Supplementary information
Voxel and Histogram analysis
We considered voxel and histogram analysis to further evaluate response to therapy.
Methods:
Voxel analysis
Voxel quantification analysis was performed on filtered data by determining the number of voxels in each tumour volume and then binning the number of occurrences of each voxel intensity value. The maximum voxel intensity was limited to 30 (single unit intervals) as none of the tumours had a voxel intensity value greater than 25. Each voxel-intensity was then normalised by injected dose and body weight and decay corrected to obtain the SUV for the voxel. The variable representing high intensity voxels (HiVox: ratio of sum of voxels with SUV≥2 to sum of all voxels) was computed. The arbitrary SUV cut-off value is typically characteristic of high FLT uptake in different malignant lesions (16, 26, 27).
Histogram analysis
After summation of voxel intensities from all target lesions, per-patient voxel intensity histograms were generated. Histogram-derived parameters were (a) mean, median voxel intensity; (b) 10th, 25th, 50th, 75th, and 90th percentiles (which indicated the voxel intensity below which the corresponding percentage of all voxel intensities lie); and (c) skew, or γ , and kurtosis, or K (measures of the asymmetry of voxel intensity distribution). Percentage change of histogram-derived parameters between baseline and post-treatment scans were then estimated. Each patient’s high voxel intensities (above 90th percentile) were set the same as for the baseline scan cut-off, to examine the fraction of higher voxels post treatment. This was to account for the fact that high voxels will always be the same against a background of large stromal component for the primary disease in pancreatic cancer.
Results: A voxel-by-voxel representation of data was also computed to appreciate the heterogeneity of the tumours (Online-Resource-2a-d). KSF-derived voxel-based data were available for all pancreatic tumours and 8 liver metastases (8 of 11, 73%) (Table-1). The variable representing high intensity voxels (HiVox) derived from KSF was further explored with respect to changes between NP and P (Online-Resource -3a-d).
Effect of treatment: Most tumours also displayed some degree of reduction in the HiVox. The mean percentage reduction in Hivox after treatment was 21%. There were no significant changes in the Hivox between NP and P on group analysis of the data (Online-Resource-4). Further histogram analysis of the voxel intensities before and after treatment, also did not show any significant difference between NP and P (Online-Resource-6).
Group analysis of the imaging data was also performed. For the variable SUV60,max, the difference between baseline and post-treatment FLT uptake showed a statistically significant decrease in the non-progressors and a significant increase in progressors. However, there were no significant changes in the SUV60,ave between NP and P (Online Resource 7).
Immunohistochemistry
We considered the tissue expression of hENT1 and Ki-67 on diagnostic tumour biopsies. Only five samples were suitable for immunohistochemistry (IHC) as the remainder of the samples were cytology samples.
Methods: Expression of hENT1 was determined by IHC using a modified version of the antigen retrieval protocol firstly described by (44). Briefly, tissue sections were de-paraffinized in xylene, rehydrated in graded alcohols and heated in a microwave oven at 900W for 30min in citrate buffer at pH 6.0. Before immunostaining, slides were cooled at room temperature and quenching of endogenous peroxidase activity was obtained by incubation with a 3% solution of H2O2 for 5min. The primary antibody for hENT-1 was incubated overnight at the concentration of 1:500 (Cat. nr. Ab48607, Abcam, Cambridge, UK). Immunostaining for Ki-67 (Leica Microsystems, Wetzlar, Germany) was carried out at 1:800 dilution, respectively. Tissue sections were incubated with the secondary antibody for 1h at room temperature and then processed using the Polymer-HRP Kit (BioGenex, San Ramon, CA, USA) with development in Diaminobenzidine and Mayer's Haematoxylin counterstaining. Appropriately chosen tissue samples were used as external positive control during each reaction to confirm its specificity. Omission of the primary antibody was used as negative control reactions, resulting in the absence of staining in all cases. All the immunostaining procedures were performed centrally within the Imperial College NHS Trust Pathology Department.
Protein expression was quantified using the histoscore (HS) method. Briefly, each tumour specimen was scored on a semiquantitative scale ranging from 0 to 300, with the final score resulting from the percentage of tumour cells staining positively (range 0–100) multiplied by staining intensity graded as negative, weak, moderate or strong (range 0–3). The Ki-67 labelling index was expressed as the percentage of immunopositive nuclei from a minimum of 500 nuclei in at least five microscopic fields. The median HS value was used as a cut-off level to discriminate high vs low expression of each biomarker. Two observers (FAM, RS) blinded to the clinical data scored all the cases and consensus was reached in case of significant discrepancy between the individual scores.
Results: Expression of hENT1 was predominantly diffuse (cytoplasmic and nuclear: Online-Resource-8). Only one sample had membrane expression of hENT1. The median HS value for hENT1 was 150 (range 0-270). The median expression of Ki-67 was 5% (range 2 – 20%). No associations were determined between hENT1 and Ki-67 expression, and FLT parameters given the small number of samples.