short zones. It is highly dependent on the occurrence of small zones and is expected to be large for fine textures.

Size Variation 3 st
Describes the variation in the size of different substructures in an image (VOI). www.impactjournals.com/oncotarget patient management and treatment tailoring.

MATERIALS AND METHODS
All patients gave written and informed consent to the treatment and imaging procedures. The study was approved by local institutional review boards or the requirement for additional approval was waived due to the retrospective character of this study.
Clinical and routinely acquired characteristics of the patient cohort including sex, age, prior therapies, site of metastases, CgA, Ki67, and administered activities are given in Table 1.
Radiopeptide therapy was performed according to the Rotterdam protocol as outlined by Kwekkeboom et al. as well as The joint IAEA, EANM and SNMMI practical guidance on a compassionate use basis, respectively [11,19]. A total of 709 treatment cycles (median, 4, range, 1-6) with a median of 7.3 Gigabecquerel (GBq) (range, 0.9-9.0 GBq) per cycle with 177 Lutetium ( 177 Lu)-labeled somatostatin analog ( 177 Lu-DOTATATE/-DOTA-D-Phe-Tyr3-octreotide (DOTATOC)) were performed. Imaging including both functional (SSTR-PET/CT) and/ or morphologic imaging (CT) modalities was conducted every 3-6 months after PRRT [11]. PFS was defined according to Response Evaluation Criteria in Solid Tumors 1.1 (RECIST 1.1) by serial radiological assessment starting from the time point of baseline imaging and/or according to clinical signs of progression [11, 39. For the calculation of overall survival (OS) the time interval between the pre-therapeutic PET examination and the date of death was used.

PET/CT imaging
Prior to PRRT, all patients underwent SSTR-PET/ CT according to EANM guidelines to assess tumor receptor expression [11]. A median of 124 Megabecquerel (MBq) (range, 61-239 MBq) of 68 Ga-DOTATATE/-TOC was injected intravenously. After 60 minutes, imaging was performed using the following devices: Bonn, Biograph 2 PET/CT (Siemens Medical Solutions, Erlangen, Germany); Wuerzburg, Biograph 64 (Siemens Medical Solutions, Erlangen, Germany); Munich, Gemini TF PET/CT (Philips Medical, Eindhoven, Netherlands) or Siemens Biograph 64 (Siemens Medical Solutions, Erlangen, Germany); Hannover, Biograph 2 (Siemens Medical Solutions, Erlangen, Germany). All data was reconstructed using iterative algorithms implemented by the manufacturer and depending on the routine protocol of the different medical centers. Scatter and attenuation correction was performed based on the different transmission data.

Image interpretation
Analysis of the dataset was performed at the university hospitals of Bonn and Wuerzburg. All image data were transferred to an Interview Fusion Workstation (Mediso Medical Imaging Systems Ltd., Budapest, Hungary). Conventional parameters (such as SUV mean /SUV max , TRE)) were derived. TRE is the product of segmented lesion volume times the mean activity uptake, analog to the total lesion glycolysis in 18 F-fluorodeoxyglucose ( 18 F-FDG) PET. For analysis of tumor heterogeneity the largest lymph node, bone and visceral metastases were assessed. For systembased analysis the mean value of these lesions for each individual system was used. For patient-based analysis, the mean value of all segmented lesions per patient was performed. Metastases smaller than 15 mm were not taken into account to avoid partial volume effects. Manual segmentation was executed in combined PET/ CT data (Figure 3.1). Several different textural parameters representing intratumoral heterogeneity were derived from every individual lesion and were divided in first order parameters (e.g., Coefficient of variation, COV), second order parameters (e.g., Entropy, Correlation) and higher order parameters (e.g., Grey Level Uniformity, Intensity Variation, Short Zone Emphasis). A detailed description can be found in Table 4, Figure 3.2, [26] and [27]. For comparison to heterogeneity parameters, conventional diagnostic parameters were evaluated: morphologic volume of the lesion, maximum standardized uptake value (SUV max ), mean SUV (SUV mean ), and TRE were assessed.
The SUV was calculated according to the body weight of the patient. In total, 872 lesions (median, 6, range, 1-23 per patient) were manually segmented.

Statistical analysis
First, Cox multiparametric regression was applied to determine independent prognostic parameters of PFS and OS as well as estimation of HR. Due to multi-colinearity of different textural parameters, only the following parameters were included in the final Cox regression: COV, Skewness, Entropy, Homogeneity, Correlation, Contrast, Short Zone Emphasis and Size Variation. For all independent parameters, ROC analysis was obtained to estimate the optimal cut-off value for the individual parameters to assess progress and OS in the follow-up period. For this purpose, the Youden index was used to maximize the sum of sensitivity and specificity [40]. The area under curve (AUC) was calculated including the exact binominal confidence intervals (95% confidence level). Statistical significance of the prognostic capability was assumed when the critical value of 0.5 was not included in the confidence interval. For the parameters showing such significance, the relationship to both PFS and OS was analyzed using Kaplan-Meier plots. Kaplan-Meier analysis was performed using thresholds established by prior ROC analysis. Differences between Kaplan-Meier curves were evaluated using nonparametric log-rank tests, considering differences with a p value smaller than 0.05 to be significant.

CONFLICTS OF INTEREST
All authors had full control of the data and information submitted for publication. RB has a noncommercial research contract with Mediso Medical Imaging Systems, RB is on the speaker's bureau for Mediso Medical Imaging Systems, and NZ is employed by Mediso Medical Imaging Systems, LP was employed by Mediso Medical Imaging Systems while largest parts of the study were performed. All other authors had full control of the data and information submitted for publication. No potential conflicts of interest were disclosed by the other authors.
Parts of this work have been presented at the Annual Meeting of the Society of Nuclear Medicine and Molecular Imaging, 2016, San Diego, United States.

Editorial note
This paper has been accepted based in part on peerreview conducted by another journal and the authors' response and revisions as well as expedited peer-review in Oncotarget.