*Article* **Prognostic and Therapeutic Potential of the OIP5 Network in Papillary Renal Cell Carcinoma**

**Mathilda Jing Chow 1,2,3, Yan Gu 1,2,3, Lizhi He <sup>4</sup> , Xiaozeng Lin 1,2,3, Ying Dong 1,2,3, Wenjuan Mei 1,2,3 , Anil Kapoor 1,2,3,\* and Damu Tang 1,2,3,\***


**Simple Summary:** Papillary renal cell carcinoma (pRCC) is an aggressive kidney cancer. Currently, there are no effective prognostic biomarkers and lack of efficacious therapies in treating pRCC. We report a novel and critical pRCC oncogenic factor OIP5. Its expression is increased in pRCC and the upregulation is associated with adverse features. High levels of OIP5 effectively predict pRCC recurrence and fatality. OIP5 promotes pRCC cell proliferation and tumor formation through complex processes. A 66-gene multigene panel (Overlap66) was constructed. Overlap66 is novel and robustly predicts pRCC recurrence and fatality. High risk pRCCs stratified by Overlap66 are associated with immune suppression. Furthermore, PLK1 is a component gene of Overlap66; PLK1 inhibitor significantly reduced OIP5-promoted pRCC cell proliferation in vitro and tumor growth in vivo. Collectively, Overlap66 can effectively stratifies high-risk pRCCs and these tumors can be treated with PLK1 inhibitors. Our findings can be explored for personalized therapy in pRCC patients.

**Abstract:** Papillary renal cell carcinoma (pRCC) is an aggressive but minor type of RCC. The current understanding and management of pRCC remain poor. We report here OIP5 being a novel oncogenic factor and possessing robust prognostic values and therapeutic potential. OIP5 upregulation is observed in pRCC. The upregulation is associated with pRCC adverse features (T1P < T2P < CIMP, Stage1 + 2 < Stage 3 < Stage 4, and N0 < N1) and effectively stratifies the fatality risk. OIP5 promotes ACHN pRCC cell proliferation and xenograft formation; the latter is correlated with network alterations related to immune regulation, metabolism, and hypoxia. A set of differentially expressed genes (DEFs) was derived from ACHN OIP5 xenografts and primary pRCCs (*n* = 282) contingent to OIP5 upregulation; both DEG sets share 66 overlap genes. Overlap66 effectively predicts overall survival (*<sup>p</sup>* < 2 <sup>×</sup> <sup>10</sup>−16) and relapse (*<sup>p</sup>* < 2 <sup>×</sup> <sup>10</sup>−16) possibilities. High-risk tumors stratified by Overlap66 risk score possess an immune suppressive environment, evident by elevations in Treg cells and PD1 in CD8 T cells. Upregulation of PLK1 occurs in both xenografts and primary pRCC tumors with OIP5 elevations. PLK1 displays a synthetic lethality relationship with OIP5. PLK1 inhibitor BI2356 inhibits the growth of xenografts formed by ACHN OIP5 cells. Collectively, the OIP5 network can be explored for personalized therapies in management of pRCC patients.

**Keywords:** OIP5; papillary renal cell carcinoma; PLK1; tumorigenesis; therapy; biomarkers

## **1. Introduction**

Renal cell carcinoma (RCC) accounts for approximately 85% of kidney cancer cases. RCC can be classified as clear cell RCC (ccRCC, 75%) and non-ccRCC (nccRCC, 25%) [1].

**Citation:** Chow, M.J.; Gu, Y.; He, L.; Lin, X.; Dong, Y.; Mei, W.; Kapoor, A.; Tang, D. Prognostic and Therapeutic Potential of the OIP5 Network in Papillary Renal Cell Carcinoma. *Cancers* **2021**, *13*, 4483. https:// doi.org/10.3390/cancers13174483

Academic Editor: José I. López

Received: 1 August 2021 Accepted: 1 September 2021 Published: 6 September 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

In the latter group, papillary RCC (pRCC) constitutes 50–64% of total incidence [2,3]. Morphologically, pRCC consists of two subtypes: type 1 pRCC (T1P) and type 2 pRCC (T2P) [4]. T1P and T2P are often associated with low nuclear grade (Fuhrman 1–2) and high nuclear grade (Fuhrman 3–4) tumors respectively [4,5], providing a clinical basis for T2P tumors having poor prognosis [6–9]. Genetically, while T1P tumors typically have alterations in the *MET* gene leading to abnormal MET activation [10], T2P tumors are heterogenous and contains: (1) mutations in the *FH* (fumarate hydratase) [11], *CDKN2A*, *SETD2*, *BAP1*, and *PBRM1* genes [12], (2) CpG island methylator phenotype (CIMP), and (3) activation of the NFR2-ARE (antioxidant response element) pathway [12]. Among the T2P tumors, CIMP subtype show a particularly low possibility of overall survival [12].

While these morphological and molecular subtyping offers a primary prognostic assessment of pRCC, significant improvement is needed to enhance patient counselling and management. Effective prediction of the risk of pRCC relapse is essential in offering personalized treatments; this risk assessment is particularly important in the light that surgery remains the primary treatment for localized pRCC with a relapse rate of nearly 40% [13]. Furthermore, therapeutic options for recurrent and metastatic pRCCs are limited and non-effective, which was partly a result of treatments being extrapolated from ccRCC studies. For instance, sunitinib is a standard of care for patients with metastatic pRCC [14], despite the therapeutic benefits being low and not as effective as for ccRCC [15]. The lack of effective prognostic biomarkers and therapeutic options highlight the unmet need for a more thorough investigation of the critical factors regulating pRCC progression.

Opa interacting protein 5 (OIP5) was discovered as an Opa (*Neisseria gonorrhoeae* opacity- associated) interacting protein [16]. The protein is highly enriched in human testis (https://www.proteinatlas.org/ENSG00000104147-OIP5/tissue, accessed on 1 August 2021) [17]; its upregulation is associated with adverse clinical features in multiple cancer types, including leukemia [18], ccRCC [19], glioma [20], and the cancers of the liver [21,22], lung [23], breast [24], gastric [25,26], and bladder [27–30]. Functionally, knockdown of OIP5 was reported to attenuate the proliferation of bladder cancer cells [29], as well as colorectal and gastric cells in vitro [25]. Building on these limited studies (*n* = 17 articles in PubMed on 5 April 2021) reporting a relevance of OIP5 in oncogenic events, much more remains unanswered for OIP5-facilitated oncogenesis, particular in the context of pRCC, as OIP5 has yet to be reported in studies related to pRCC.

We provide the first comprehensive analysis of OIP5's oncogenic contributions in pRCC. OIP5 expression is significantly upregulated in pRCC; high levels of OIP5 correlate with adverse clinical characteristics of the disease, including stage, histological subtype (T2P), molecular subtype (CIMP), and lymph node metastasis. OIP5 expression robustly stratifies the risk of pRCC progression (progression-free survival) and fatality (overall survival and disease-specific survival). Functionally, OIP5 promotes pRCC cell proliferation in vitro and xenograft growth in vivo. Mechanistically, OIP5 facilitates pRCC progression along with network alterations; these changes show robust prognostic efficacies for rapid pRCC progression and fatality risk. Those of high-risk tumors display alterations in immune cell subsets including increases in the regulatory T (Treg) cell population. Treg cells are a major contributor to tumor-associated immune suppression [31]. Additionally, we identified polo-like kinase 1 (PLK1) as an OIP5-related gene in pRCC; the inhibition of PLK1 reduced OIP5-derived promotion of pRCC xenograft growth in vivo. Collectively, we report here (1) novel multigene sets derived from the OIP5 network that effectively predict the shortening of progression-free survival (PFS), overall survival (OS), and disease-specific survival (DSS) of pRCC, (2) an immune suppressive environment in pRCC tumors with OIP5 upregulation, and (3) inhibition of PLK1 as a potentially effective therapy in pRCC harboring OIP5 upregulation.

#### **2. Materials and Methods**

#### *2.1. Cell Lines, Plasmid, and Retrovirus Infection*

ACHN pRCC cell line and 786-O ccRCC cell line were purchased from ATCC and cultured in MEM and RPMI1640 respectively (Gibco, Carlsbad, CA, USA), both supplemented with 1% Penicillin-Streptomycin (Gibco, Carlsbad, CA, USA) and 10% fetal bovine serum (Life Technologies, Burlington, ON, USA). Cell lines were routinely checked for Mycoplasma contamination using a PCR kit (Abm, Cat#: G238). OIP5 cDNA plasmid was obtained from Origene (Cat: RG202255, Rockville, MD, USA) and subcloned into pBABE-puro retroviral plasmid (From Dr. Tak Mak at University of Toronto). Packing of retrovirus and the subsequent transfection were performed following our published conditions [32].

#### *2.2. Invasion and Soft Agar Assay*

Insert chambers with a control or matrigel membrane (8 µM pore size) for 24-well plates (Life sciences Corning® BioCoat™, Glendale, AZ, USA) was used for invasion assay following manufacturer's instructions. Cells (10<sup>4</sup> ) were seeded into the top chamber; serum-free medium and 10% serum medium was added to the top and bottom chamber, respectively. Cells passing through the membrane were stained with crystal violet (0.5%). Soft agar assay was performed following our published conditions [32].

#### *2.3. Colony Formation Assay and Proliferation Assay*

Growth curves were generated by seeding 10<sup>5</sup> cells/per well into 6-well tissue culture plates. Cell numbers were counted every 2 days. Colony formation assay was conducted by seeding cells in six-well plates with 100, 500, 1000 cells for ACHN, and 100, 300, 500 for 786-O. Colonies were fixed by fixation buffer (2% formaldehyde) and stained by crystal violet (0.5%) after cultured for 2 weeks. Colony numbers were counted and analyzed.

#### *2.4. Western Blot*

Cell lysates were prepared, and western blot was carried out as we have previously published [32]. Antibodies used included Anti-Flag M2 (1:1500, Sigma-Aldrich, Oakville, ON, Canada) and Anti-OIP5 (1:500, Sigma-Aldrich, Oakville, ON, Canada).

#### *2.5. Immunohistochemistry (IHC)*

Kidney cancer TMA (KD29602) was purchased from US Biomax (Dervood, MD, USA). Slide was baked at 60 ◦C for 1 h, then de-paraffinized in 100% xylene and 70% EtOH series. Antigen retrieval buffer was prepared with sodium citrate buffer (PH = 6) in the steamer for 20 min. OIP5 (1:50, Sigma-Aldrich, Oakville, ON, Canada) antibodies were incubated at 4 ◦C overnight. Secondary anti-rabbit antibodies (Vector Laboratories, 1:200), VECTASTAIN ABC and DAB solution (Vector Laboratories) were subsequently added to the slides and incubated following our IHC protocol. Washes were performed by 1× PBS and distilled water. Slides were counterstained by haematoxylin (Sigma Aldrich, Oakville, ON, Canada) and image analysis was conducted with ImageScope software (Leica Microsystems Inc., Richmond Hill, ON, Canada). Staining intensity scores were calculated into HScore by the formula [HScore = (%Positive) × (Intensity) + 1]. Statistical analysis was performed by student *t*-test, and *p* < 0.05 was considered statistically significant.

Xenograft tumors were paraffin embedded and cut serially by microtome. OIP5 (1:50, Sigma-Aldrich), Anti-Phospho-Histone H3 (Ser 10) (1:200, Upstate Biotechnology Inc., Lake Placid, NY, USA), CDK2 (1:200, Santa Cruz, Dallas, TX, USA), and PLK1 (1:300, Novus Biologicals, Toronto, ON, Canada) antibodies were used in the analyses for the xenograft tumors.

#### *2.6. Xenograft Tumor Formation and Treatment with PLK1 Inhibitor*

ACHN OIP5 and ACHN EV were suspended in 0.1 mL MEM/Matrigel (BD) mixture with 1:1 volume and implanted subcutaneously into the left flank of 8-week-old non-

obese diabetic/severe combined immunodeficiency (NOD/SCID) male mice (The Jackson Laboratory). The mice were monitored post-injection of cancer cells through observation and palpation. The size of the tumors was measured every two days by caliper. Tumor volume was calculated based on the formula V = L × W2 × 0.52. BI2536 PLK1 inhibitor (Selleckchem, Burlington, ON, Canada) was dissolved in 0.1 N HCl and diluted by 0.9% NaCl. Diluted BI2536 or 0.9% NaCl (negative control) was injected to mice intravenously via tail vein with a dosage of 50 mg/kg. The mice were euthanized when the tumor volume reached 1000 mm<sup>3</sup> . The xenograft tumor, together with all the major organs, were photographed and collected. All tumors were cut in half, with one half fixed with 10% formalin (VWR, Mississauga, ON, Canada) and the other half stored in −80 ◦C. The formalin-fixed tissue was processed by department of Histology (St. Joseph's Health care, Hamilton, ON, Canada) and embedded in paraffin. All the animal works were performed according to the protocols approved by McMaster University Animal Research Ethics Board (16-06-23).

#### *2.7. RNA Sequencing Analysis*

RNA sequencing analysis was carried out following our established conditions [33]. RNA was extracted from ACHN EV (*n* = 3) and ACHN OIP5 (*n* = 3) xenografts using a miRNeasy Mini Kit (Qiagen, No. 217004) according to the manufacturer's instructions. RNA-seq libraries were generated with TruSeq Ribo Profile Mammalian Kit (Illumina, RPHMR12126) according to manufacturer's instruction. These libraries were sequenced in a paired end setting by Harvard Bauer Core Facility using Nextseq 500/550. RNA-seq reads were processed and analyzed using Galaxy (https://usegalaxy.org/, accessed on 31 May 2020). Specifically, low quality reads and adaptor sequences (AGATCGGAAGAG-CACACGTCTGAACTCCAGTCA: forward strand and AGATCGGAAGAGCGTCGTG-TAGGGAAAGAGTGT: reverse strand) were first removed. Alignment and read counts were performed using HISAT2 and Featurecounts respectively. Differential gene expression was determined using DESeq2. KEGG analysis and GSEA (Gene Set Enrichment Analysis) were also performed using Galaxy; the FGSEA (fast preranked GSEA) was used for GSEA analysis. Enrichment analyses were carried out using Metascape (https://metascape.org/gp/index.html#/main/step1, accessed on 1 September 2020) [34].

#### *2.8. RNA Sequencing Analysis*

Cox proportional hazards (Cox PH) regression analyses were performed using the R survival package. The PH assumption was tested. Cutoff points were estimated using Maximally Selected Rank Statistics (the Maxstat package, https://cran.r-project.org/web/ packages/maxstat/maxstat.pdf, accessed on 8 August 2020). The TCGA PanCancer Atlas pRCC dataset available from cBioPortal [35,36] was used.

#### *2.9. Examination of Gene Expression*

Gene expressions were determined using the UALCAN platform (ualcan.path.uab. edu/home, accessed on 31 March 2021) [37].

#### *2.10. Statistical Analysis*

Kaplan-Meier survival analyses and logrank test were conducted by R *Survival* package and tools provided by cBioPortal. Cox regression analyses were performed using R *survival* package. Time-dependent receiver operating characteristic (tROC) analyses were carried out with R *timeROC* package. ROC and precision-recall (PR) profiles were constructed using the PRROC package in R. Two-tailed Student *t*-test, one-way ANOVA, and two-way ANOVA were performed for statistical analysis of two and more than two groups respectively, with *p* < 0.05 to be considered statistically significant. Tukey's test was performed for post-hoc analysis. Statistical analysis was conducted by GraphPad Prism 7 and data were presented as mean ± SEM/SD. A value of *p* < 0.05 was considered statistically significant.

#### **3. Results**

#### *3.1. Association of OIP5 Upregulation with pRCC Tumorigenesis and Progression*

OIP5 was reported to be a component gene in a multigene set predicting the risk of prostate cancer recurrence [38]; its upregulation associates with adverse features in ccRCC and bladder cancer [19,29], supporting a general involvement of OIP5 in urogenital cancers. To investigate this possibility, we examined OIP5 expression in pRCC using a tissue microarray (TMA) containing 40 pairs of pRCC and 74 pairs of ccRCC tumors with the adjacent non-tumor kidney (AJK) tissues from 20 and 37 patients, respectively. The pRCC patient population (*n* = 20) consists of 11 men and 9 women with most tumors being at T1 stage (Table 1). In comparison to AJK tissues, pRCC tumor tissues expressed a significant OIP5 upregulation (Figure 1A,B). OIP5 expression was further increased in advanced T stage tumors (Figure 1B). Consistent with a previous report, OIP5 upregulation occurred in Grade 2–3 ccRCC tumors compared to the AJK tissues; nonetheless, we could not demonstrate OIP5 upregulation in Grade 1 ccRCC compared to the AJK tissues (Figure S1), suggesting a role of OIP5 in ccRCC progression. By using the TCGA RNA-sequencing data organized by UALCAN (ualcan.path.uab.edu/home, 31 March 2021) [37], OIP5 upregulation at the mRNA level in pRCC tissues was observed (Figure 1c); the upregulations reflects the level of severity and the order of unfavorable outcome of pRCC with higher expression levels in T2P over T1P tumors, CIMP tumors over other subtypes (Figure 1D), stage 3 tumors over stages 1–2 tumors, stage 4 over stage 3 tumors (Figure 1E), and N1 (lymph node metastasis) over N0 tumors (Figure 1F). Consistent with its associations with adverse tumor features, OIP5 expression robustly stratifies pRCC tumors into a high- and low-risk group based on overall survival possibility (Figure 1G). Among the 10 patients in the OIP5-high group, seven died in a rapid time course (Figure 1G). Collectively, these observations support a strong association of OIP5 with pRCC tumorigenesis and progression.

**Table 1.** The clinical parameters of pRCC patients included in TMA.


Age: median (Q1/quartile 1–Q3) *n*: number of cases. All patients were without lymph node metastasis (N0) and distant metastasis (M0).

#### *3.2. OIP5-Mediated Enhancement of pRCC Tumorigenesis along with Network Alterations*

Attributed to the uncommon status of pRCC, there are only limited number of confirmed pRCC cell lines available. ACHN is the most widely used and confirmed metastatic pRCC cell line; the cells have the typical feature of *c-MET* polymorphism detected in pRCC [39,40]. ACHN is likely the only confirmed metastatic pRCC cell line [39]. To analyze the functional impact of OIP5 on pRCC tumorigenesis, we stably expressed OIP5 in ACHN cells (Figure 2A). In comparison to ACHN EV (empty vector) cells, ACHN OIP5 cells displayed elevated abilities for proliferation (Figure 2B), colony formation (Figure 2C; Figure S2A), invasion (Figure 2D; Figure S2B), and growth in soft agar (Figure 2E; Figure S2C). We have also established the EV and OIP5 stable lines in the commonly used 786-O ccRCC cells, and OIP5 overexpression did not affect all of the above oncogenic events observed in ACHN cells in vitro (data not shown), which suggests a certain level of specificity of OIP5 in promoting pRCC. In vivo, OIP5 enhanced the growth of ACHN cell-produced xenografts compared to tumors produced by ACHN EV cells (Figure 2F); mice bearing ACHN OIP5 tumors reached endpoint faster compared to animals with ACHN EV cell-produced tumors (Figure 2G). The overexpression of OIP5 in ACHN OIP5 tumors was confirmed (Figure S3A). The ACHN OIP5 tumors show a significant increase of CDK2 expression largely in the nuclei (Figure S3B); the functions of this are not clear as no upregulations of the relevant cyclins (cyclin A and cyclin E) was observed (data not shown).

was 15.48 months.

**Figure 1.** Upregulation of OIP5 associates with adverse features of pRCC and predicts poor overall survival. (**A**) IHC staining for OIP5 was performed using a RCC TMA; typical images of OIP5 staining in the adjacent kidney (AJK) and pRCC tumor tissues are presented. (**B**) Quantification of OIP5 IHC staining by H-score in the indicated tissues; means ± standard deviations (SDs) are graphed. Statistical analyses were performed using 2-tailed Student's *t*-test; \*\*\*: *p* < 0.001 compared to the respective AJK tissues, \$\$\$: *p* < 0.001 compared to T1 tumors. (**C**–**F**) OIP5 mRNA expressions in the indicated setting were analyzed using the TCGA dataset organized by UALCAN [37]. Student's *t*-test (**C**) and other indicated paired statistics were provided by UALCAN. \*: *p* < 0.05, \*\*: *p* < 0.01, \*\*\*: *p* < 0.001 compared to normal kidney tissues; \$: *p* < 0.05, \$\$: *p* < 0.01 compared to T1P (**D**), Stage 2 (**E**), and N0 (**F**); ##: *p* < 0.01, ###: *p* < 0.001 compared to T2P (**D**) and Stage 3 (**E**). (**G**) Survival analysis was performed using the TCGA Pancancer pRCC dataset within cBioPortal. Logrank test was performed. Cutoff point used to separate the high- and low-OIP5 expression groups was ≥ 2 z-score or 2SD. The graph was produced using tools provided by cBioPortal. The median months overall survival for patients in the high-OIP5 group **Figure 1.** Upregulation of OIP5 associates with adverse features of pRCC and predicts poor overall survival. (**A**) IHC staining for OIP5 was performed using a RCC TMA; typical images of OIP5 staining in the adjacent kidney (AJK) and pRCC tumor tissues are presented. (**B**) Quantification of OIP5 IHC staining by H-score in the indicated tissues; means ± standard deviations (SDs) are graphed. Statistical analyses were performed using 2-tailed Student's *t*-test; \*\*\*: *p* < 0.001 compared to the respective AJK tissues, \$\$\$: *p* < 0.001 compared to T1 tumors. (**C**–**F**) OIP5 mRNA expressions in the indicated setting were analyzed using the TCGA dataset organized by UALCAN [37]. Student's *t*-test (**C**) and other indicated paired statistics were provided by UALCAN. \*: *p* < 0.05, \*\*: *p* < 0.01, \*\*\*: *p* < 0.001 compared to normal kidney tissues; \$: *p* < 0.05, \$\$: *p* < 0.01 compared to T1P (**D**), Stage 2 (**E**), and N0 (**F**); ##: *p* < 0.01, ###: *p* < 0.001 compared to T2P (**D**) and Stage 3 (**E**). (**G**) Survival analysis was performed using the TCGA Pancancer pRCC dataset within cBioPortal. Logrank test was performed. Cutoff point used to separate the high- and low-OIP5 expression groups was ≥2 z-score or 2SD. The graph was produced using tools provided by cBioPortal. The median months overall survival for patients in the high-OIP5 group was 15.48 months.

*3.2. OIP5-Mediated Enhancement of pRCC Tumorigenesis along with Network Alterations*

Attributed to the uncommon status of pRCC, there are only limited number of con-

pRCC cell line; the cells have the typical feature of *c-MET* polymorphism detected in pRCC [39,40]. ACHN is likely the only confirmed metastatic pRCC cell line [39]. To analyze the functional impact of OIP5 on pRCC tumorigenesis, we stably expressed OIP5 in ACHN cells (Figure 2A). In comparison to ACHN EV (empty vector) cells, ACHN OIP5

shown).

cells displayed elevated abilities for proliferation (Figure 2B), colony formation (Figure 2C; Figure S2A), invasion (Figure 2D; Figure S2B), and growth in soft agar (Figure 2E; Figure S2C). We have also established the EV and OIP5 stable lines in the commonly used 786-O ccRCC cells, and OIP5 overexpression did not affect all of the above oncogenic events observed in ACHN cells in vitro (data not shown), which suggests a certain level of specificity of OIP5 in promoting pRCC. In vivo, OIP5 enhanced the growth of ACHN cell-produced xenografts compared to tumors produced by ACHN EV cells (Figure 2F); mice bearing ACHN OIP5 tumors reached endpoint faster compared to animals with ACHN EV cell-produced tumors (Figure 2G). The overexpression of OIP5 in ACHN OIP5 tumors was confirmed (Figure S3A). The ACHN OIP5 tumors show a significant increase of CDK2 expression largely in the nuclei (Figure S3B); the functions of this are not clear as no upregulations of the relevant cyclins (cyclin A and cyclin E) was observed (data not

**Figure 2.** OIP5 promotes oncogenic processes of ACHN cells in vitro and in vivo. (**A**) ACHN empty vector (EV) and OIP5 stable lines. Western blot was carried out using anti-OIP5 and Actin antibodies. OIP5 expression was normalized to Actin and presented at fold changes to OIP5 expression in EV cells. (**B**) ACHN EV and ACHN OIP5 cells were seeded in 6-well plate at 105 cell/well; cell numbers were recorded at the indicated days. Experiments were repeated three times; means ± SDs are graphed. Statistical analysis was performed using 2-way ANOVA. \*\*\*: *p* < 0.001 between the two curves. (**C**) The indicated cells were seeded at the indicated number in 6-well plates. Colonies were formed following 2 weeks culture. Experiments were repeated three times; means ± SDs are graphed. \*\*: *p* < 0.01 compared to the respective EV by Student's *t*-test (2-way). (**D**,**E**) Invasion and soft agar assays were repeated 3 times; means ± SDs are graphed. \*\*: *p* < **Figure 2.** OIP5 promotes oncogenic processes of ACHN cells in vitro and in vivo. (**A**) ACHN empty vector (EV) and OIP5 stable lines. Western blot was carried out using anti-OIP5 and Actin antibodies. OIP5 expression was normalized to Actin and presented at fold changes to OIP5 expression in EV cells. (**B**) ACHN EV and ACHN OIP5 cells were seeded in 6-well plate at 10<sup>5</sup> cell/well; cell numbers were recorded at the indicated days. Experiments were repeated three times; means ± SDs are graphed. Statistical analysis was performed using 2-way ANOVA. \*\*\*: *p* < 0.001 between the two curves. (**C**) The indicated cells were seeded at the indicated number in 6-well plates. Colonies were formed following 2 weeks culture. Experiments were repeated three times; means ± SDs are graphed. \*\*: *p* < 0.01 compared to the respective EV by Student's *t*-test (2-way). (**D**,**E**) Invasion and soft agar assays were repeated 3 times; means ± SDs are graphed. \*\*: *p* < 0.01, \*\*\*: *p* < 0.001 compared to the respective EV control by Student's *t*-test (2-way). (**F**,**G**) Xenografts were produced in NOS/SCID mice (5 mice per group) using ACHN EV cells and ACHN OIP5 cells. Means ± SEM (standard error of the mean) are graphed; \*\*\*: *p* < 0.001 between the two curved by two-way ANOVA (**F**). Kaplan-Meier curve; statistical analysis was performed using logrank test (**G**).

To further analyze factors and networks utilized by OIP5 in enhancing ACHN cellproduced xenografts, RNA-sequencing (RNA-seq) was performed on ACHN EV and ACHN OIP5 tumors at three per group. Gene set enrichment analysis (GSEA) was conducted on differentially expressed genes obtained in the setting of OIP5 vs. EV. When enrichment in the oncogenic gene sets (C6) collection was analyzed using FGSEA (fast gene set enrichment analysis), we observed that genes downregulated (DN) in cells with activation (UP) of ERB2, MEK, and mTOR were also downregulated in ACHN OIP5 tumors compared to ACHN EV tumors (Figure 3A), suggesting OIP5 suppressing those genes that are downregulated by ERB2, MEK, and mTOR. Similarly, ACHN OIP5 tumors also display

downregulation of EGFR-downregulated genes (Table S1). The serine/threonine kinase 33 (STK33) is a synthetic lethal interacting protein of KRAS mutant, i.e., cells expressing KRAS mutant rely on STK33 for survival [41]. Knockdown of STK33 in acute myeloid leukemia cells led to upregulation of a set of genes (STK33-UP) [41], suggesting a potential inhibition of these genes by STK33. These gene expressions were also reduced in ACHN OIP5 tumors (Figure 3A; Table S1). To test the reliability of the enrichment obtained by FGSEA, GSEA was further conducted using a more stringent platform: EGSEA. Ensemble gene set enrichment analysis produces a consensus gene set ranking (enrichment) with the combination of multiple (up to *n* = 12) algorithms [42]. With the maximal stringent condition using all 12 algorithms, EGSEA revealed within the top 12 ranks the downregulation of the ERB2- and MEK-suppressed gene sets in ACHN OIP5 tumors (Figure S4); downregulation of genes in cells with STK33 knockdown was observed in multiple setting (Figure S4) which is consistent with the enrichments derived from using FGSEA (Table S1). All top 12 ranked gene sets obtained by EGSA (Figure S4) are also included in those produced by FGSEA (Table S1). It is intriguing that VEGFA-suppressed genes in HUVEC (human umbilical vein endothelial cell) cells were also downregulated in ACHN OIP5 tumors (Figure S4; Table S1). Based on the overall gene set enrichment within the oncogenic gene set (C6, MSigDB) collection (Table S1), we can summarize that in ACHN OIP5 xenografts, the RB pathway is inhibited and the signaling processes of STK33, BMI1, EZH2, MYC, WNT, VEGFA, and EGFR/ERB2 are enhanced (Figure 3B).

We further examined gene set enrichment within the Hallmark gene set collection using FGSEA. The analyses revealed downregulations of inflammatory response, TNFα\_via\_ NFκB signaling (NFκB-regulated genes in response to TNFα), and complement gene expression (Hallmark\_Component, normalized enrichment score/NES: −1.48, padj 0.013) (Figure S5A; Table S2). Additionally, ACHN OIP5 xenografts exhibited upregulations in gene sets regulating fatty acid metabolism and cholesterol homeostasis (Figure 3C; Table S2). These enrichments were also produced by EGSEA (Figure S5B,C). Several processes are enhanced in ACHN OIP5 tumors, which include oxidative phosphorylation, the expression of E2F and MYC targets, EMT (epithelial mesenchymal transition), mTORC1 signaling, and adipogenesis (Table S2). Enrichment in glycolysis in ACHN OIP5 tumors was obtained by FGSEA (Table S2), which was also confirmed by KEGG pathway analysis using EGSEA (Figure S6). Evidence thus suggests a metabolic switch to Warburg metabolism in ACHN OIP5 tumors.

#### *3.3. Association of OIP5-Related Differentially Expressed Genes with CIMP Subtype*

In comparison to other pRCC subtypes, CIMP tumors have a Warburg metabolic shift [12], indicating an association between OIP5-affected genes and CIMP. This notion is supported by the elevation of OIP5 expression in CIPM pRCC tumors (Figure 1D). To investigate this possibility, we firstly defined the differentially expressed genes (DEGs) in ACHN OIP5 tumors vs. ACHN EV tumors as those with *p*.adj < 0.05 and fold change > |1.5|; a total of 1128 DEGs were derived (Table S3). In these DEGs, the top upregulated genes include WNT7A, FGF1, CNTN1 [43], SOX2, and others, which are known for their facilitative roles in tumorigenesis. The top 20 clusters enriched in these DEGs contain those that regulate urogenital system development, blood vessel morphogenesis, hippo pathway, cell surface receptor signaling, pathway in cancer, epithelial cell proliferation, and others (Figure 4A; Table S4). Individual terms in these enriched clusters form a network connection (Figure S7A). These pathways are clearly relevant to tumorigenesis. DEGs are clustered in ACHN OIP5 tumors vs. ACHN EV tumors (Figure 4B).

sets affected).

**Figure 3.** OIP5 induces network alterations during pRCC tumorigenesis. (**A**) GSEA (gene set enrichment analysis) on differentially expressed genes derived from the comparison of ACHN OIP5 tumors to ACHN EV tumors was performed with FGSEA within the Galaxy platform. The MSigDB oncogenic gene sets (C6) collection was used. (**B**) Summary of the major oncogenic gene sets affected in ACHN OIP5 tumors (see Table S1 for individual gene sets affected). (**C**) Enrichment of the indicated gene set within the MSigDB hallmark gene sets collection (see Table S2 for individual gene **Figure 3.** OIP5 induces network alterations during pRCC tumorigenesis. (**A**) GSEA (gene set enrichment analysis) on differentially expressed genes derived from the comparison of ACHN OIP5 tumors to ACHN EV tumors was performed with FGSEA within the Galaxy platform. The MSigDB oncogenic gene sets (C6) collection was used. (**B**) Summary of the major oncogenic gene sets affected in ACHN OIP5 tumors (see Table S1 for individual gene sets affected). (**C**) Enrichment of the indicated gene set within the MSigDB hallmark gene sets collection (see Table S2 for individual gene sets affected).

*3.3. Association of OIP5-Related Differentially Expressed Genes with CIMP Subtype*

In comparison to other pRCC subtypes, CIMP tumors have a Warburg metabolic shift [12], indicating an association between OIP5-affected genes and CIMP. This notion is supported by the elevation of OIP5 expression in CIPM pRCC tumors (Figure 1D). To investigate this possibility, we firstly defined the differentially expressed genes (DEGs) in ACHN OIP5 tumors vs. ACHN EV tumors as those with p.adj < 0.05 and fold change >


**Table 2.** Characterization of Overlap66 genes.


**Table 2.** *Cont.*

1: prediction of overall survival determined by univariate Cox analysis; +, −, and N: gene expression positively, negatively, and not predicting OS, respectively. NS: not significant. 2: expression status in CIMP tumors, "Up": upregulation compared to T2P, "Down": downregulation compared to T1P, 2a: in comparison to T2P as the comparison to T1P being not significant, N: no changes. 3: tumor (*n* = 290) in comparison to normal tissues (*n* = 30). 4: these genes are in Overlap21. 5: these genes are in Overlap21plus. Expression analysis in "CIMP" and "Tumor" using the TCGA data (UALCAN). \*: *p* < 0.5, \*\*: *p* < 0.01, \*\*\*: *p* < 0.001.

> To confirm the relevance of these DEGs derived from ACHN cell-produced xenografts in pRCC pathogenesis, we analyzed their relationship to DEGs derived from primary pRCCs relative to OIP5 expression. In the TCGA Pancancer pRCC dataset within cBioPortal, high OIP5 expression robustly separates pRCC tumors into a high and low risk group based on their overall survival (OS) possibilities (Figure 1G). From these two groups, we obtained 873 DEGs defined by *q* < 0.05 and fold change ≥ |2| (Table S5). These primary pRCC-derived DEGs share 66 overlap DEGs (Overlap66) with the xenograft-derived DEGs (Table 2; Figure 4C). The alterations in their expressions in normal kidney tissues (*n* = 30) and pRCC tumors (*n* = 290) at different stages are presented in Figure S8. The genes with further elevations in Stage 3–4 tumors include SLC7A11, PCSK5, STC2, PLK1, TK1, TRIB3, and SRXN1 (Figure S8).

**Figure 4.** Pathway enrichment of OIP5 DEGs. DEGs were first defined as p.adj < 0.05 and fold changes > |1.5| in the comparison of ACHN OIP5 tumors (*n* = 3) vs. ACHN EV tumors (*n* = 3). (**A**) Pathway enrichment in these DEGs (Table S3) was then performed using the Metascape [34] platform. (**B**) Clustering of DEGs in ACHN OIP5 tumors and ACHN EV **Figure 4.** Pathway enrichment of OIP5 DEGs. DEGs were first defined as *p*.adj < 0.05 and fold changes > |1.5| in the comparison of ACHN OIP5 tumors (*n* = 3) vs. ACHN EV tumors (*n* = 3). (**A**) Pathway enrichment in these DEGs (Table S3) was then performed using the Metascape [34] platform. (**B**) Clustering of DEGs in ACHN OIP5 tumors and ACHN EV tumors. (**C**) The number of overlapping genes between primary (patient) pRCC-derived DEGs and DEGs obtained from xenografts at fold change > |1.5|. (**D**,**E**) The indicated DEGs were analyzed for expression in the histological subtypes of pRCC using the UALCAN platform [37]. DEGs positively (**D**) and negatively (**E**) predict shortening of OS (see Table 2 for details). \*\*: *p* < 0.01, \*\*\*: *p* < 0.001 in comparison to normal kidney tissues.

Among these 66 DEGs, 8 and 41 genes are not known for associations with cancer and ccRCC respectively (Table S6A); only PLK1 was reported to be a component gene in a prognostic multigene of pRCC (Table S6A). Overlap66 is novel to pRCC. Forty-six out of 66 DEGs significantly predict overall survival (OS) possibility with some being individually efficacious based on their *<sup>p</sup>* values: 6.73 <sup>×</sup> <sup>10</sup>−<sup>10</sup> for PCSK5, 1.3 <sup>×</sup> <sup>10</sup>−<sup>10</sup> for C116orf75 (RMI2), 1.84 <sup>×</sup> <sup>10</sup>−<sup>13</sup> for ATAD2, and others (Table S6A). Furthermore, 33 DEGs retain their predictive significance after adjusting for age at diagnosis, sex, and T stage (Table S6A).

The potentials of the 33 DEGs as prognostic biomarkers are in accordance with their expression status in CIMP. C116orf75, SRXN1, TK1, and TRIB3 positively predict poor OS (Table 2; Table S6A); they are notably upregulated in CIMP tumors (Figure 4D). In reverse, CCDC106, CX3CL1, LYNX1, and SPATA18 are negatively associated with poor OS (Table 2; Table S6A); their expressions are particularly downregulated in CIMP tumors (Figure 4E). In all 46 genes with their expression associated with OS shortening, 11 show no alterations in gene expression in CIMP tumors (Table 2); for the remaining 35 genes, their positive and negative predictions of OS shortening correlate with their respective upregulation and downregulation in CIMP tumors (Table 2). This correlation of expression was not observed in tumors vs. non-tumor tissues (Table 2). In view of CIMP tumors having the poorest OS possibility [12], the association of these gene expression with CIMP tumors supports their potential as prognostic biomarkers.

#### *3.4. Robust Prognostic Biomarker Potential of Overlap66 and Its Sub-Multigene Panels*

Following the above analyses, we examined the OS-related prognostic potential of Overlap66 as a multigene panel. The expression data for these DEGs along with the relevant clinical data were retrieved from the Pancancer pRCC dataset within cBioPortal. Risk scores for individual tumors were calculated as ∑(coef<sup>i</sup> × Geneiexp)<sup>n</sup> (coef<sup>i</sup> : Cox coefficient of gene<sup>i</sup> , Geneiexp: expression of Gene<sup>i</sup> , *n* = 66). Coefs were obtained using the multivariate Cox model. Overlap66 risk scores efficiently predict OS shortening using both univariate (UV) and multivariate (MV) Cox models (Figure S7B). The MV model consists of the risk scores, age at diagnosis, sex, and T stage (Figure S7B). With the cutoff point optimized using the Maximally Selected Rank Statistics (Figure S9), Overlap66 effectively stratifies the risk of fatality (possibility of OS) and relapse (progression-free survival/PFS) (Figure 5A,B). The discriminations of OS and PFS are with time-dependent area under the curve (tAUC) value of 94.6–91.3% in the time frame of 12.4 month (M) to 57.2 M (Figure 5A) and 93.7–86.7% for 10.8 M to 50.6 M (Figure 5B), respectively. Collective evidence supports Overlap66 being a novel and robust prognostic multigene panel for pRCC.

We further validated Overlap66 risk score in stratification of pRCC fatality risk using a recently developed R package: contpointr (https://github.com/thie1e/cutpointr, accessed on 1 May 2021). An optimal cutoff point was obtained with Kernel smoothing model coupled with 1000 bootstrapping. This cutoff point classifies pRCC fatality risk at 0.78 sensitivity and 0.84 specificity or the sum of sensitivity and specificity (sum\_sens\_spec) value of 1.62 (Figure 6A). Risk stratifications of out-of-bag bootstrap samples (*n* = 1000) occurred most frequently at sum\_sens\_spec 1.6 (Figure 6B), which closely approximates sum\_sens\_spec 1.62 associated with the optimal cutoff point on the full cohort (Figure 6A). The fatality risk stratifications of the in-bag samples (*n* = 1000, average 63.2% of full samples) and the out-of-bag samples (*n* = 1000) were at the median sum\_sens\_spec values of 1.62 and 1.60 respectively. Taken together, these bootstrap analyses reveal a good outof-sample performance of Overlap66 in classification of pRCC fatality risk, supporting Overlap66's application in real world. This potential is strengthened by the effectiveness of the risk classification with a range of cutoff points (Figure 6B,C).

**Figure 5.** Stratification of the possibilities of overall survival (OS) and progression-free survival (PFS) by Overlap66 and Overlap21. (**A**,**B**) Cutoff points were determined by Maximally Selected Rank Statistics for the risk scores of Overlap66 (see Figure S9) and Overlap21. Kaplan Meier curves for OS (**A**) and PFS (**B**) are constructed, using the R survival package, with the populations at risk in the indicated follow-up period included. Statistical analyses were performed using logrank test. The median months of OS and PFS are indicated. Time-dependent ROC (receiver operating characteristic; tROC) curves were generated using the R *timeROC* package; time-dependent area under the curve (AUC) values for the indicated multigene sets are shown. (**C**,**D**) ROC and precision-recall (PR) curves for Overlap66 and Overlap21 in predicting OS and PFS possibilities were produced using the R PRROC. **Figure 5.** Stratification of the possibilities of overall survival (OS) and progression-free survival (PFS) by Overlap66 and Overlap21. (**A**,**B**) Cutoff points were determined by Maximally Selected Rank Statistics for the risk scores of Overlap66 (see Figure S9) and Overlap21. Kaplan Meier curves for OS (**A**) and PFS (**B**) are constructed, using the R survival package, with the populations at risk in the indicated follow-up period included. Statistical analyses were performed using logrank test. The median months of OS and PFS are indicated. Time-dependent ROC (receiver operating characteristic; tROC) curves were generated using the R *timeROC* package; time-dependent area under the curve (AUC) values for the indicated multigene sets are shown. (**C**,**D**) ROC and precision-recall (PR) curves for Overlap66 and Overlap21 in predicting OS and PFS possibilities were produced using the R PRROC.

We further validated Overlap66 risk score in stratification of pRCC fatality risk using a recently developed R package: contpointr (https://github.com/thie1e/cutpointr, 1 May 2021). An optimal cutoff point was obtained with Kernel smoothing model coupled with 1000 bootstrapping. This cutoff point classifies pRCC fatality risk at 0.78 sensitivity and 0.84 specificity or the sum of sensitivity and specificity (sum\_sens\_spec) value of 1.62 (Figure 6A). Risk stratifications of out-of-bag bootstrap samples (*n* = 1000) occurred most frequently at sum\_sens\_spec 1.6 (Figure 6B), which closely approximates sum\_sens\_spec

1.62 associated with the optimal cutoff point on the full cohort (Figure 6A). The fatality risk stratifications of the in-bag samples (*n* = 1000, average 63.2% of full samples) and the out-of-bag samples (*n* = 1000) were at the median sum\_sens\_spec values of 1.62 and 1.60 respectively. Taken together, these bootstrap analyses reveal a good out-of-sample performance of Overlap66 in classification of pRCC fatality risk, supporting Overlap66's ap-

sification with a range of cutoff points (Figure 6B,C).

**Figure 6.** Validation of Overlap66 risk score in stratification of pRCC fatality risk. Cutoff points were estimated using Kernel smoothing method coupled with bootstrapping (*n* = 1000). The average inbag and out-of-bag (OOB) bootstrap samples are 63.2% and 36.8% of the full sample size respectively. The analysis was performed using the cutpointr R package (https://github.com/thie1e/cutpointr, 21 July 2021). (**A**) ROC curve with the optimal cutoff point indicated (arrow); sens: sensitivity, spec: specificity, and the sum\_sens\_spec: 1.62. (**B**) Distribution of out-of-bag (OOB) metric values. The most predictions occur in these OOB samples (*n* = 1000) at the sum\_sens\_sepc value 1.6. The region marked by the 2 dotted lines includes a range of sum\_sens\_sepc values that frequently stratify the fatality risk with high accuracy. The red dot represents a sum\_sens\_sepc value 1.55. (**C**) Classification of pRCC tumors into a high- and low-risk group using two indicated cutpoints; the **Figure 6.** Validation of Overlap66 risk score in stratification of pRCC fatality risk. Cutoff points were estimated using Kernel smoothing method coupled with bootstrapping (*n* = 1000). The average in-bag and out-of-bag (OOB) bootstrap samples are 63.2% and 36.8% of the full sample size respectively. The analysis was performed using the cutpointr R package (https://github.com/thie1e/cutpointr, accessed on 21 July 2021). (**A**) ROC curve with the optimal cutoff point indicated (arrow); sens: sensitivity, spec: specificity, and the sum\_sens\_spec: 1.62. (**B**) Distribution of out-of-bag (OOB) metric values. The most predictions occur in these OOB samples (*n* = 1000) at the sum\_sens\_sepc value 1.6. The region marked by the 2 dotted lines includes a range of sum\_sens\_sepc values that frequently stratify the fatality risk with high accuracy. The red dot represents a sum\_sens\_sepc value 1.55. (**C**) Classification of pRCC tumors into a high- and low-risk group using two indicated cutpoints; the sum\_sens\_sepc 1.62 cutoff point was obtained using Kernel method and the sum\_sens\_sepc 1.55 cutoff point (see the red dot in panel (**B**)) was derived using Maximally selected LogRank statistics (see Figure S9). The *p* value is for both separations.

sum\_sens\_sepc 1.62 cutoff point was obtained using Kernel method and the sum\_sens\_sepc 1.55 cutoff point (see the red dot in panel B) was derived using Maximally selected LogRank statistics (see Figure S9). The *p* value is for both separations. We subsequently optimized Overlap66. As OIP5 expression was at 1.9 folds in ACHN OIP5 tumors compared to ACHN EV tumors (Table S3), we defined a subgroup of DEGs as those with p.adj < 0.05 and fold ≥ |1.9| in ACHN OIP5 tumors compared to We subsequently optimized Overlap66. As OIP5 expression was at 1.9 folds in ACHN OIP5 tumors compared to ACHN EV tumors (Table S3), we defined a subgroup of DEGs as those with *p*.adj < 0.05 and fold ≥ |1.9| in ACHN OIP5 tumors compared to ACHN EV tumors. These DEGs (*n* = 298) share 21 overlap genes (Overlap21) with primary pRCCderived DEGs (Figure S7C, Table S6B). As expected, Overlap21 is a subgroup of Overlap66 (Table 2). Overlap21 risk scores predict OS possibility under both UV and MV Cox models with comparable efficiency as Overlap66, evident by Hazard ratio (HR) and 95% confident

pRCC-derived DEGs (Figure S7C, Table S6B). As expected, Overlap21 is a subgroup of

interval (CI) (Figure S7B). Similar prediction efficiencies for PFS between Overlap21 and Overlap66 were also observed (Figure S7B). Overlap21 effectively stratifies the risk of mortality and PFS; the discriminations possess high tAUC values (Figure 5A,B). In comparison, Overlap21 seems marginally less effective compared to Overlap66 in the discriminations of OS and PFS (Figure 5A,B). Nonetheless, the Overlap21-mediated predictions are clearly effective. Similar to Overlap66, Overlap 21 risk score is an independent predictor of poor OS after adjusting age at diagnosis, sex, and T stage (Table 3).


**Table 3.** Univariate and multivariate Cox analysis of Overlap66 and Overlap21 risk scores for pRCC OS.

Analyses were performed using the TCGA PanCancer pRCC dataset. Age: at diagnosis. Sex: male vs. female. Tstage 1: T stage 3 + 4 vs. Tstage 0: T stage 1 + 2. i and ii: in analysis with Overlap66 (i) and Overlap21 (ii). \*, \*\*, and \*\*\*: *p* < 0.05, *p* < 0.01, and *p* < 0.001 respectively

> The utility of Overlap21 in assessing pRCC fatality risk is further illustrated by its impressive separation of disease-specific survival (DSS) risk (Figure 7A,C). DSS is more specific compared to OS in addressing factors contributing to cancer-caused deaths. Overlap66 did not perform well in DSS estimation (data not shown), which might be attributable to the small number of events (disease-specific death *n* = 27) in the context of the large number of variables (*n* = 66 in Overlap66). We thus generated Overlap21plus by using Overlap21 as the basis, and the rest of DEGs within Overlap66 were added if they remain risk factors for decreased OS after adjusting age at diagnosis, sex, and T stages (Table S6A). However, Overlap21plus was not superior to Overlap21 in the estimation of OS and PFS (data not shown). Nonetheless, the risk score of Overlap21plus predicts DSS risk in a comparable efficiency as Overlap21 (Figure S7B); its ability to classify DSS possibility was marginally superior to Overlap21 (Figure 7A–C).

> Instead of using time-dependent ROC (receiver-operating characteristic) in evaluating the performance of Overlap66, Overlap21, and Overlap21plus for their prognostic prediction, we further examined their prediction performance using the intact population (i.e., without the time component) by both ROC-AUC and PR-AUC curves. The precisionrecall (PR) curve is used to account for the imbalance nature of dataset; the event rates are 14.6% (41/280) for OS, 18.9% (53/280) for PFS, and 9.6% (27/280) for DSS, which are much less than 50%. PR-curve was suggested to evaluate biomarker's discriminative performance [44]. According to both ROC-AUC and PR-AUC curves, Overlap66 predicts OS and PFS possibilities better than Overlap21 (Figure 5C,D), while Overlap21plus holds a slight edge over Overlap21 in estimating DSS possibility (Figure 7D,E).

#### *3.5. Alterations in Immune Cell Subsets in High-Risk pRCC Tumors*

Tumor-associated immune cells play critical role in tumor initiation and progression [45,46], suggesting alterations of immune components in Overlap66-stratified high-risk pRCC tumors compared to those of low-risk. To examine this possibility, we profiled all 22 leukocyte subsets in 280 primary pRCC tumors within the TCGA Pancancer dataset using CIBERSORTx (https://cibersortx.stanford.edu/index.php, accessed on 21 July 2021) [47]. Significant alterations in several immune cell subsets between high-risk (*n* = 32) and low-risk tumors (*n* = 248) were detected (Figure 8). Increases in B naïve cells, T follicular helper

cells (Tfh), CD4 T memory (activated) cells, and CD8 T (*p* = 0.075) cells were detected in high-risk local pRCC tumors (Figure 8A), indicating persistent immune reactions towards tumors; this scenario is not uncommon, evident by the co-existence of ATM-derived tumor surveillance (antioncogenic actions) with oncogenic actions during cancer initiation and progression [48]. However, CD8 T cells expressed an upregulation of programmed cell death protein 1 (PDCD1 or PD1) (Figure 8B), a major mechanism contributing to CD8 T cell exhaustion in cancer [49]. Additionally, T regulatory (Treg) cells suppress T cells activation via downregulation of CD80/86 in antigen-presenting dendritic cells [50] and a significant elevation of Treg cells was observed in high-risk pRCC tumors (Figure 8A). Alterations in M1 and M2 composition in high-risk pRCCs (Figure 8A) are consistent with the contributions of tumor-associated macrophages in cancer progression [51]. Decreases in macrophages M2 in high risk pRCC tumors is supported by a downregulation of β-2-adrenergic receptor (ADRB2) in these tumors (Figure 8C); the receptor was associated with M2 macrophages [52]. Reductions of activated mast cells in high-risk pRCC tumors (Figure 8A) suggest a downregulation of immune reactions in facilitating pRCC progression. While B naïve cells, CD8 T cells, M2 macrophages, and activated master cells are similarly clustered in both Overlap66 stratified high- and low-risk pRCCs (Figure S10), activated CD4 T memory cells, Tfh, Treg, and M1 macrophages in the high-risk tumors display different clustering patterns from their counterparts in the low-risk pRCCs (Figure 8D–G). Collectively, changes in immune components in high-risk pRCC tumors stratified by Overlap66 risk scores favor the development of an immune suppressive microenvironment, which might be a mechanism underpinning pRCC progression. This concept provides additional evidence supporting Overlap66 being a novel and effective prognostic biomarker for pRCC. *Cancers* **2021**, *13*, x FOR PEER REVIEW 17 of 25

**Figure 7.** Stratification of the risk of disease-specific survival (DSS) by Overlap21 and Overlap21plus. (**A**,**B**) Separation of a high- and low-risk group based on DSS by Overlap21 and Overlap21plus risk scores. (**C**) tROC analysis. (**D**,**E**) ROC-AUC and PR-AUC curves for the indicated events. **Figure 7.** Stratification of the risk of disease-specific survival (DSS) by Overlap21 and Overlap21plus. (**A**,**B**) Separation of a high- and low-risk group based on DSS by Overlap21 and Overlap21plus risk scores. (**C**) tROC analysis. (**D**,**E**) ROC-AUC and PR-AUC curves for the indicated events.

*3.5. Alterations in Immune Cell Subsets in High-Risk pRCC Tumors*

Tumor-associated immune cells play critical role in tumor initiation and progression

22 leukocyte subsets in 280 primary pRCC tumors within the TCGA Pancancer dataset using CIBERSORTx (https://cibersortx.stanford.edu/index.php, 21 July 2021) [47]. Significant alterations in several immune cell subsets between high-risk (*n* = 32) and low-risk tumors (*n* = 248) were detected (Figure 8). Increases in B naïve cells, T follicular helper cells (Tfh), CD4 T memory (activated) cells, and CD8 T (*p* = 0.075) cells were detected in high-risk local pRCC tumors (Figure 8A), indicating persistent immune reactions towards tumors; this scenario is not uncommon, evident by the co-existence of ATM-derived tumor surveillance (antioncogenic actions) with oncogenic actions during cancer initiation and progression [48]. However, CD8 T cells expressed an upregulation of programmed cell death protein 1 (PDCD1 or PD1) (Figure 8B), a major mechanism contributing to CD8 T cell exhaustion in cancer [49]. Additionally, T regulatory (Treg) cells suppress T cells activation via downregulation of CD80/86 in antigen-presenting dendritic cells [50] and a significant elevation of Treg cells was observed in high-risk pRCC tumors (Figure 8A). Alterations in M1 and M2 composition in high-risk pRCCs (Figure 8A) are consistent with the contributions of tumor-associated macrophages in cancer progression [51]. Decreases in macrophages M2 in high risk pRCC tumors is supported by a downregulation of β-2 adrenergic receptor (ADRB2) in these tumors (Figure 8C); the receptor was associated with M2 macrophages [52]. Reductions of activated mast cells in high-risk pRCC tumors (Figure 8A) suggest a downregulation of immune reactions in facilitating pRCC progression. While B naïve cells, CD8 T cells, M2 macrophages, and activated master cells are similarly clustered in both Overlap66 stratified high- and low-risk pRCCs (Figure S10),

#### *3.6. Critical Contributions of PLK1 to OIP5-Promoted Growth of pRCC Tumors*

PLK1 (Polo-like kinase 1) is one of the upregulated DEGs identified in relation to OIP5 upregulation in both xenograft tumors and primary pRCC, i.e., a component gene of Overlap66 (Table 2). In the same manner, both LYPD6 and PCSK5 were upregulated in primary pRCC tumors with elevated OIP5 expression and in ACHN OIP5 xenografts determined by RNA-seq (Table 2). By using real-time PCR, we confirmed LYPD6 (fold 2.32 ± 0.2/SD, *p* < 0.5) and PCSK5 (fold 2.6 ± 0.1, *p* < 0.05) upregulations in ACHN OIP5 tumors (*n* = 4) compared to ACHN EV tumors (*n* = 6). PLK1 upregulation in xenografts produced by ACHN OIP5 cells compared to those derived from ACHN EV cells was demonstrated by RNA-seq and real-time PCR (Figure 9A,B)). In primary pRCC tumors, OIP5 expression correlates with PLK1 expression with a Pearson correlation value of 0.7 (UALCAN, ualcan.path.uab.edu/home, 1 March, 2021). OIP5, which is also known as Mis18β, is an essential component of the Mis18 complex that is required to load a histone H3 variant CENP-A (centromere protein A) to centromere of newly synthesized DNA strand in early G1 phase [53,54]. PLK1 contributes to CENP-A loading via phosphorylation of M18BP1, a component of the Mis18 complex [55]. In line with this knowledge, we examined whether PLK1 kinase activity plays a role in OIP5-promoted pRCC growth.

PLK1 inhibitors have been developed and approved by FDA as Orpha Drug Designation for cancer therapy [56,57]. The PLK1 inhibitor BI2536 caused G2/M arrest with concurrent reduction in G1 phase in ACHN OIP5 cells without apparent effects on cell cycle distributions of ACHN EV cells at the conditions used (Figure 9C). We then treated mice bearing ACHN EV or ACHN OIP5 cell-produced xenograft tumors with BI2536 when tumors reached 100 mm<sup>3</sup> . In the vehicle treatment group, the OIP5 tumors grew significantly faster compared to the EV tumor (Figure 9D). Administration of BI2536 had no effects on the growth of ACHN EV tumors but significantly inhibited the growth of ACHN OIP5 tumors (Figure 9E,F). In the presence of BI2536, ACHN OIP5 tumor showed marginally slower growth compared to ACHN EV tumors (Figure 9G). Inhibition of PLK1 significantly increases the survival of mice bearing ACHN OIP5 tumor (Figure 9H). As ACHN is a metastatic pRCC cell line [39], evidence supports inhibition of PLK1 being an option in treating metastatic pRCCs with OIP5 upregulation. Collectively, the above observations indicate synthetic lethality between OIP5 and PLK1 in metastatic pRCCs.

tic biomarker for pRCC.

activated CD4 T memory cells, Tfh, Treg, and M1 macrophages in the high-risk tumors display different clustering patterns from their counterparts in the low-risk pRCCs (Figure 8D–G). Collectively, changes in immune components in high-risk pRCC tumors stratified by Overlap66 risk scores favor the development of an immune suppressive microenvironment, which might be a mechanism underpinning pRCC progression. This concept provides additional evidence supporting Overlap66 being a novel and effective prognos-

**Figure 8.** Changes in immune cells in pRCC tumors with high risk of fatality. RNA-seq profiles for 280 pRCC tumors were retrieved from cBioPortal and analyzed for immune cell profiles using the LM22 signature matrix and the CIBERSORTx program (https://cibersortx.stanford.edu/index.php, 21 July 2021) [47]. The analysis setting was with B-mode batch correction and 500 permutations (https://cibersortx.stanford.edu/index.php, 21 July 2021). (**A**) The abundance of the indicated immune cell subsets was determined by their immune fraction scores. Means ± SEMs in high-risk and low-risk tumors stratified by Overlap66 risk score are graphed; \*: *p* < 0.05; \*\*: *p* < 0.01; and \*\*\*: *p* < 0.001 in comparison to low-risk tumors **Figure 8.** Changes in immune cells in pRCC tumors with high risk of fatality. RNA-seq profiles for 280 pRCC tumors were retrieved from cBioPortal and analyzed for immune cell profiles using the LM22 signature matrix and the CIBERSORTx program (https://cibersortx.stanford.edu/index.php, accessed on 21 July 2021) [47]. The analysis setting was with B-mode batch correction and 500 permutations (https://cibersortx.stanford.edu/index.php, accessed on 21 July 2021). (**A**) The abundance of the indicated immune cell subsets was determined by their immune fraction scores. Means ± SEMs in high-risk and low-risk tumors stratified by Overlap66 risk score are graphed; \*: *p* < 0.05; \*\*: *p* < 0.01; and \*\*\*: *p* < 0.001 in comparison to low-risk tumors by 2-tailed *t*-test. (**B**,**C**) Boxplots for the expression of PDCD1 and ADRB2 in low- and highrisk pRCC tumors; statistical analyses were conducted using Welch *t*-test with *p*-value adjusted with the Holm–Bonferroni (Holm) method. (**D**–**G**) Clustering of the indicated immune cell types associated with low risk (Overlap66−) and high risk (Overlap66+) tumors by tSNE (t-distributed stochastic neighbor embedding); the marked clusters are enriched with high-risk tumors. The graph was produced using CIBERSORTx (https://cibersortx.stanford.edu/index.php, accessed on 21 July 2021).

**Figure 9.** PLK1 inhibitor reduces OIP5-mediated pRCC tumorigenesis. (**A**,**B**) RNA-seq and real-time PCR analyses of PLK1 expression in ACHN EV and ACHN OIP5 tumors. RNA-seq was performed in 3 each from ACHN EV and ACHN OIP5 tumors. (**C**) ACHN EV and ACHN OIP5 cells were treated with DMSO (−) or the PLK1 inhibitor (BI2536) in 40 nM PLK1 inhibitor for 72 h, followed by quantification of cell cycle distributions. Experiments were repeated 3 times; means ± SEMs are graphed. \*\*: *p* < 0.01, \*\*\*: *p* < 0.001, \*\*\*\*: *p* < 0.0001 in the indicated comparisons by 2-tailed Student's test. (**D**–**G**) Mice bearing ACHN EV or ACHN OIP5 tumors were treated with vehicle or BI2536 (50 mg/kg) intravenously. The overall profiles of tumor growth in the vehicle treated setting (**D**); tumor volumes were recorded following treatments (**E**–**G**). Statistical analyses were performed using 2-tailed Student's *t*-test; \*: *p* < 0.05, \*\*\*: *p* < 0.001. (**H**) Kaplan Meier curve for the indicated mice reaching endpoints. Statistical analysis was performed using logrank test. **Figure 9.** PLK1 inhibitor reduces OIP5-mediated pRCC tumorigenesis. (**A**,**B**) RNA-seq and real-time PCR analyses of PLK1 expression in ACHN EV and ACHN OIP5 tumors. RNA-seq was performed in 3 each from ACHN EV and ACHN OIP5 tumors. (**C**) ACHN EV and ACHN OIP5 cells were treated with DMSO (−) or the PLK1 inhibitor (BI2536) in 40 nM PLK1 inhibitor for 72 h, followed by quantification of cell cycle distributions. Experiments were repeated 3 times; means ± SEMs are graphed. \*\*: *p* < 0.01, \*\*\*: *p* < 0.001, \*\*\*\*: *p* < 0.0001 in the indicated comparisons by 2-tailed Student's test. (**D**–**G**) Mice bearing ACHN EV or ACHN OIP5 tumors were treated with vehicle or BI2536 (50 mg/kg) intravenously. The overall profiles of tumor growth in the vehicle treated setting (**D**); tumor volumes were recorded following treatments (**E**–**G**). Statistical analyses were performed using 2-tailed Student's *t*-test; \*: *p* < 0.05, \*\*\*: *p* < 0.001. (**H**) Kaplan Meier curve for the indicated mice reaching endpoints. Statistical analysis was performed using logrank test.

#### **4. Discussion 4. Discussion**

Papillary RCC is a minor type of RCC compared to ccRCC which composes 75–80% of RCC cases. Nonetheless, pRCC can be as aggressive as ccRCC, particularly the T2P tumors which usually have more aggressive potential than ccRCC. As a minor RCC type, research on pRCC falls short compared to ccRCC. Therefore, the current understanding on pRCC remains limited, which presents a major concern particularly considering pRCC being associated with poor prognosis. The situation calls for improvement in risk assessment and personalized therapies in managing pRCC. We provide here the first evidence for OIP5 being an important oncogenic factor of Papillary RCC is a minor type of RCC compared to ccRCC which composes 75–80% of RCC cases. Nonetheless, pRCC can be as aggressive as ccRCC, particularly the T2P tumors which usually have more aggressive potential than ccRCC. As a minor RCC type, research on pRCC falls short compared to ccRCC. Therefore, the current understanding on pRCC remains limited, which presents a major concern particularly considering pRCC being associated with poor prognosis. The situation calls for improvement in risk assessment and personalized therapies in managing pRCC.

pRCC. This concept is supported by multiple pieces of evidence with respect to the impact of OIP5 upregulation on the tumorigenesis of pRCC cells in vitro and in vivo as well as the association of OIP5 upregulation with primary pRCC. Although we have made exten-We provide here the first evidence for OIP5 being an important oncogenic factor of pRCC. This concept is supported by multiple pieces of evidence with respect to the impact of OIP5 upregulation on the tumorigenesis of pRCC cells in vitro and in vivo as well as the association of OIP5 upregulation with primary pRCC. Although we have made extensive

efforts to knockdown OIP5 in ACHN cells, the attempts were not successful, suggesting OIP5 being essential for ACHN cell survival. This plausibility is in accordance with OIP5 initiating multiple processes critical for pRCC tumorigenesis, including those regulating urogenital system development, immune reaction, and others. Among these features is the expression status of OIP5 and its related DEGs within Overlap66 in CIMP. Although it remains to be determined whether OIP5 and these DEGs contribute to CIMP, this possibility seems likely. Among the pRCC subtypes, CIPM tumors are associated with a metabolic shift towards Warburg metabolism, which include enhancement of glycolysis, fatty acid and lipid metabolism, and hypoxia [12]. These are typical pathways enriched in ACHN OIP5 tumors (Figure 3; Figure S4–S6; Table S2).

OIP5 may also utilize other pathways in promoting pRCC. As an essential component (Mis18β) in the Mis18 complex, OIP5 is required for CENP-A loading and thus centromere formation [53,54]. This process is essential for genome stability, evident by the centromeremediated chromosome segregation. In line with this concept, genes with function in maintaining genome stability are overrepresented in Overlap66; RMI2 (RecQmediated genome instability 2; C16ORF75) [58], RAD54B (RAD54 homolog B) [59], and PLK1 [60] all play roles in genome stability. Furthermore, pathway enrichment analysis of Overlap66 DEGs revealed the top pathways enriched being GO:0071168: protein localization to chromatin (*p* < 0.0001), GO:0140013: meiotic nuclear division (*p* < 0.0001), GO:0006790: sulfur compound metabolic process (*p* < 0.001), and GO:0000724: double-strand break repair via homologous recombination (*p* < 0.001).

One of the neighboring genes to OIP5 is OIP5-AS1 (OIP5 antisense RNA 1). According to GRCh38.p13 (Genome Reference Consortium Human Build 38 patch release 13) released in Feb 28, 2019, the OIP5 genes runs from 41,332,591 to 41,309,273 on chromosome 15 (https://www.ncbi.nlm.nih.gov/gene/11339, accessed on 29 August 2021), while the OIP5-AS1 gene runs from 41,282,697 to 41,313,338 on chromosome 15 (https://www.ncbi. nlm.nih.gov/gene/729082, accessed on 29 August 2021). While both genes have an overlap region of 4065 nucleotides, there is no evidence suggesting a regulatory relationship between OIP5 and OIP5-AS1 [61]. OIP5-AS1 encodes a long non-coding RNA (lnRNA) and possesses oncogenic activities via regulating a set of microRNAs [61]. For instance, OIP5- AS1 was reported to sponge miR-143-3P to enhance cervical cancer [62] and miR-186a-5p to facilitate hepatoblastoma [63]. However, the involvement of OIP5-AS1 in pRCC remains unknown. In view of both OIP5 and OIP5-AS1 being pro-oncogenesis and their adjacent genetic locations, potential functional connections between both in pRCC pathogenesis and progression is worthy of future investigation. In supporting this possibility, we noticed OIP5-AS1 being upregulated (1.37 folds, *p* = 0.00464 and *q* = 0.0459) in pRCC tumors expressing high levels of OIP5 compared to those with low levels of OIP5 expression (Figure 1G).

OIP5 is a tumor-associated antigen (TAA), owning to its largely restricted expression in human testis and its upregulation in multiple cancers [64,65]. We noticed that testisassociated proteins are also enriched in Overlap66, including OIP5, TEX15 (testis expressed 15, meiosis, and synapsis associated), SPAG1 (sperm associated antigen 1), and SPATA18 (spermatogenesis associated 18) (Table 2). It is thus tempting to propose an involvement of some testis events in pRCC tumorigenesis. OIP5 possesses a robust prognostic potential (Figure 1G). This predictive power is significantly strengthened in OIP5-derived multigene sets: Overlap66, Overlap21, and Overlap21plus. Because of the small number of component genes and its effectiveness in predicting OS, PFS, and DSS, Overlap21 may offer primary clinical application with the other two provide assisting roles. These multigene sets possess great potential to be implemented into clinical applications. This possibility is supported by a very good out-of-sample performance of Overlap66 in stratification of pRCC fatality risk (Figure 6) and these stratifications can be effective using a range of cutoff points (Figure 6B,C). Clinical applications of Overlap66 and its-related multigene panels may significantly improve our ability in predicting prognosis and potentially even the development of personalized therapies.

Although a recent phase 2 clinical trial suggests the MET inhibitor cabozantinib improves PFS and OS in patients with metastatic pRCC compared to a current standard of care with sunitinib [66], much more needs to be done to confirm its efficacy. The dependence on MET signaling is likely much less in T2P compared to T1P, which needs to be considered in using MET inhibitors in treating patients with T2P tumors. Our finding of PLK1 inhibitor being effective in inhibition of ACHN OIP5 tumor growth may have significant clinical applications in treating metastatic pRCC with OIP5 upregulation; this will offer a venue for potential utilization of personalized therapy in pRCC. This possibility can be readily explored as volasertib, a PLK1 inhibitor, has been granted Orpha Drug Designation status in treating AML (acute myeloid leukemia) in 2014 and rhabdomyosarcoma in 2020 (https://oncoheroes.com/press-releases-content/2020/10/14/volasertib-a-potentialnew-treatment-for-rhabdomyosarcoma-receives-orphan-drug-designation-from-the-us-fda, accessed on 31 May 2021) by FDA. Even for BI2356 used in this study, its clinical safety was deemed acceptable based on multiple phase II clinical trials (NCT00701766, NCT00376623, and NCT00526149) on solid cancer. Intriguingly, we observed changes in immune cells in pRCC tumors stratified by Overlap66, an OIP5-derived multigene panel, including increases of Treg cells and PD1 upregulation in CD8 T cells (Figure 8). This suggests that these patients might benefit from rescuing of CD8 T cell exhaustion via PD1-based immune therapies. Treg action can be suppressed via CTLA-4 immune therapy. In this regard, combinations of PLK1 inhibitor and PD1 or CTLA-4 immune therapies might optimize personalized treatment. Collectively, this research enhances our understanding of pRCC and suggests novel means in predicting pRCC prognosis and in developing personalized therapy. Nonetheless, additional work is required to realize these potentials.

#### **5. Conclusions**

We report here a novel and thorough investigation of OIP5's contributions to pRCC. OIP5 upregulations robustly predict the survival possibility of pRCC patients. The multigene panel Overlap66, a portion of the OIP5 network, possesses an impressive prognostic potential in predicting pRCC progression, disease-specific survival, and overall survival; the predictions are associated with an excellent out-of-sample performance, indicating its potential clinical applications. Furthermore, PLK1 is among Overlap66 and displays synthetic lethality with OIP5; inhibition of PLK1 using BI2356 only suppresses the growth of xenograft tumors generated by ACHN OIP5 cells but not the growth of tumor produced by ACHN EV cells, supporting a targeted and personalized therapy for pRCCs with OIP5 elevations. Collectively, combinational use of Overlap66 and PLK1 inhibitors may open an era of personalized therapy in pRCC.

#### **6. Patents**

A US provisional patent (63/202,616) has been filed.

**Supplementary Materials:** The following are available online at https://www.mdpi.com/article/10 .3390/cancers13174483/s1, Figure S1: OIP5 upregulation in ccRCC; Figure S2: OIP5 enhances ACHN cell proliferation, colony formation, and growth in soft agar; Figure S3: Increases in OIP5 and CDK2 expression in ACHN OIP5 tumors; Figure S4: Gene set enrichment in ACHN OIP5 tumors; Figure S5: Enrichment of h-Hallmarks Gene Set collection in ACHN OIP5 tumors; Figure S6: Enhancement of the glycolysis pathway in ACHN OIP5 tumors; Figure S7: Prediction of OS, PFS, and DSS possibilities by Overlaop66; Figure S8: Heatmap of Overlap66 gene expressions in the indicated tissues; Figure S9: Cutoff point of Overlap66 risk score in the estimation of OS shortening; Figure S10: Clustering of the indicated immune cell subsets with tSNE; Table S1: Enrichment of the C6 oncogenic gene sets collection within OIP5-related DEGs; Table S2: Enrichment of the Hallmark gene sets collection within OIP5-related DEGs; Table S3: Differentially expressed genes derived from the comparison between ACHN OIP5 xenografts and ACHN EV tumors; Table S4: Pathway enrichment in OIP5 related DEGs (*p*.adj < 0.05 and fold change > |1.5|) derived from the comparison between ACHN OIP5 xenografts and ACHN EV tumors; Table S5: DEGs derived from primary pRCC with respect to

OIP5 expression; Table S6A: Overlap66 genes' association with OS possibility and their oncogenic functions; Table S6B: Overlap21 and Overlap21plus gene lists.

**Author Contributions:** M.J.C. performed analyses of OIP5 expression in primary pRCC using TMA, constructed ACHN EV and OIP5 stable lines, examined the impact of OIP5 on a set of oncogenic events in ACHN cell in vitro and on the cell's ability of forming xenografts in vivo, and investigated the inhibition of PLK1 inhibitor on the growth of ACHN OIP5 cell-derived xenografts. Y.G. contributed to production of xenograft tumors. M.J.C. and L.H. performed RNA-seq analyses. X.L. and Y.D. examined OIP5-affected immune reactions. M.J.C., X.L. and W.M. contributed to the molecular manipulations of OIP5. A.K. contributed to clinical analysis of primary pRCC. D.T. performed prognostic analyses. All authors contributed to data interpretation. M.J.C., A.K. and D.T. designed this research. A.K. and D.T. supervised the research. M.J.C. and D.T. prepared the manuscript. M.J.C., Y.G., X.L. and A.K. revised the manuscript. All authors have read and agreed to the published version of the manuscript.

**Funding:** Y.G. is supported by Studentship provided by Ontario Graduate Student Fellowship and Research Institute of St Joe's Hamilton. Grant support was from Canadian Cancer Society (grant #: 319412) and CIHR to DT. This research was also supported by funds from Urological Cancer Centre for Research and Innovation (UCCRI), St Joseph's Hospital, Hamilton, ON L8N 4A6, Canada.

**Institutional Review Board Statement:** The study was conducted according to the guidelines of the Declaration of Helsinki and approved by the Institutional Review Board of McMaster University (16-06-23) and Hamilton Integrated Research Ethics Board (11-3472).

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Data is present in this article and Supplementary Materials.

**Acknowledgments:** The results shown here are in part based on data generated by the TCGA Research Network (http://cancergenome.nih.gov/, accessed on 1 September 2020).

**Conflicts of Interest:** All authors declare no competing interests.

#### **References**


## *Perspective* **Metastasis, an Example of Evolvability**

**Annick Laruelle 1,2, Claudia Manini <sup>3</sup> , Elena Iñarra 1,4 and José I. López 5,6,\***


**Simple Summary:** Cancer is a complex disease. Modern molecular technologies are progressively unveiling its genetic and epigenetic complexity, but still many key issues remain unknown. Considering cancer as a social dysfunction in a community of individuals has provided new perspectives of analysis with promising results. This narrative considers both approaches with respect to the metastatic process, the final cause of death in most patients affected by this disease.

**Abstract:** This overview focuses on two different perspectives to analyze the metastatic process taking clear cell renal cell carcinoma as a model, molecular and ecological. On the one hand, genomic analyses have demonstrated up to seven different constrained routes of tumor evolution and two different metastatic patterns. On the other hand, game theory applied to cell encounters within a tumor provides a sociological perspective of the possible behaviors of individuals (cells) in a collectivity. This combined approach provides a more comprehensive understanding of the complex rules governing a neoplasm.

**Keywords:** cancer; metastasis; genomic analysis; microenvironment; tumor ecology; game theory

#### **1. Introduction**

Modern treatment modalities are obtaining longer survivals and even cure in a significant percentage of patients with cancer, transforming the disease into a chronic-like condition. However, cancer still remains today the leading cause of death in Western countries. Metastatic dissemination is responsible for more than 90% of tumor deaths and is a challenge for modern oncology. However, metastasis is an extremely inefficient process. It has been estimated that only 0.01% of circulating tumor cells succeed in developing it [1]. The acquisition of motility by tumor cells involves a complex self-organized and self-regulated systemic cellular organization emerging from Hopfield-like dynamics [2]. These dynamics tightly regulate cell migration and other critical functions like associative memory in the cell, as it has been demonstrated very recently in unicellular organisms like amoebae [3].

The initialization of the metastatic process is possible only under regulated cellular metabolic conditions. These particular conditions allow the occurrence of epithelial-tomesenchymal transition processes that enable epithelial cells to acquire amoeboid motility through the development of a cascade of molecular events. Unveiling the intricate interactions between tumor cells themselves on the one hand, and between tumor and host cells on the other is not totally understood and remains one of the main next frontiers in oncology. Such an approach will benefit from an ecological perspective, a viewpoint that will be considered later in this narrative.

**Citation:** Laruelle, A.; Manini, C.; Iñarra, E.; López, J.I. Metastasis, an Example of Evolvability. *Cancers* **2021**, *13*, 3653. https://doi.org/10.3390/ cancers13153653

Academic Editors: Emilie Mamessier-Birnbaum and Aldo M. Roccaro

Received: 17 June 2021 Accepted: 20 July 2021 Published: 21 July 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

Evolvability is an ecological term that reflects the adaptive ability of a species to generate and maintain a heritable phenotypic diversification to prevent extinction [4]. At least theoretically, metastasis may be conceived as an example of tumor evolvability achieved by genetic and epigenetic modifications that confer migratory abilities to cells to escape to more "favorable" ecosystems. A significant amount of studies are being published focusing on the intricate genomic/epigenomic complexity of this issue, unveiling the multiple routes that enable malignant cells to acquire locomotion capacities. However, very few studies have analyzed the collective pressures and cell-to-cell interactions that may explain the reasons for which malignant cells decide to migrate far away. This new approach implies considering tumors as a sort of social dysfunction [5] and analyzing malignant cells and their microenvironment from an ecological viewpoint [6].

In this perspective, we approach the metastatic process from genomic/epigenomic, ecological, and sociological perspectives. On the one hand, we review some of the last salient genomic/epigenomic findings related to the acquisition of metastatic capacities focusing on a paradigmatic example of intratumor heterogeneity and metastatic selection: The clear cell renal cell carcinoma (CCRCC). Tumor/non-tumor cell interactions (endothelial cells, tumor-infiltrating lymphocytes, and cancer-associated fibroblasts) are also mentioned as a substantial part of the adaptive processes leading to tumor migration. Finally, we analyze cell-to-cell interactions using a game theory approach and hypothesize that metastasis may be simply a specific response of a subset of tumor cells. Such response would consist of searching for collective stability far away from the primary tumor to improve their collective wellness and prevent extinction.

#### **2. Molecular Approach**

The molecular characterization of CCRCC has been largely analyzed [7]. Clonal and sub-clonal evolutions generate in some of these carcinomas a high variability of metastatic patterns. While some tumors typically develop only one metastatic clone, others are able to develop several ones with specific capacities and topographic affinities [8]. CCRCC is a paradigmatic example of intratumor heterogeneity (ITH) [9]. Actually, some deterministic constraints with prognostic impact [10] and two different metastatic patterns [11] have been identified in this tumor. A rule-based classification system supported by unsupervised clustering comparing different genomic, histological, and clinical data has shown up to seven evolutionary subtypes [10]. Multiple clonal drivers, *BAP-1* driven, and *VHL* wild type displayed a punctuated evolution and accelerated clinical progression. On the other hand, CCRCC related to *PBRM-1* gene dysfunction (PBRM-1 → SETD2, PBRM-1 → PI3K, PBRM-1 → SCNA) pursued a less aggressive clinical progression and a branched evolution. The seventh subtype detected, the *VHL* mono-driver, showed a monoclonal structure without additional driver mutations.

Losses of 9p and 14q are molecular alterations linked with metastatic ability in CCRCC [11]. Tumors that follow a punctuated model of evolution display high chromosomal complexity but low ITH. Here, a single clone with high fitness fixes early in its evolution and occupies a significant percentage of the tumor mass following a Darwinian pattern. These cases develop multiple metastases early in their evolution (rapid progression). The branching pattern, however, is also a Darwinian example of tumor evolution, but in this case, there is high clonal and sub-clonal diversification. As a result, chromosomal complexity is low, but ITH is high. Branching-type tumors develop few metastases late in their evolution (attenuated progression).

Hypoxia promotes metastases, but this factor is not homogeneously distributed across the tumor. Actually, a tumor spatial specialization pattern has been very recently detected in CCRCC. Tumor areas with metastatic abilities are located in the tumor interior in a multi-regional analysis of 79 cases [12]. Tumor necrosis and higher Ki-67 index, histological grade, and chromosomal complexity were detected in the tumor center, where microenvironmental pressures and the struggle for survival are supposedly higher due to higher levels of hypoxia.

The metastatic capacity is associated with the acquisition of a stem cell-like phenotype because of the evolvability of a subset of tumor cells, the so-called metastasis-initiating cells [13]. Metastatic dissemination (local invasion, intravasation, blood and/or lymphatic circulation, extravasation and extravascular spread), dormancy, immune evasion, organ colonization, and local tropism development conform to the stepwise process followed by this special subset of traveling tumor cells. This concatenated sequence of events needs a dynamic epigenetic remodeling to adapt these cells to the ever-changing environments. Abnormal DNA methylation (methyl-binding proteins, post-translational histone modifications, miRNAs, lncRNAs, RNA methylation) is one of the epigenetic mechanisms most extensively studied and an actionable target for future treatments [14].

#### **3. Microenvironmental Context**

A question arises at this point: How many times can tumor cells migrate? In other words, can a metastasis metastasize? The answer is yes, as it has been demonstrated in genomic studies analyzing the clonality patterns of the widespread tumor seeding in renal, prostate, breast, pancreatic, and colo-rectal cancers [15,16]. Such complex polyclonal dissemination seems a cornerstone in the development of prostate cancer clones resistant to castration, as demonstrated by Gundem et al. [15].

The relationship between tumor cells and their microenvironment is crucial in cancer progression and metastatic process. Immune cells, neovascular endothelia, and fibroblasts are the best-studied tumor-associated cells. These stromal elements dynamically interact between and with tumor cells. In addition, this interplay is becoming a promising therapeutic target, with immune checkpoint blockade (PD-1, PD-L1, CTLA-4) and anti-angiogenesis as two of the most representative examples of precision therapies in CCRCC [17] and other neoplasms. In fact, the combination of both treatments in a subset of these patients has a positive effect based on the synergic actions of antiangiogenic and checkpoint blockers [17].

PD-1 and its ligand PD-L1 (B7-H1) blockade has been implemented in clinical practice as a promising therapeutic strategy in CCRCC and other tumors, but the immunohistochemical selection of candidates, with different antibodies and cut-offs, is still controversial [18]. By contrast, the soluble fraction of PD-L1 has been associated with prognosis and metastatic status in these tumors [19]. Immune checkpoint blockade seems especially useful in the so-called inflamed tumors, a group of aggressive CCRCC with abundant tumor-infiltrating lymphocytes linked to *BAP-1* gene inactivation and tendency to early metastases [20]. On the other hand, B7-H3, another immune regulator of the B7 family, has been directly implicated in the metastatic process [21], expanding the list of candidates for checkpoint blockade in the clinical practice.

CCRCC is a neoplasm associated with *VHL* gene malfunction that creates a pseudohypoxic status in neoplastic cells. This condition induces the initialization of the VEGF cascade [9], ending in neoangiogenesis. A new definition of microvessel density in CCRCC has been recently published [22]. These authors consider the sum of classical microvessel density and vasculogenic mimicry as the total microvessel density and correlate this sum with prognosis [22]. Since these neoplasms are highly vascularized, the antiangiogenic tyrosine-kinase inhibitors have had a relevant role in the therapeutic armamentarium, particularly in metastatic patients [23].

Not all CCRCC, however, display the same degree of angiogenesis. Predominantly angiogenic examples are typically associated with *PBRM-1* gene inactivation [20] and show a pancreatic tropism [24]. Interestingly, pancreatic metastases in angiogenic-type low-grade CCRCC have been detected significantly later in the tumor evolution compared with metastases in the rest of the sites [11].

Cancer-associated fibroblasts (CAF) are essential elements in the tumor microenvironment [25]. The interplay between tumor cells and CAF activates the epithelial-tomesenchymal transition process providing a stem cell phenotype to epithelial tumor cells, thus enabling tumor cell migration [26]. In this interaction, CAF produce fibroblast activation protein-α (FAP), a protein associated with biological aggressiveness [27] and early

metastases [28] in CCRCC. Other proteins produced by CAF, however, also contribute to tumor progression [29]. Coercive feedback signaling from cancer cells to CAF does exist, thus assuring the maintenance of FAP production and perpetuating the loop [26]. has opened up new perspectives to understand cancer dynamics. The Hawk–Dove game (Figure 1) is a classical model in biology [31]. It assumes a population whose members have bilateral encounters to divide a resource (v), with some cost (c) associated with fight (c > v). In every encounter, each individual can behave as a

Game theory [30] is a branch of applied mathematics initially applied to economics that investigates interactions between individuals. It has been used to examine complex problems in biology [31] and, more recently, in oncology [32]. Considering cancer as a social dysfunction [5] in which cells interact following different strategies, game theory

Cancer-associated fibroblasts (CAF) are essential elements in the tumor microenvironment [25]. The interplay between tumor cells and CAF activates the epithelial-to-mesenchymal transition process providing a stem cell phenotype to epithelial tumor cells, thus enabling tumor cell migration [26]. In this interaction, CAF produce fibroblast activation protein-α (FAP), a protein associated with biological aggressiveness [27] and early metastases [28] in CCRCC. Other proteins produced by CAF, however, also contribute to tumor progression [29]. Coercive feedback signaling from cancer cells to CAF does exist, thus assuring the maintenance of FAP production and perpetuating

#### **4. Sociological Approach** "hawk" and escalate to a fight or as a "dove" and back down. Therefore, if one individual

**4. Sociological Approach** 

the loop [26].

Game theory [30] is a branch of applied mathematics initially applied to economics that investigates interactions between individuals. It has been used to examine complex problems in biology [31] and, more recently, in oncology [32]. Considering cancer as a social dysfunction [5] in which cells interact following different strategies, game theory has opened up new perspectives to understand cancer dynamics. acts as a hawk while the other acts as a dove, the "hawk" gets the resource (v) and the dove gets nothing (0). If two "hawks" meet, there is a fight, the winner receives the resource (v) and the other faces the cost of the fight (−c). On average, the two hawks receive (v − c)/2. If two individuals act as dove, they share the resource. On average, the two doves get v/2. The concept of evolutionarily stable strategy (ESS) is used to solve the game. ESS in

The Hawk–Dove game (Figure 1) is a classical model in biology [31]. It assumes a population whose members have bilateral encounters to divide a resource (v), with some cost (c) associated with fight (c > v). In every encounter, each individual can behave as a "hawk" and escalate to a fight or as a "dove" and back down. Therefore, if one individual acts as a hawk while the other acts as a dove, the "hawk" gets the resource (v) and the dove gets nothing (0). If two "hawks" meet, there is a fight, the winner receives the resource (v) and the other faces the cost of the fight (−c). On average, the two hawks receive (v − c)/2. If two individuals act as dove, they share the resource. On average, the two doves get v/2. game theory captures the resilience of a strategy against another in the sense that it assumes that most members of the population play the ESS but a small proportion of members, here termed mutants, chooses a different strategy. In this context, each mutant's expected payoff is smaller than the expected payoff of an individual who plays the ESS. Consequently, mutants are driven out of the population. As a consequence, the Hawk–Dove game has a unique ESS, which consists of a fraction, v/c, of the population which plays hawk while the remaining population plays dove.


*Cancers* **2021**, *13*, x FOR PEER REVIEW 4 of 6

**Figure 1. Hawk–Dove game matrix**. Each of the two individuals chooses one strategy. The matrix summarizes the payoffs of the four possible results for the individual choosing the row strategy playing against an individual choosing the column strategy (v = resource, c = cost). **Figure 1.** Hawk–Dove game matrix. Each of the two individuals chooses one strategy. The matrix summarizes the payoffs of the four possible results for the individual choosing the row strategy playing against an individual choosing the column strategy (v = resource, c = cost).

> This description is appropriate for analyzing the interactions between cancer cells and cancer-associated fibroblasts in CCRCC [25] that we have previously mentioned in this narrative. The resource, in this case, is the diffusible factor fibroblast activation protein-α (FAP) that improves cancer fitness, while the cost is the effort of producing and excreting it to the medium. It is thus theoretically possible to predict that a homogeneous neoplasm will achieve its ESS. However, there is no such thing as a perfectly homogeneous tumor in real life, so a heterogeneous population must be considered*.* In the heterogeneous Hawk–Dove game The concept of evolutionarily stable strategy (ESS) is used to solve the game. ESS in game theory captures the resilience of a strategy against another in the sense that it assumes that most members of the population play the ESS but a small proportion of members, here termed mutants, chooses a different strategy. In this context, each mutant's expected payoff is smaller than the expected payoff of an individual who plays the ESS. Consequently, mutants are driven out of the population. As a consequence, the Hawk–Dove game has a unique ESS, which consists of a fraction, v/c, of the population which plays hawk while the remaining population plays dove.

> [33], the population is divided into two types. In consequence, each individual conditions This description is appropriate for analyzing the interactions between cancer cells and cancer-associated fibroblasts in CCRCC [25] that we have previously mentioned in this narrative. The resource, in this case, is the diffusible factor fibroblast activation protein-α (FAP) that improves cancer fitness, while the cost is the effort of producing and excreting it to the medium.

> It is thus theoretically possible to predict that a homogeneous neoplasm will achieve its ESS. However, there is no such thing as a perfectly homogeneous tumor in real life, so a heterogeneous population must be considered. In the heterogeneous Hawk–Dove game [33], the population is divided into two types. In consequence, each individual conditions his/her action depending on his/her opponent's type. In the heterogeneous game, the previously described ESS is not stable any longer. Therefore, the game will have two ESSs. Each ESS is characterized by one of the two types of individuals being discriminated. In such a context, one type of individual receives a higher payoff than the other. Furthermore, the payoff of the discriminated individuals in the heterogeneous game is smaller than the payoff obtained in the homogeneous one.

> However, there is an unanswered question in oncology: Why does a malignant tumor metastasize? From an ecological viewpoint, the goal of cell migration is tumor survival. From a sociological perspective, however, it can be hypothesized that the ultimate common

endpoint of neoplastic cells that migrate far away is to achieve a higher payoff elsewhere. Here, the set of discriminated cells in the heterogeneous population that governs the primary tumor might develop metastatic abilities just to escape to a friendlier environment where a higher payoff is initially possible, at least theoretically.

#### **5. Conclusions**

Tumor spatial specialization with metastasizing subclones located in the tumor interior, the demonstrated ability of metastases to metastasize, and the sociological tumor cell interactions unveiled by the Hawk–Dove game reinforce the storyline of this perspective in the sense that the search for a better environment by tumor cells is a constant event in malignant tumors.

**Author Contributions:** A.L., C.M., E.I. and J.I.L. conceived and wrote the article. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Conflicts of Interest:** The authors have no conflicts of interest.

#### **References**


## *Article* **Restaging the Biochemical Recurrence of Prostate Cancer with [ <sup>68</sup>Ga]Ga-PSMA-11 PET/CT: Diagnostic Performance and Impact on Patient Disease Management**

**Aloÿse Fourquet <sup>1</sup> , Lucien Lahmi <sup>2</sup> , Timofei Rusu <sup>1</sup> , Yazid Belkacemi <sup>3</sup> , Gilles Créhange <sup>4</sup> , Alexandre de la Taille <sup>5</sup> , Georges Fournier <sup>6</sup> , Olivier Cussenot <sup>7</sup> and Mathieu Gauthé 1,8,\***


**Simple Summary:** We aimed to evaluate the diagnostic performance, impact on patient disease management, and therapy efficacy prediction of [68Ga]Ga-PSMA-11 PET/CT on 294 patients with biochemical recurrence of prostate cancer. We established a composite standard of truth for the imaging based on all clinical data available collected during the follow-up period with a median duration of follow-up of 17 months. Using this methodology, we found that the overall per-patient sensitivity and specificity were both 70%, the patient disease management was changed in 68% of patients, and that [68Ga]Ga-PSMA-11 PET/CT impacted this change in 86% of patients. The treatment carried out on the patient was considered effective in 78% of patients; in 89% of patients when guided by [68Ga]Ga-PSMA-11 PET/CT versus 61% of patients when not guided by [68Ga]Ga-PSMA-11 PET/CT.

**Abstract:** Background: Detection rates of [68Ga]Ga-PSMA-11 PET/CT on the restaging of prostate cancer (PCa) patients presenting with biochemical recurrence (BCR) have been well documented, but its performance and impact on patient management have not been evaluated as extensively. Methods: Retrospective analysis of PCa patients presenting with BCR and referred for [68Ga]Ga-PSMA-11 PET/CT. Pathological foci were classified according to six anatomical sites and evaluated with a threepoint scale according to the uptake intensity. The impact of [68Ga]Ga-PSMA-11 PET/CT was defined as any change in management that was triggered by [68Ga]Ga-PSMA-11 PET/CT. The existence of a PCa lesion was established according to a composite standard of truth based on all clinical data available collected during the follow-up period. Results: We included 294 patients. The detection rate was 69%. Per-patient sensitivity and specificity were both 70%. Patient disease management was changed in 68% of patients, and [68Ga]Ga-PSMA-11 PET/CT impacted this change in 86% of patients. The treatment carried out on patient was considered effective in 89% of patients when guided by [ <sup>68</sup>Ga]Ga-PSMA-11 PET/CT versus 61% of patients when not guided by [68Ga]Ga-PSMA-11 PET/CT (*p* < 0.001). Conclusions: [68Ga]Ga-PSMA-11 PET/CT demonstrated high performance in locating PCa recurrence sites and impacted therapeutic management in nearly two out of three patients.

**Citation:** Fourquet, A.; Lahmi, L.; Rusu, T.; Belkacemi, Y.; Créhange, G.; de la Taille, A.; Fournier, G.; Cussenot, O.; Gauthé, M. Restaging the Biochemical Recurrence of Prostate Cancer with [68Ga]Ga-PSMA-11 PET/CT: Diagnostic Performance and Impact on Patient Disease Management. *Cancers* **2021**, *13*, 1594. https://doi.org/10.3390/ cancers13071594

Academic Editor: Jonas Cicenas

Received: 20 February 2021 Accepted: 26 March 2021 Published: 30 March 2021

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2021 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

**Keywords:** prostatic neoplasms; positron-emission tomography; decision making

#### **1. Introduction**

Prostate cancer (PCa) is the most prevalent cancer in men worldwide, accounting for approximately 21% of all diagnosed cancers [1]. Up to 40% of patients with PCa initially treated with curative intent will experience biochemical recurrence (BCR) [2,3], which is defined following radical prostatectomy by two consecutive rising prostatespecific antigen (PSA) values >0.2 ng/mL, or after primary radiation therapy by any PSA increase ≥2 ng/mL higher than the PSA nadir value, regardless of the serum concentration of the nadir [4,5]. Accurately locating the recurrence site(s) is essential for optimizing patient management, as localized or oligometastatic recurrences could be eligible for salvage targeted treatments with curative intent, such as local therapy [6] or stereotactic radiation therapy [7]. Conventional imaging modalities, such as bone scan and computed tomography (CT), have limited utility in this setting, especially when PSA serum levels are below 10 ng/mL [8]. [18F]fluorocholine positron emission tomography associated with computed tomography (PET/CT) was demonstrated to have better performance than conventional imaging but may also fail to locate recurrence at low PSA levels [9].

The prostate-specific membrane antigen (PSMA) is a transmembrane protein that is over-expressed by up to 1000-fold in almost all PCa cells [10,11]. The recent introduction of PET/CT using a PSMA radioligand for imaging of PCa BCR has shown promising results due to its performance in detecting lesions, even at very low PSA levels, impacting on the therapeutic management of PCa patients [12–14].

Although the detection rates of <sup>68</sup>Ga-PSMA PET/CT have been well documented, its sensitivity, specificity, impact on patient management, and therapy efficacy prediction have not been evaluated as extensively.

The purpose of this study was to evaluate the diagnostic performance, impact on patient disease management, and therapy efficacy prediction of PET/CT using a PSMA ligand radiolabelled with gallium-68, the [68Ga]Ga-PSMA-11, on the restaging of PCa patients presenting with BCR.

#### **2. Materials and Methods**

#### *2.1. Population*

Patients presenting with BCR of PCa who were addressed to our department for [ <sup>68</sup>Ga]Ga-PSMA-11 PET/CT were consecutively included and retrospectively analyzed. These patients had shown no sign of distant metastases at [18F]fluorocholine PET/CT; for this reason, they were referred to [68Ga]Ga-PSMA-11 PET/CT, based on the French regulation for compassionate use of pharmaceutical, which is authorized on an individual basis by the National Medicine Agency.

Inclusion criteria for patients were as follows: 1—histologically confirmed PCa previously treated with curative intent; 2—no known history of PCa distant metastases (invaded locoregional pelvic lymph node at diagnosis was not considered as metastatic according to the 2009 TNM classification for staging PCa [15]); 3—currently presenting a biochemical recurrence defined as two consecutive rising PSA values above 0.2 ng/mL following radical prostatectomy or any PSA increase greater than or equal to 2 ng/mL higher than the PSA nadir value, regardless of the nadir value, for non-surgical first-line definitive treatments [4].

Exclusion criteria were as follows: 1—PCa with known distant metastases; 2—patients with persistent PSA after prostatectomy (PSA ≥ 0.1 ng/mL) [15] or radiation therapy (nadir PSA < 2 ng/mL with testosterone recovered if previous androgen deprivation therapy (ADT) [5]); 3—patients who were never treated with curative intent for PCa; 4—the presence of a second active neoplasm other than PCa.

This research implied no intervention on the patient. According to French regulations, the approval of an institutional review board was not necessary for performing this retrospective analysis of already available data. Patients were informed that their data collected for the [68Ga]Ga-PSMA-11 PET/CT would be analyzed and published anonymously, and did not object.

## *2.2. [68Ga]Ga-PSMA-11 PET/CT Imaging Procedure*

Gallium-68 was obtained from a [68Ge]Ge/ [68Ga]Ga radionuclide generator (GalliaPharm, Eckert & Ziegler Radiopharma GmBH, Berlin, Germany) and used for radiolabelling of PSMA-11 according to the manufacturer's instructions (IASON GmbH). Patients did not require specific preparation before the injection. Patients received 1–2 MBq/kg of the radiotracer injected in saline via an infusion line.

Images were acquired using a Gemini TF16 (Philips Medical Systems, Cleveland, OH, USA) or a Biograph mCTflow (Siemens Healthcare, Erlangen, Germany) PET/CT. Both PET/CT scanners included time-of-flight technology. Dynamic images were acquired on the pelvis immediately after [68Ga]Ga-PSMA-11 injection (10 images of one-minute duration each) and from vertex to mid-thigh 60 to 90 min after injection. On the Gemini TF16 PET/CT, the pelvis was imaged for 3 min, and every other bed position was imaged for 2 min in 3D mode with a 576 mm FOV and a 144 × 144 matrix. Images were reconstructed from 3 iterations and 33 subsets using the OSEM weighted method. Low-dose CT without contrast-enhancement was performed prior to PET acquisition (120 kVp, 80 mA.s, slice thickness 2.5 mm, pitch 0.813, rotation time 0.5 s, FOV 600 mm). On the Biograph mCTflow PET/CT, the scanning speed was set to 0.7 cm/min over the pelvis and 0.9 cm/min for the rest of the acquisition field. Images were taken in 3D mode with a 780 mm FOV and a 200 × 200 matrix. Images were reconstructed from 2 iterations and 21 subsets using the OSEM weighted method. Low-dose CT without contrast-enhancement was performed prior to PET acquisition (CareDose® automatic modulation for keV and mA.s, slice thickness 2 mm, pitch 0.813, rotation time 0.5 s, FOV 500 mm). A harmonization in PET images between the two scanners by using EQ.PET, a NEMA-referenced SUV across technologies was performed [16].

## *2.3. [68Ga]Ga-PSMA-11 PET/CT Image Analysis*

[ <sup>68</sup>Ga]Ga-PSMA-11 PET/CTs were read on-site on the day of image acquisition (routine unmasked reading) by local nuclear physicians with at least 4 years of experience in reading PET/CT and 6 months of experience in reading [68Ga]Ga-PSMA-11 PET/CT. An expert nuclear medicine physician with 10 years of experience in reading PET/CT and 4 years of experience in reading [68Ga]Ga-PSMA-11 PET/CT, who was blinded to all clinical data, performed a retrospective reading of the [68Ga]Ga-PSMA-11 PET/CTs of the patients who met the inclusion criteria (masked retrospective reading). Anonymized images presented in a random order were independently reviewed on a dedicated workstation (Syngo.via, Siemens Healthcare).

The expert reader assessed uptakes across six anatomical sites and attributed them value on a 3-point qualitative scale according their intensity: 0—no suspicious uptake (at best equal to muscle background); 1—equivocal uptake (between background in muscles and vessels); 2—malignant uptake (higher than background in vessels) [17]. CT images were used for anatomic allocation of a suspicious focus and to facilitate diagnosis. We considered six anatomical sites: prostate/prostatic lodge, pelvic lymph nodes (up to the common iliac lymph nodes), paraaortic lymph nodes, lymph nodes above the diaphragm, bone, and viscera. If at least one suspicious uptake (equivocal or malignant) was detected in an anatomical site, the entire areas was quoted as equivocal or malignant. The intensity of the most intense abnormal uptake was determined by the maximum standardized uptake value (SUVmax) for each anatomical site during the retrospective reading. Based on the results of [68Ga]Ga-PSMA-11 PET/CT, we categorized patients as oligometastatic if they present between 1 and 3, 4, or 5 distant malignant uptakes excluding the prostate/prostatic lodge (oligo-3, oligo-4, and oligo-5, respectively); polymetastatic if more than 5 distant malignant uptakes were detected [18].

## *2.4. Follow-Up and Evaluation of [68Ga]Ga-PSMA-11 PET/CT Impact on Patient Disease Management*

After imaging, the clinical follow-up was performed for each patient by his referring physician. Clinicians decided for each patient management plan during multidisciplinary meetings dedicated to urological cancers. These multidisciplinary meeting boards were constituted by a urologist, a radiation oncologist, a medical oncologist, a pathologist, a radiologist, and a nuclear medicine physician. They analyzed all clinical data available before and after [68Ga]Ga-PSMA-11 PET/CT and then decided the management of the patients. We defined the impact of [68Ga]Ga-PSMA-11 PET/CT as any change in management decided by clinicians during the multidisciplinary meeting triggered by [68Ga]Ga-PSMA-11 PET/CT.

The treatment carried out on the patient was considered to be effective if the PSA declined by more than 50% (compared to the baseline value) following treatment modification or if the PSA remained stable (maximum variation of 10% compared to baseline) on at least 2 assays performed at least 3 weeks apart when surveillance was decided [19].

#### *2.5. Standard of Truth*

Existence of a PCa lesion was established for each patient according to a composite standard of truth (SOT) based on all clinical data that were available during the followup period: histological findings, results of other imaging, follow-up imaging and PSA evolution. Histological findings when available were considered as the strongest criteria. When histological confirmation was not available, SOT criteria were as follows:


#### *2.6. Statistical Analysis*

Data were analyzed using the IBM SPSS software. We considered a *p*-value less than 0.05 to be statistically significant. We performed logistic regression to search for a relationship between [68Ga]Ga-PSMA-11 PET/CT positivity (patient-based and per anatomical site) and initial ISUP grade group, initial d'Amico group risk, PSA in surgical patients (closest assay to the [68Ga]Ga-PSMA-11 PET/CT), PSA doubling time in months, and PSA velocity in ng/mL/year in all patients. Receiver operating characteristic (ROC) analysis was used to determine area under the curve and cut-off values of PSA parameters in relation to [68Ga]Ga-PSMA-11 PET/CT positivity. Comparisons of [68Ga]Ga-PSMA-11 PET/CT detection rates (at least one suspicious abnormality suggestive of PCa) and impact on patient management to parameters in relation with [68Ga]Ga-PSMA-11 PET/CT positivity were performed by chi-squared test, a Fisher's exact test, or a Student's *t*-test according to the type of variable. Detection rates and accuracies between the two PET/CT scanners were compared by chi-squared test. Therapy efficacies when guided or not by [68Ga]GaPSMA-11 PET/CT imaging were compared via Fisher's exact test. The agreement between retrospective masked and routine unmasked [68Ga]Ga-PSMA-11 PET/CT readings, overall and per anatomical site, were assessed using Cohen's kappa coefficient (0–0.20: very weak; 0.21–0.40: weak; 0.41–0.60: moderate; 0.61–0.80: strong; 0.81–1.0: very strong).

#### **3. Results**

#### *3.1. BCR Patient Characteristics*

Between June 2016 and November 2018, 294 consecutive patients who met the inclusion criteria were retrospectively included (Table 1). No eligible patient was excluded. One-hundred and ninety-three patients were presenting with first PCa BCR among whom 159 had prostatectomy.

**Table 1.** Patient characteristics.


\* Evaluated on 285 patients; 95% confidence intervals are presented between brackets.

The mean time from PCa diagnosis to the first BCR was 42 months [95%CI: 37–46], longer for ISUP 1–2 patients (49 months [95%CI: 43–56]) than for ISUP 3–5 patients (33 months [95%CI: 28–39]) (*p* < 0.001; Student *t* test).

Sixteen patients were considered lost to follow-up (no follow-up data available). The median duration of follow up after [68Ga]Ga-PSMA-11 PET/CT for the 278 assessable patients was 17 months [95%CI: 14–19]. A patient-based SOT was feasible in 176 patients (60%) among whom histological confirmation was available in 27 patients.

## *3.2. [68Ga]Ga-PSMA-11 PET/CT Positivity Rates and Performance*

On the 294 patients, 140 (48%) where scanned on the Gemini TF16 and 154 (52%) on the Biograph CTflow.

At least one abnormal focus was found in 237 patients (81%) on routine unmasked reading and in 229 patients (78%) on masked retrospective reading. The overall and per anatomical site detection rates, irrespective of PSA, for readings, as well as the SUVmax of detected foci, are presented in Table 2.

**Table 2.** [ <sup>68</sup>Ga]Ga-PSMA-11 PET/CT positivity rates in prostate cancer patients investigated due to a biochemical recurrence (irrespective of total prostate-specific antigen serum values). Results of routine unmasked and retrospective masked readings, both by anatomical site and overall, are presented. Median maximum standard uptake values (SUVmax) per anatomical site are presented with their range brackets. Agreement was evaluated with Cohen's kappa coefficient κ.


\*: 7 carcinomatosis, 7 pleura/lung, 2 testis, 1 liver, 1 penile, 1 intramedullary spinal, 1 rectal. \*\*: 4 carcinomatosis, 3 liver, 3 testis, 1 penile, 1 pancreas.

> Based on the masked retrospective reading results, we identified a relationship between [68Ga]Ga-PSMA-11 PET/CT detection rates and PSA serum level in surgical patients: 37%, 56%, 71%, and 87% for PSA serum levels of 0.20–0.49 ng/mL (*n* = 57), 0.50–0.99 ng/mL (*n* = 45); 1.00–1.99 ng/mL (*n* = 59), and ≥2.00 ng/mL (*n* = 91), respectively, if equivocal findings were considered negative for malignancy; and 59%, 67%, 81%, and 88%, respectively, if equivocal findings were considered positive for malignancy.

> From the ROC analysis, the best cut-off value of PSA to perform a [68Ga]Ga-PSMA-11 PET/CT in surgical patients (*n* = 252) was 1 ng/mL (area under curve of 0.59), whether considering equivocal findings as positive or negative for malignancy.

> Overall, the mean PSA serum value of patients with negative [68Ga]Ga-PSMA-11 PET/CT was significantly lower than that of the patients with positive examinations, whether considering equivocal findings as positive for malignancy (1.4 vs. 3.8 ng/mL: *p* = 0.03) or negative for malignancy (1.8 vs. 3.9 ng/mL; *p* = 0.03).

> A relationship was found between [68Ga]Ga-PSMA-11 PET/CT positivity on bone and initial ISUP grade group. However, the rate of bone foci positive was statistically higher for ISUP 3–5 patients than for ISUP 1–2 patients only when considering equivocal findings as negative for malignancy (14% vs. 25%; *p* = 0.03 ).

> We did not find any relationship between [68Ga]Ga-PSMA-11 PET/CT positivity and the other tested parameters.

Furthermore, we did not identify a difference in [68Ga]Ga-PSMA-11 PET/CT positivity between patients presenting with a first BCR and patients presenting with a second or third episode of BCR (*p* = 0.1).

According to [68Ga]Ga-PSMA-11 PET/CT results (retrospective masked reading results, equivocal findings considered positive for malignancy), 65 patients (22%) had no detectable disease (60 surgeries), 34 patients (12%) presented an isolated focus in the prostate/prostatic lodge (24 surgeries), 132 patients (45%) were categorized oligo-3 (116 surgeries), 142 (48%) oligo-4 (125 surgeries), 149 (51%) oligo-5 (130 surgeries), and 46 (16%) had more than five distant malignant foci (37 surgeries).

The dynamic images acquired on the pelvis provided an additional diagnostic information in 6/294 patients (2%), all of whom had surgery, since an abnormal focus in the prostatic lodge was detected on this acquisition but masked by the physiologic urinary uptake in the bladder on the 60 to 90 min after injection PET acquisition.

The overall and per anatomical site diagnostic performances of [68Ga]Ga-PSMA-11 PET/CT based on the 176 patients on whom a SOT was feasible are presented in Table 3.

**Table 3.** [ <sup>68</sup>Ga]Ga-PSMA-11 PET/CT performances in prostate cancer patients investigated due to a biochemical recurrence (irrespective of total prostate-specific antigen serum values). Results with equivocal findings considered positive for malignancy and with equivocal results negative for malignancy are both presented. Patient-based and region-based analyses with the number of cases on which the standard of truth was feasible.


We found no differences in detection rates or in accuracies between the two scanners.

## *3.3. Impact of [68Ga]Ga-PSMA-11 PET/CT on BCR Patient Management and Therapy Efficacy Prediction*

The impact of [68Ga]Ga-PSMA-11 PET/CT was assessable for the 278 patients for whom follow-up data were available. Patient disease management changed in 189/278 (68%) cases, and [68Ga]Ga-PSMA-11 PET/CT impacted this change in 162 cases (86%), 21 being minor changes (Table 4).



imaging on the prostate) and a solitary fibrous tumor (false positive of imaging on viscera).

This impact was statistically higher when PSA was greater than 1 ng/mL in surgical patients (97/145 = 67% vs. 43/95 = 45%; *p* < 0.001), but only when PSA was superior to 2 ng/mL in non-surgical patients (21/31 = 68% vs. 1/7 = 14%). Overall, the impact was statistically higher when PSA doubling time was less than one year (113/176 = 65% vs. 46/95 = 48%: *p* = 0.01), and tended to be higher for ISUP 3–5 patients (83/131 = 63% vs. 77/145 = 53%; *p* = 0.08).

The therapy efficacy was assessable for 257 patients (87%) for whom sufficient followup data were available. Treatments with curative intent consisted in 107 radiation therapies focused on abnormalities detected by [68Ga]Ga-PSMA-11 PET/CT, 12 salvage lymphadenectomies, two focal irreversible electroporations, one cryosurgery, and one left orchidectomy (isolated CaP metastasis of the testis histologically proven). Eleven patients with negative [68Ga]Ga-PSMA-11 PET/CT were treated by radiation therapy of the prostatic lodge in accordance with the guidelines for salvage radiation therapy after prostatectomy [20]. ADT was started for 67 patients among whom three benefited from novel androgen axis drugs (such abiraterone or enzalutamide). Surveillance was finally decided for 75 patients, while it was only indicated in 70 after multidisciplinary meeting, as five patients, in whom PSA presented a long doubling time, refused the offered treatment.

The treatment carried out on the patient was considered effective according to the defined criteria in 78% (200/257) of patients overall, 89% (138/155) when guided by [ <sup>68</sup>Ga]Ga-PSMA-11 PET/CT versus 61% (62/102) when not guided by [68Ga]Ga-PSMA-11 PET/CT (*p* < 0.001). It was considered effective in 84% (112/133) when a treatment with curative intent was performed, 94% (60/64) when ADT was started, and 47% (28/60) when surveillance was decided. The treatment carried out on the patient was considered effective in 85%, 86%, and 87% when a treatment with curative intent was performed in oligo-3, oligo-4, and oligo-5 patients, respectively.

## *3.4. Agreement between Routine Unmasked and Retrospective Masked [68Ga]Ga-PSMA-11 PET/CT Readings*

The agreements between routine unmasked and retrospective masked readings are presented in Table 2. Overall agreement was strong (k = 0.68). The agreement was moderate for the prostate/prostatic lodge (k = 0.54) and viscera (k = 0.56), strong for lymph nodes above the diaphragm (k = 0.73) and bone (k = 0.74), and very strong for pelvic lymph nodes (k = 0.90) and paraaortic lymph nodes (k = 0.84). Both readings were similar in 92% (272/294) of cases. In the 22 cases for which readings were different, the findings on masked readings were considered more accurate according to the follow-up in 6% (17/294: 10 on prostate/prostatic lodge, 2two on bone, two on lymph nodes above the diaphragm, one on pelvic lymph nodes, one on paraaortic lymph nodes, and one on lung) versus 2% (5/294: three on the prostate/prostatic lodge and two on pelvic lymph nodes) for routine reading findings.

#### **4. Discussion**

## *4.1. [68Ga]Ga-PSMA-11 PET/CT Performances*

PSMA-11 labelled with gallium-68 a is the most studied ligand for imaging of PCa, especially patients with BCR, for whom detection rates were largely reported, but this study is one of the largest series presenting [68Ga]Ga-PSMA-11 PET/CT performances (i.e., sensitivity, specificity, and accuracy), based on a composite SOT. In the present study, [ <sup>68</sup>Ga]Ga-PSMA-11 PET/CT demonstrated an overall positivity rate of 78% for restaging PCa patients with BCR, which is consistent with that of 76% reported in a recent metaanalysis [21]. Our per anatomical site positivity rates were also in agreement with an updated version of this meta-analysis [13], except for the extrapelvic lymph nodes positivity rate as we chose to analyze this location in two separate areas (paraarotic lymph nodes and lymph nodes above the diaphragm). We decided to do this because patient management may significantly differ for PCa recurrence between these locations. We also confirmed the relationship between [68Ga]Ga-PSMA-11 PET/CT positivity rate and PSA serum levels

that has already been reported several times [22–24], observing positivity rates in surgical patients comparable to those reported by Perera et al. [13].

In our study, we established a composite SOT for PCa on 176 patients overall and at least a hundred patients for each anatomical area. As a histological confirmation of the detected abnormalities was only available in 27 patients, the SOT was primarily based on clinical follow-up including the PSA response to targeted treatment for imaging findings during a median follow-up period of 17 months. Using those criteria, we found overall sensitivity and specificity of both 70%, lower than that of 86% reported by Perera et al. [21], which were only based on histopathologic correlation with <sup>68</sup>Ga-PSMA PET/CT abnormal findings and, therefore, did not take into account anatomical areas without abnormal PSMA uptake. [68Ga]Ga-PSMA-11 PET/CT accuracies for anatomical areas were above 90%, except for the prostate/prostatic lodge area, for which image reading was impaired by the urinary physiologic uptake in the bladder. To improve reading in this location, we performed dynamic PET acquisition over the pelvis immediately after [68Ga]Ga-PSMA-11 injection as it was published that this imaging sequence increases the detection rate of local recurrence [25]. In our study these early images provided additional information in 2% (6/294) of patients, all of whom received surgery, by detecting pathological foci in prostate lodge that were then masked by the urinary physiologic uptake in the bladder during the 60 min post-IV PET images. Considering these results, adding dynamic PET images to [ <sup>68</sup>Ga]Ga-PSMA-11 PET/CT imaging protocol should be considered for surgical patients.

## *4.2. [68Ga]Ga-PSMA-11 PET/CT's Impact on PCa Management and Therapy Efficacy Prediction*

We found that [68Ga]Ga-PSMA-11 PET/CT impacted patient disease management in 58% of cases, resulting in an increased proportion of treatments with curative intent. These findings are consistent with the 54% reported by a recent meta-analysis [14]. Moreover, we found that [68Ga]Ga-PSMA-11 PET/CT's impact was significantly higher when PSA was greater than 1 ng/mL in surgical patients but only when PSA was greater than 2 ng/mL in non-surgical patients. All patients considered, the impact was significantly higher when PSA doubling time was less than one year and tended to be higher for ISUP 3–5 patients.

In this study, the treatment carried out on patient was considered effective in 91% of cases when guided by [68Ga]Ga-PSMA-11 PET/CT versus 40% when not guided by [ <sup>68</sup>Ga]Ga-PSMA-11 PET/CT (*p* < 0.001). We previously reported similar findings in a small cohort of 30 castration-resistant PCa patients restaged by [68Ga]Ga-PSMA-11 PET/CT [26].

## *4.3. [68Ga]Ga-PSMA-11 PET/CT Reading Agreement*

In our study we analyzed [68Ga]Ga-PSMA-11 PET/CTs according to a three-point scale to introduce diagnostic uncertainty in imaging reading, as pitfalls and equivocal findings are inseparable from any diagnostic procedure. Thus, we found equivocal results in approximately 10% for both routine unmasked and retrospective masked readings, which is relatively low. We also noted that the proportion of equivocal findings decreased when PSA increased. Systematic approaches to the interpretation of PSMA imaging studies, using a five-point scale, were recently proposed to classify imaging findings and better reflects the likelihood of the presence of PCa [27,28]. However, we could not use those approaches for the routine readings that were already performed and chose not to use it for the retrospective readings as we wanted to ensure that evaluation between readings would be comparable. Furthermore, we could not use the 2021 EANM standardized guidelines for PSMA-PET as we conducted this research in 2019–2020.

We found an overall strong agreement between [68Ga]Ga-PSMA-11 PET/CT routine unmasked and retrospective masked readings (k = 0.68), which is comparable to a previously report on a more heterogeneous series of 50 patients (k = 0.62) [29].

Agreement was moderate (k = 0.54) for the prostate/prostatic lodge, slightly lower than previously reported (k = 0.62) [29], likely because our series reports a large proportion of surgical patients. The reading in the prostate/prostate bed is impaired by the physiological uptake of the urine in the bladder, especially in operated patients. Dynamic images aim

to improve the reading in the prostate/prostatic lodge. The unmasked reader may look less attentively at the dynamic images because he relies on the clinical data, such as mpMRI or PSA serum value or doubling time that may influence its diagnosis (more equivocal findings in the prostate/prostatic lodge). The experienced masked reader relies only on the dynamic images to improve its reading in the prostate bed.

We found that agreement was strong to very strong for all lymph nodes areas (k = 0.73 for lymph nodes above the diaphragm, k = 0.84 for paraaortic lymph nodes, and k = 0.90 for pelvic lymph nodes), similar to previously reported values of k = 0.74, considering all lymph node areas [29]. However, we distinguished invasion of the lymph nodes between pelvic and paraaortic regions and above the diaphragm, as we assumed that therapeutic management for involved lymph nodes differed between these areas. We found a strong agreement in readings for bone (k = 0.74), which is comparable to that previously reported [29]. Finally, we found a moderate agreement in readings for viscera (k = 0.56), likely due to the low incidence of visceral metastases in patients with BCR [30] and to atypical locations (like penile, testis, intramedullary spinal cord) or challenging location for imaging (30% of visceral lesions were peritoneal carcinomatosis) [31,32].

We assumed that the disagreements between readings might be explain by the higher number of equivocal foci found by the routine unmasked reading. Indeed, the knowledge of clinical parameters such as PSA serum level or doubling time may influence the reading.

#### *4.4. Limitations*

This study has several limitations. The primary one, shared by most imaging studies addressing the search for metastatic disease, is the lack of sufficient histological proof for most of the suspected metastases, which were primarily characterized based on follow-up data. Indeed, obtaining a histopathological evidence for asymptomatic and possibly benign lesions, or locations that are negative on imaging, is ethically questionable and hardly feasible in practice. We chose to base our SOT on the variation of PSA, excluding patients with a change in their ADT regimen after [68Ga]Ga-PSMA-11 PET/CT and on histological findings, if available. In this work, a SOT was feasible for 60% of patients, with a 17-month median duration of follow-up, which allowed us to calculate overall performances of [ <sup>68</sup>Ga]Ga-PSMA-11 PET/CT, as well as performances per anatomical areas. Furthermore, we were able to determine from the ROC analysis that the best cut-off values of PSA level to perform a [68Ga]Ga-PSMA-11 PET/CT in operated patients was 1 ng/mL. These findings need to be confirmed by other large studies, as most available data report <sup>68</sup>Ga-PSMA PET/CT detection rates [30,33,34].

The second major limitation of this work was its retrospective design. However, this study is one of the larger homogenous cohorts of BCR patients which has evaluated imaging performance based on a composite SOT.

In this work, we reported [68Ga]Ga-PSMA-11 PET/CT performance according to PSA serum levels which were not evaluated the day of the PET, in the same laboratory, but corresponded to the value of the closet assays to the examinations. Thus, we assume that we may have overestimated detection rates, especially for low PSA levels.

In this work, [68Ga]Ga-PSMA-11 PET/CT acquisition time varied from 60 to 90 min after [68Ga]Ga-PSMA-11 injection, which may be important as increased lesion detection was reported with delayed imaging times up to four hours [35]. However, acquisition times were within the acceptable range of 50 to 100 min that is recommended by the current guidelines for <sup>68</sup>Ga-PSMA PET/CT [35]. Thus, we assume that the limited variation in acquisition times in our study did not significantly affect the results.

Finally, another notable feature is that all patients in our study who underwent [ <sup>68</sup>Ga]Ga-PSMA-11 PET/CT had previously show no sign of distant metastases at [18F]fluorocholine PET/CT and were referred for this reason to [68Ga]Ga-PSMA-11 PET/CT based on the French regulation for compassionate use of pharmaceutical, which is authorized on an individual basis by the National Medicine Agency. Therefore, our study population may not reflect that of other international studies, and results may differ due to this selection method. However, we found comparable positivity rates than that reported on larger series and metanalyses [13,36].

Because of these limitations, the promising performances and impact rate on patient disease management of [68Ga]Ga-PSMA-11 PET/CT need to be confirmed in a larger prospective study.

#### **5. Conclusions**

In conclusion, [68Ga]Ga-PSMA-11 PET/CT demonstrated reliable performance in locating recurrence sites of prostate cancer and motivated disease management changes in almost two out of three patients. Those performances and impact rates were better when PSA serum level were above 1 ng/mL.

Comparison between routine unmasked and retrospective masked readings demonstrated that this imaging modality is highly reproducible, especially for the detection of pelvic and paraaortic lymph nodes.

The use of PSMA radioligands with PET/CT should be considered, when available, as a first line imaging modality for biochemical recurrence of prostate cancer.

**Author Contributions:** Conception and design: M.G.; acquisition of data: M.G., A.F., L.L. and G.C.; analysis and interpretation of data: M.G., A.F. and L.L.; drafting the manuscript: M.G., A.F., Y.B., A.d.l.T., O.C., G.C. and G.F.; critical revision of the manuscript: M.G., Y.B., A.d.l.T., O.C. and G.F.; statistical analysis: M.G.; technical and material support: M.G. and T.R. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** According to French regulation, the approval of an institutional review board was not necessary for performing this retrospective analysis of already available data.

**Informed Consent Statement:** According to French regulation, the written consent was not necessary for performing this retrospective analysis of already available data.

**Data Availability Statement:** The data that support the findings of this study are available from the corresponding author, M.G., upon reasonable request.

**Acknowledgments:** The authors are grateful to the following physicians for their confidence in referring their patients: Chauveinc, Radiothérapie—Institut de radiothérapie Hartmann, Levallois-Perret; Simon, Radiothérapie—Hôpital de la Pitié-Salpétrière, Paris; Quero, Radiothérapie—Hôpital Saint-Louis, Paris; Graff Caillaud, Radiothérapie—IUCT Oncopole, Toulouse.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**

