Study population
This study included consecutive patients who underwent staging with 68Ga-PSMA-11-PET for newly diagnosed intermediate or high-risk prostate cancer at the University Hospital Zurich from April 2016 to May 2018. All patients with no radical prostatectomy (RPE) specimen available were excluded. The local ethics committee approved the study protocol (BASEC Nr. 2018–01284) and all patients gave a general written informed consent for use of their data. Relevant clinico-pathological characteristics such as patients’ age at the time of operation, tumour stage and (modified) Gleason Score respective WHO/ISUP prognostic grade group were collected.
Histopathological parameters and Immunohistochemistry
Sixty-two formalin-fixed, paraffin-embedded (FFPE) RPE specimen were evaluated on 2 µm hematoxylin & eosin (H&E) stained sections. One representative slide from the RPE specimen was chosen for further investigation, harbouring the largest area of tumour and therefore defining the dominant tumour lesion.
Staging and grading were done according to the WHO/ISUP/UICC guidelines (21, 22). Separate grading of the dominant tumour lesion was done and used for further correlation analysis. The tumour area and maximum diameter on each slide were measured digitally. Very small carcinoma lesions (maximum diameter < 5 mm) were excluded from statistical analysis because of a natural resolution limit of PSMA-PET scans due to partial-volume effects (23).
A newly developed type of growth pattern (infiltrative vs. expansive) of each prostate cancer lesion was determined. We defined infiltrative growth as entrapped benign glands within the carcinoma complexes. An expansive growth pattern showed a tumour infiltration of pure carcinoma glands (without intermingled benign glands) within an area of at least 3 circles of 5 mm2 (radius 1.26 mm).
International Society of Urologic Pathology (ISUP) and World Health Organisation (WHO) guidelines, according to WHO/ISUP 2014 guidelines according to WHO/ISUP 2014 guidelines
Immunohistochemical staining for PSMA (DAKO, M3620, clone 3E6, 1:25) was performed as described previously (24). The predominant PSMA expression patterns were visually quantified using a four-tiered system (0 = negative, 1+ = weak, 2+ = moderate, 3+ = strong) for both membranous and cytoplasmic PSMA expression by two board certified, experienced genito-urinary pathologists (J.H.R, N.J.R.). Examples of expression patterns are shown in Fig. 1. Furthermore, tumour areas without PSMA expression were quantified in steps of 5%, 10% and further 10% increments in relation to the total tumour area, as percentage PSMA-negative tumour area (PSMA%neg) as a consent of both pathologists. Heterogeneity was defined by differences in the staining pattern of at least 5% of the representative tumour slide (Fig. 2).
Slides were digitalized (Nanozoomer NDP digital slide scanner C9600-12) using the Hamamatsu NDP.view 2.8.24 Software.
Imaging
Patients underwent clinical routine 68Ga-PSMA-11-PET/computed tomography (CT) on a Discovery VCT 690 PET/CT (GE Healthcare, Waukesha, WI, USA) or on a Discovery MI PET/CT (GE Healthcare, Waukesha, WI, USA) or 68Ga-PSMA-11-PET/MRI (SIGNA PET/MR, GE Healthcare, Waukesha, WI, USA) after a single injection of 68Ga-PSMA-11 (mean dose ± standard deviation (SD) 130 ± 18 MBq, range 81-171 MBq). The institutional protocol is in agreement with the EANM and SNMMI procedure guidelines (25). Details are given in the supplements.
Imaging analysis
The acquired PET/CT and PET/MR images were analysed in a dedicated review workstation (Advantage Workstation, Version 4.6 or 4.7, GE Healthcare), which enables the review of the PET and the CT or MR images side by side and in fused mode. Every patient was discussed in a multidisciplinary set up including a pathologist and a nuclear medicine physician and radiologist with the selected pathology slide available alongside with the PET data. The corresponding area on PET images with the dominant tumour lesion was identified and PSMA uptake quantified using the maximum standardized uptake value (SUVmax). There is a wide range of proposed cutoffs to detect significant prostate cancer from SUVmax 3.15 (20) to up to SUVmax 9.1 (26). For visual identification an clear uptake above background might be more efficient than a absolute cutoff, and given that there were no lesions in the central zone in our cohort, and to select clear positive lesions we decided to take a PSMA uptake of SUVmax ≥ 5 as definition of PSMA-PET positivity (18).
An additional analysis for SUVmax ≥ 4 is given in the supplements, to rule out a systematic underestimation. For the correlation between immunhistochemical (IHC)-parameters and PET quantification the spatial resolution of the PET scanners was taken into account. Therefore, a tumour diameter of 5 mm or more on histology was considered necessary for accurate quantification of PSMA-accumulation limiting the impact of partial-volume effect (23).
Correlation of histopathological and immunohistochemical parameters with SUVmax values
Correlations between histological parameters, immunhistochemical PSMA expression patterns and SUVmax values were calculated using Mann-Whitney U test, Kruskal–Wallis test and Pearson’s correlation. An optimal cutoff for PSMA%neg was determined using Receiver operating characteristic (ROC) analysis. We investigated the association between a combination of histological parameters and immunhistochemical PSMA expression patterns with a negative PSMA-PET scan using a multiple logistic regression analysis.
Statistical analysis
Normal distribution was tested using the Kolmogorov–Smirnov test. Comparisons were calculated with Mann-Whitney U test for binary variables and Kruskal–Wallis test for multiple variables. Correlations were done using bivariate Pearson’s correlation. Discrimination was evaluated using area under the receiver operating characteristic (ROC) curve (AUC). The variables entered in the multiple logistic regression analysis were selected by univariable logistic regression with a p-value cut-off point of 0.05. For the logistic regression analyses ordinal variables were treated as continuous. Multicollinearity was assessed using Variable Inflation Factors (VIF). Two-sided p values < 0.05 were considered statistically significant. Correlations and ROC curve analysis were performed using SPSS Version 26 (IBM, Armonk, New York, USA). Logistic regression analyses were performed using R (R version 4.0.2; R Foundation for Statistical computing, Vienna, Austria). Graphs were generated using GraphPad Prism v8.