Prediction of Biochemical Recurrence Based on Molecular Detection of Lymph Node Metastasis After Radical Prostatectomy

Take Home Message Our study shows that molecular detection of lymph node micrometastases with a panel of prostate epithelial markers seems to be more useful for early detection of biochemical recurrence than histology alone or using markers of stem cells, epithelial-to-mesenchymal transition, or reactive stroma.


Introduction
Organ-confined prostate cancer (PC) is cured by radical prostatectomy (RP) in only 70-80% of patients. This is mainly due to early dissemination of cancer cells and formation of occult metastases not manifest at the time of treatment of the primary tumor [1]. It is currently difficult to unequivocally identify high-risk patients in need of close, long-term follow-up.
The presence of histologically detectable pelvic lymph node (LN) metastases (pN1) is an important predictor of disease recurrence [2]. However, over 20% of patients classified as LN negative (pN0) will suffer recurrence despite effective local therapy [3]. This suggests that a conventional histopathological analysis of pelvic LNs may miss small metastatic foci. The discovery rate of micrometastases can be increased by immunohistochemical staining with antibodies against cytokeratins and prostate-specific antigen (PSA) [4,5]. Additional sensitivity is attained by quantitative real-time polymerase chain reaction (RT-qPCR) measuring the expression of PSA and prostate-specific membrane antigen (PSMA) [6][7][8]. A molecular LN analysis can identify pN0 patients with a higher risk of biochemical recurrence [6][7][8] and is superior to histopathological LN status [9]. The outcome of patients with occult LN metastases detected by molecular analyses is similar to that of pN1 patients [10].
Most studies on LN micrometastasis in PC patients have focused on the detection of prostate epithelial markers such as PSA and PSMA [6][7][8]. However, other markers may be more helpful in detecting LN micrometastases. In fact, PC circulating tumor cells (CTCs) present in peripheral blood [11] and early disseminating cancer cells found in bone marrow have a stem cell (SC) phenotype [12] and undergo epithelial-to-mesenchymal transition (EMT) [13], and may therefore escape detection based on epithelial markers. Metastasis-initiating cells most likely represent a subpopulation of CTCs and, consequently, probably also express SC features [14]. On the contrary, the presence of a reactive stroma in the primary tumor as well as alteration of the LN microenvironment was shown to predict recurrencefree survival after RP [15]. There are currently only limited data on the ideal number and type of markers for the molecular detection of PC recurrence.
In this prospective study, we assessed the expression of a panel of molecular markers in LNs from patients with organ-confined PC who were followed up for 10 yr. We investigated whether the additional assessment of mRNA expression of markers of prostate SC-like, EMT, and reactive stroma determined in our laboratory could increase the detection rate of LN micrometastases in PC patients. Finally, we evaluated the predictive value of molecular detection of LN metastases for biochemical recurrence. The length and width of the metastatic deposits were measured. If necessary, an immunohistochemical analysis was carried out.

Gene expression analysis
RNA extraction, cDNA synthesis, and RT-qPCR were performed as previously described [16]. Gene symbols and corresponding expression assays are listed in Supplementary Table 1.
Control LN specimens were obtained from four female patients undergoing surgery for noncancerous reasons and from one patient undergoing Millin's prostatectomy.

Immunohistochemistry
Immunohistochemical staining was performed on 47 deparaffinized LN sections of six pN1 patients with the primary antibodies listed in Supplementary Table 2, as previously described [17].

Statistical analysis
Statistical analysis was performed with GraphPad Prism version 6.0d (www.graphpad.com) and R version 4.0.3. (www.r-project.org). The Mann-Whitney test was used to compare mRNA expression between LNs from PC and control patients.
The Kruskal-Wallis test was applied to assess the association of LN status with clinical variables. The Spearman correlation coefficient rho was used to assess the presence and magnitude of monotonous trends between the level of evidence of LN metastases (pN0/molN0 < pN0/mo lN1 < pN1/molN1) and ordinal clinical risk factors. The median bRFS as well as bRFS at 10 yr together with 95% confidence intervals (CIs) was calculated using the Kaplan-Meier method and compared between LN status using the log-rank test. Hazard ratios (HRs) and 95% CIs were calculated using Cox regression models. All statistical tests performed were two sided, and a p value of <0.05 was considered statistically significant.

Expression of candidate marker genes in LNs of control and PC patients
We have compared the expression levels of candidate genes in seven LNs from three patients with macroscopic PC LN metastases and in 11 LNs from five control patients.
All markers of epithelial cells (PSA, EpCAM, PSCA, PSMA, NKX3-1, and AGR2) were expressed at a significantly higher level in LNs from PC patients than in those from control patients (Fig. 1A). In contrast, the levels of expression of the EMT markers (SNAIL, TWIST, and CXCR4) were either higher in control patients or not different between the two groups of patients (Fig. 1B) With the exception of TROP2, all proven or putative markers of SCs (ALDH1A1, NANOG, SOX2, OCT4, KLF4, EGR1,  BMI1, LGR5, LGR6, LRIG1, TSPAN7, and TSPAN13) were expressed in LNs from control patients to similar or even higher levels than in PC patients (Fig. 2). TROP2 expression was barely detectable in LNs from controls but could be measured in a few LNs from a PC patient.
Based on these results, we have selected the panel of epithelial markers to be further assessed for molecular detection of micrometastases in our cohort of 60 patients.

3.2.
Immunohistochemical detection of prostate epithelial markers in LNs from PC patients Protein expression of PSA, PSCA, EpCAM, PSMA, NKX3-1, and AGR2 was analyzed on LN sections from six pN1 patients. Tumor cells in LNs showed cytoplasmic expression of PSA, PSCA, and AGR2; membrane and cytoplasmic expression of PSMA and EpCAM; and nuclear expression of NKX3-1. Lymphoid cells were negative for all markers (Fig. 3).

PC patient and LN characteristics
The clinical and pathological characteristics of the patients are summarized in Table 1 Table 3).
In general, the pattern of expression of PSA, PSCA, EpCAM, PSMA, NKX3-1, and AGR2 was very heterogeneous among LNs of the same patient, as shown for representative LNs of six pN1 patients (Fig. 4). 3 Eleven of the 12 pN1 patients (91.7%) showed expression of one or multiple epithelial markers (pN1/molN1). Of the 21 pN0 patients who relapsed, eight (38.1%) showed no expression of any molecular marker (pN0/molN0), while 13 (61.9%) were positive for one or more marker type (pN0/molN1). Twelve patients overall were positive in the molecular analysis with no sign of recurrence. Fifteen patients were negative in the analysis and had no sign of recurrence (  (Fig. 5). The positive predictive value of molecular detection of epithelial cell markers was 66.7% (95% CI 49.0-81.4%) and the negative predictive value was 62.5% (95% CI 40.6-81.2%).

Discussion
Molecular detection of micrometastases in LNs from PC patients with localized disease has the potential to refine the diagnosis of high-risk PC patients. In this study, we investigated whether an extended panel of markers for prostate epithelial cells, SCs/progenitor cells, EMT, and reactive stroma would be useful to reveal the presence of occult metastases. Our data confirm previous reports demonstrating that molecular screening of LNs could identify patients at risk of biochemical recurrence, although they had been classified to have pN0 by histopathology [9]. However, 38.1% of the pN0 patients not showing any molecular marker expression (pN0/molN0) experienced biochemical recurrence. This absence of molecular positivity may be due to several reasons such as local recurrence rather than nodal metastases or due to the insufficient sensitivity of our detection assay as well as possibly quenching of the positive PCR result by the LN microenvironment of otherwise prognostic markers. On the contrary, all pN1 patients except one were also positive in the molecular screening. Nevertheless, when each   LN was considered, there was not always an overlap between RNA expression data and histopathological findings. This discrepancy may be due to the fact that, since only half of each LN was analyzed by one of the methods, the metastatic foci may have been located in one half only. Small metastases might also have been missed by the histopathological analysis since it was limited to one section per LN. In addition, in 33.3% of the cases with positive PCR results, the patients did not experience recurrence during the follow-up time. It is conceivable that in these cases, the molecular markers identify a population of dormant cancer cells in the LN. This state has been called ''micrometastatic dormancy'' by Ruppender and colleagues [18] and defines a group of tumor cells that cannot grow due to a restrictive proliferation/apoptosis equilibrium. It is currently not clear whether the population of patients with molecular detection of LN metastases might benefit from adjuvant therapies, and further research should address this question.
Notably, our results validate epithelial markers as the best markers for the detection of LN metastases. Besides the widely used epithelial markers PSA, EpCAM, and PSMA, we also measured the mRNA expression levels of PSCA, AGR2, NKX3-1, TROP2, and the TMPRSS2-ERG fusion gene. PSCA is a protein overexpressed in the majority of PC patients, and we have previously shown that SCs/progenitor cells from the primary tumor of PC patients express PSCA [19]. High PSCA expression has been correlated with poor prognosis and progression toward neuroendocrine PC [20], and is being investigated as a target for CAR T-cell therapy for castration-resistant PC [21]. Likewise, AGR2 is overexpressed in PC and is implicated in the control of cellular senescence [22]. The homeobox gene NKX3-1 involved in differentiation of the normal prostate epithelium is a very sensitive and specific marker of PC metastases [23]. TROP2 is a marker of prostate basal cells with SC characteristics, and high TROP2 expression by immunohistochemistry was recently shown to predict biochemical recurrence after RP [24]. Presence of TMPRSS2-ERG fusion in the LN did not prove to be a prognostic factor in a clinical study performed at our institution [25].
Our findings suggest that only the use of a panel of markers will allow the consistent detection of microscopic LN metastases, since not all markers are necessarily coexpressed. This might also have implications for the development of liquid biopsy panels. The most common liquid biopsy biomarkers are CTCs, circulating tumor DNA (ctDNA), and extracellular vesicles. It was recently shown that the levels of ctDNA are below the threshold for detection in serially collected plasma samples over 24 mo after RP and therefore not suitable for the detection of recurrence [26], while it very well recapitulates the genomic landscape detected in tissue biopsies of metastatic PC [27]. The number of CTCs positively correlates with biochemical recurrence after RP; yet, there is considerable phenotypic heterogeneity regarding the expression of androgen receptor and cytokeratine [28]. It is possible that a more extensive panel of prostate epithelial markers could improve the utility of CTCs as predictors of recurrence.
Our hypothesis that markers of stemness and EMT may be more sensitive than epithelial markers in detecting micrometastasis could not be validated, since all the markers analyzed were already expressed in control LNs to the same or an even higher extent than in metastatic LNs. Our choice of SC markers was based on our report in the BM18 xenograft model of PC that cells surviving castration and able to reinitiate tumor growth are characterized by the coexpression of ALDH1A1 or NANOG together with the  luminal marker NKX3-1 [16]. These castration-resistant cells also show mRNA expression of OCT4 and SOX2, LGR5 and LGR6, KLF4, EGR1, BMI1, LRIG1, and TSPAN7 and TSPAN13. For the markers of EMT used, there is a large body of evidence that SNAIL, TWIST, and CXCR4 are representative of PC, and correlate with metastases and poor outcome [29,30]. The presence of EMT, as assessed by the coexpression of CK8 and vimentin, was predictive of biochemical recurrence after RP, irrespective of other clinical factors such as Gleason grade, pathological stage, or surgical margins [31].
The concept of the tumor microenvironment as a decisive factor in the metastatic process is well established. However, similarly to what was observed with markers for stemness and EMT, the markers of reactive stroma did not discriminate between metastatic and control LNs. Our analysis was limited to four genes, ASPN, POSTN, SPARCL1, and MCAM, which we determined as potential markers of stromal activation in a xenograft model of PC bone metastasis [17]. It is conceivable that other stroma markers might be more suitable for the detection of LN metastases.
Potential limitations of our work are the cutting of the LN in half, which might miss tumor foci in the histopathological or molecular workup as well as the hypothesis-driven selection of markers that might differ from markers selected by high throughput screening.

Conclusions
In conclusion, the expression of markers of stemness, EMT, and reactive stroma in the LN microenvironment precludes their use in the detection of micrometastases. Consequently, prostate epithelial markers remain the best candidates for RP with LN dissection. Additional trials are needed to validate the clinical utility of molecular detection of epithelial prostate markers as prognostic biomarkers.
Author contributions: Bernhard Kiss had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Thalmann.
Critical revision of the manuscript for important intellectual content: Özdemir, Arnold, Kiss, Thalmann.
Administrative, technical, or material support: None.
Financial disclosures: Bernhard Kiss certifies that all conflicts of interest, including specific financial interests and relationships and affiliations relevant to the subject matter or materials discussed in the manuscript (eg, employment/affiliation, grants or funding, consultancies, honoraria, stock ownership or options, expert testimony, royalties, or patents filed, received, or pending), are the following: None.
Funding/Support and role of the sponsor: This study was supported by a grant from the Swiss National Foundation (320030_130862).