Blood Volume as a new functional image-based biomarker of progression in metastatic renal cell carcinoma

RECIST v1.1 has limitations in evaluating progression. We assessed Dynamic Constrast Enhanced Computed Tomography (DCE-CT) identified Blood Volume (BV) for the evaluation of progressive disease (PD) in patients with metastatic renal cell carcinoma (mRCC). BV was quantified prospectively at baseline, after one month, then every three months until PD. Relative changes (ΔBV) were assessed at each timepoint compared with baseline values. The primary endpoint was Time to PD (TTP), the secondary endpoint was Time to the scan prior to PD (PDminus1). Cox proportional hazard models adjusted ΔBV for treatments and International mRCC Database Consortium factors. A total of 62 patients had analyzable scans at the PD timepoint. Median BV was 23.92 mL × 100 g−1 (range 4.40–399.04) at PD and 26.39 mL × 100 g−1 (range 8.70–77.44) at PDminus1. In the final multivariate analysis higher ΔBV was statistically significantly associated with shorter Time to PD, HR 1.11 (95% CI 1.07–1.15, P < 0.001). Also assessed at PDminus1, higher ΔBV was significantly associated with shorter time to PD, HR 1.14 (95% CI 1.01–1.28, P = 0.031). In conclusion, DCE-CT identified BV is a new image-based biomarker of therapy progression in patients with mRCC.

The inclusion criteria for the DaRenCa Study-1 were: histologically verified clear cell mRCC, no prior oncologic treatment, measurable metastatic disease according to the RECIST v.1.1 criteria, favorable or intermediate risk MSKCC group, Karnofsky Performance Status ≥ 70% and adequate kidney function (serum creatinine < 150 micromol/L). The inclusion criteria for AIS were: histologically verified mRCC, no prior oncologic treatment and adequate kidney function (estimated glomerular filtration rate (eGFR < 35 ml/min).
Using the same cohort of patients as in this current study, the association between baseline BV and survival outcome adjusted for baseline features 13 and the association between DCT-CT parameters and early treatment response have been published [15][16][17] .
Approval by the Central Denmark Region Ethics Committee and The Central Denmark Data Protection Agency was granted and written informed consent was obtained before inclusion started. The study was performed in accordance to the approved guidelines by the Central Denmark Region Ethics Committee and The Central Denmark Data Protection Agency. DaRenCa-1 was registered at ClinicalTrials.gov (identifier NCT01274273) and approved by the Danish Medicines Agency.
Electronic medicinal charts were used to retrieve information about baseline clinical factors, treatments and baseline IMDC prognostic factors 14 .
A total of 105 patients (DaRenCa-1, N = 76 and AIS, N = 29) and 483 analyzable DCE-CT scans were included in the study, where 62 and 64 patients had an analysable DCE-CT scan at PD and the scan prior to PD (PDmi-nus1), respectively, Fig. 1.

CE-CT and DCE-CT.
A routine contrast enhanced (CE) CT was performed at baseline and every 3 months until progression and was assessed according to RECIST 1.1 2,3 . All clinical decisions were based on routine CE-CT scan results. DCE-CT was performed at baseline, after 1 month of therapy and every three months until progression. Initially, a DCE-CT scan of a single target lesion was performed, followed by a routine CE-CT scan of the thorax, abdomen, and pelvis. Patients remained supine for 10 min between the scans. Based on prespecified protocol criteria, an experienced radiologist selected a representative target lesion that was optimal for functional imaging. The target lesions scanned with DCE-CT were located in the lung (N = 21), pleura (N = 5), supraclavicular/ thoracic lymph nodes (N = 16), retroperitoneal lymph nodes (N = 8), kidney (N = 14), kidney bed (N = 6), adrenal gland (N = 3), bone (N = 12), liver (N = 8), pancreas (N = 7), extra/intra abdominal soft tissue (N = 3 and N = 2, respectively); the same DCE-CT technique was used irrespectively of target lesion location.
Before each DCE-CT scan, 60 ml iodixanol (Visipaque, GE Healthcare) 270 mg I/mL at 6 mL/s was administered intravenously. Before each routine CE-CT, iodixanol (Visipaque, GE Healthcare) 270 mg I/ml based on body weight (maximum 180 ml) at 4 mL/s was administered intravenously. In the event of minor reactions to iodixanol, patients were subsequently given iohexol (Omnipaque, GE Healthcare) 300 mg I/mL instead.
Routine CE-CT scans were obtained using attenuation based current modulation; 120 kVp of peak voltage, 0.75 s of rotation time, a collimation of 64/128 × 0.625 mm, and a pitch of 0.925. 4D imaging analysis. The prototype software program Advanced Perfusion and Permeability Application, Philips (Philips, Healthcare) was used to analyze DCE-CT data in four dimensions (4D). After loading the dynamic data, a spatial filtration and motion correction was performed using a non-rigid registration. The software program used the deconvolution method 19 to calculate BV (mL × 100 g −1 ) and display corresponding BV maps.
Data were then loaded into Intellispace 6.0 Multimodality Tumor Tracking (Philips, Healthcare). A semiquantitative three dimensional (3D) sculpt-tool was used to delineate the target lesion as the Volume of Interest (VOI) using the morphological DCE-CT images at arterial peak enhancement. When the 3D analysis was combined with the time dimension in DCE-CT due to repeated measurements, it resulted in a 4D analysis. A www.nature.com/scientificreports/ radiologist, blinded to the treatment group and survival outcomes, performed all the analyses. This particular method has previously shown excellent interobserver correlations 17 . MATLAB (v. R2015b, MathWorks Inc.) was used to analyze the dynamic BV data based on the VOI on DCE-CT images at peak arterial enhancement. Histogram values of BV were extracted based on the DCE-CT VOI using in-house programmed scripts. The median values for BV was calculated for each histogram and used for assessment, as this value previously has shown the best correlation with patient outcome in mRCC 17 . Statistical analysis. Compared to baseline, relative changes in BV were calculated in percent (%) at each scan timepoint (X timepoint ) until RECIST v1.1 defined PD: www.nature.com/scientificreports/ The association between baseline factors and Time to progression (TTP), defined as the time between baseline and the scan timepoint of RECIST v1.1 defined PD or cancer related death, whichever came first, was examined using a univariate Cox proportional hazards models expressed as hazards ratios (HR) with 95% confidence intervals (CI). Baseline univariate factors, including individual IMDC risk factors, prior nephrectomy, age and gender, with P < 0.10 and treatment groups were included in the multivariate Cox proportional hazards models. ΔBV was assessed as continuous variables presented as 20-percent point increasements at each timepoint. Univariate and multivariate Cox proportional hazards models were used to examine the association between ΔBV and PD using two different endpoints. The primary endpoint was TTP and the secondary endpoint was Time to the scan prior to PD (PDminus1), i.e. to assess if PD could be detected on the scan one timepoint earlier.
For patients experiencing cancer related death or PD due to clinical evaluation or supplementary imaging, a consensus was made regarding classification of the latest DCE-CT scan being either PD (N = 9) or PDminus1 (N = 4) events.
The effect between treatment groups and ΔBV were examined by constricting the multivariate analyses to the patients treated with angiogenesis inhibitors and the immunotherapy, respectively. The proportional hazards assumptions were tested graphically by Schoenfeld residuals against the time and were fulfilled.
The median follow-up time in alive patients was assessed using the reverse Kaplan-Meier survival curves. A Receiver Operating Characteristic (ROC) analysis was performed to identify a possible cut-off at PD and PDminus1 for the relative change in the continuous DCE-CT parameters that were deemed significant in the multivariate Cox regression analysis. The ROC generated area under the curve (AUC) < 0.8 was considered not to have predictive value and was therefore not eligible for estimating a cut-off value 20 . All tests were two-sided and P values below 0.05 were considered as statistically significant. IBM SPSS Statistics for Windows (Version 27.0, IBM Corp.) was used to perform all statistical analyses.

Results
Patients. Baseline patient characteristics are presented in Table 1.

Multivariate analysis of relative changes in BV.
In the final multivariate analysis, higher ΔBV were independently associated with a shorter time to PD, HR 1.11 (95% CI: 1.07-1.15, P < 0.001), Fig. 2 and Table 3.
Receiver operating characteristic analysis. AUC

Discussion
This study is the first to demonstrate that higher DCE-CT identified ΔBV is a new image-based biomarker of therapy progression in patients with mRCC. Our results show that patients with a 20%-point higher ΔBV have www.nature.com/scientificreports/ a 11% higher risk of having RECIST defined PD on the current conventional CT scan, indicating that the relative change in BV can be used as a biomarker to predict progression at the timepoint of RECIST v.1.1 defined progression. Furthermore, we find that patients with a 20%-point higher ΔBV without RECIST v1.1 defined PD on the current conventional CT scan have a 14% higher risk of RECIST v1.1 defined PD on the subsequent scan, suggesting that this biomarker may predict progression at the scan prior to RECISTv.1.1 defined progression. These findings strengthen the utility of BV as a biomarker for progression. Therefore, DCE-CT identified BV may have the potential to be used as a helper to RECIST v1.1 in identifying PD in patients with mRCC and may have the potential to support clinical decision-making during treatment monitoring in mRCC, when RECIST v1.1 is uncertain. However, it was not possible to define a cut off value for BV, limiting the implementation in the clinical setting. Further development in DCE-CT functional imaging in a larger cohort is encouraged, and is needed before it can reach clinical daily life in treatment decision making. The association between ΔBV and PD was independent of treatment group at the timepoint of RECIST v.1.1 defined progression. However, at the timepoint prior to RECISTv.1.1 defined progression, the association was only significant for patients treated with IL-2 based therapies. These findings could indicate that BV increases more rapidly in patients treated with angiogenesis inhibitor. Further research assessing this matter in a larger cohort is encouraged.
RECIST v1.1 relies solely on morphological information and because changes in tumor size may lag behind pathophysiological changes within the tumor, it is a suboptimal response evaluation tool for patients treated with targeted therapy [21][22][23][24] . Characterizing unequivocal progression and changing therapy to the next treatment line at the appropriate time using RECIST v1.1 can be difficult. In the randomized phase III trial comparing pazopanib and sunitinib as first-line therapy in mRCC patients, only half of the patients continued to second-line treatment after PD 8,9 . Pseudoprogression has not been described in the literature on IL-2 immunotherapy and was not observed in this current study. However, pseudoprogression is a challenge during checkpoint immunotherapy and was highlighted in the study of Escudier et al., where the survival rate almost doubled in patients treated with nivolumab beyond RECIST v1.1 defined progression 25 . These studies illustrate and highlight the clinical dilemma of stopping therapy too late or too early.
The increasing need for a better response evaluation tool has led to attempts to improve the RECIST v.1.1 criteria. Choi was the first to combine morphological and functional information measured as CT contrast uptake (Hounsfield unit) in a target lesion on CE-CT. However, the CT contrast uptake was only measured in a single slice making it a major limitation due to intratumoral heterogenicity 26 . The method in this current study evaluated the entire target lesion and thus took into account the intratumoral heterogenicity, making this method superior to the single slice method.
DCE-CT derived BV is a robust parameter, independent of changes in cardiac output, as described by Miles et al. 12 . BV values can be affected by motion and beam hardening artifacts 19 , which can be minimized by instruction in shallow breathing, and by avoiding target lesions close to prosthetics or areas with high concentrations of contrast media agents.
Several studies, focusing on the prognostic significance of DCE-CT parameters at baseline, as well as the evaluation of early treatment response in mRCC, have been conducted 13,[15][16][17]27,28 . The study of Drljevic-Nielsen et al. showed that high baseline BV was a favorable independent prognostic factor for survival outcome 13 , while the studies of Mains et al. showed that early reduction in BV was associated with favorable outcomes, whereas only a smaller reduction or an increase in BV were associated with worse outcomes [15][16][17] . Our study is the first to demonstrate that a relative higher BV, when compared with baseline, was associated with a higher risk of PD and thus a worse outcome. Summarized, a higher BV at baseline is a favorable feature, whereas an higher ΔBV during treatment is an unfavorable feature.To our knowledge no previous studies have assessed the potential of DCE-CT parameter BV as imaging-biomarker for progression, making our study a first of a kind.
Several limitations to this study must be noted. Firstly, motion artifacts occurred even though patients were instructed in shallow breathing. Secondly, the increased radiation dose in DCE-CT induces a larger stochastic risk for a radiation-induced cancer. However, patients with mRCC have a reduced life expectancy making the risk for radiation-induced cancer very low. Thirdly, of the 105 included patients at baseline, ΔBV was only analyzable for 64 patients at PDminus1 and 62 patients at PD, which could explain the low AUC in the ROC-analyses. Furthermore, it illustrates that DCE-CT is a demanding technique, which may limit the translation of DCE-CT from a research tool to a clinial tool. Fourthly, there is a risk of introducing target lesion selection bias, due to www.nature.com/scientificreports/ the relative short Z-axis of DCE-CT (8 cm). A single target lesion was chosen, but this lesion was not necessarily representative of other target lesions in the same patient due to intertumoral heterogenicity 29 . A fifth limitation is that the clinical decision making based on RECIST1.1 in this study represents a bias, as RECIST v1.1 may not to PDminus1 (28,85 mL × 100 g −1 ). CE-CT contrast enhanced Computed-Tomography, DCE-CT dynamic Contrast enhanced Computed-Tomography, PDminus1 scan prior to progressive disease.