Liquid Biopsy Landscape in Patients with Primary Upper Tract Urothelial Carcinoma

Simple Summary The standard of care for patients diagnosed with upper tract urothelial carcinoma (UTUC) is surgery. However, the current standard tools for the diagnosis and preoperative risk stratification of primary UTUC leave much to be desired. Using a simple blood sample, we can detect rare events, such as circulating tumor cells (CTCs) and large extracellular vesicles (LEVs), which may reveal biomarkers indicative of disease and provide insight into progression. This study demonstrates a wide variety of CTCs and LEVs that are identifiable in the blood of patients with UTUC at higher counts compared to normal donors. The data presented indicate that a blood-based liquid biopsy can detect UTUC and help guide clinical decisions. Abstract Urothelial carcinomas (UCs) are a broad and heterogeneous group of malignancies, with the prevalence of upper tract urothelial carcinoma (UTUC) being rare, accounting for only 5–10% of total malignancies. There is a need for additional toolsets to assist the current clinical paradigm of care for patients with UTUC. As a non-invasive tool for the discovery of cancer-related biomarkers, the liquid biopsy has the potential to represent the complex process of tumorigenesis and metastasis. Herein, we show the efficacy of the liquid biopsy as a source of biomarkers for detecting UTUC. Using the third-generation high-definition single-cell assay (HDSCA3.0) workflow, we investigate liquid biopsy samples collected from patients with UTUC and normal donors (NDs) to provide critical information regarding the molecular and morphological characteristics of circulating rare events. We document several important findings from the liquid biopsy analysis of patients diagnosed with UTUC prior to surgery: (1) Large extracellular vesicles (LEVs) and circulating tumor cells (CTCs) are detectable in the peripheral blood. (2) The rare-event profile is highly heterogeneous. (3) Clinical data elements correlate with liquid biopsy analytes. Overall, this study provides evidence for the efficacy of the liquid biopsy in understanding the biology of UTUC with the future intent of informing clinical decision making, ultimately improving patient outcomes.


Introduction
Urothelial carcinomas (UCs) are a broad and heterogeneous group of malignancies that are among the most frequently occurring tumor indications in Europe and the United States [1]. Within these, the prevalence of upper tract urothelial carcinoma (UTUC) is rare, accounting for only 5-10% of total malignancies and an annual incidence of 2 cases per 100,000 individuals [2]. UTUCs are classified as non-muscle-invasive (Tis, Ta, T1) or muscle-invasive (T2-T4), indicative of tumor invasion into the muscularis, peripelvic fat, renal parenchyma, periureteric fat, and/or other organs [3]. Approximately 50-60% of cases present as muscle-invasive at diagnosis [4]. While low-grade tumors are unlikely to spread from the kidney or ureter, high-grade UTUCs have a higher potential to recur elsewhere in the body. Bladder recurrence occurs in 22-47% of patients, while approximately 17% of patients manifest concomitant bladder cancer [5,6].
The current standard tools for the diagnosis and preoperative risk stratification of primary UTUC include urine cytology, cross-sectional imaging, and ureteroscopic (URS) biopsy, the procedure that informs the need for kidney-sparing surgery (KSS) for low-risk disease and radical nephroureterectomy (RNU) for high-risk disease [7]. While minimally invasive, URS biopsy leaves more to be desired in the accurate matching of biopsy grade and final pathology (66% in low-grade and 97% in high-grade tumors) [8]. Additionally, under-grading and under-staging rates from URS biopsy remain a concern, at 32% and 46%, respectively [8]. Furthermore, analyses show that, for patients with UTUC who undergo RNU, concordance between URS biopsy clinical staging and RNU pathological staging was found to be 34.5%, underscoring the significant discrepancy between clinical stage and final pathology [9]. Moreover, URS biopsy is associated with a higher rate of intravesical recurrence following RNU [10,11]. Taken together, such evidence necessitates the need for additional toolsets to assist the current clinical paradigm of UTUC patient care. Herein, we discuss the efficacy of the liquid biopsy as a source of biomarkers for detecting UTUC and informing treatment options to improve patient outcomes.
The detection and enumeration of circulating tumor cells (CTCs) via the liquid biopsy, typically accomplished through the collection of peripheral blood (PB), are deemed viable prognostic markers for relapse [12][13][14]. A study by Claps et al. illustrated that a downward trend of CTCs present in patient liquid biopsy was indicative of progression-free survival and increased overall survival [15]. Similarly, a meta-analysis by Zhang et al. investigating the relationship between UTUC and the presence of CTCs in the liquid biopsy showed that the presence of CTCs was indicative of negative patient outcomes with regard to tumor staging and metastasis [14]. These studies support the assertion that CTCs found in PB are associated with poorer patient prognosis. The current study will make use of the thirdgeneration liquid biopsy workflow for the systematic visualization and characterization of CTCs and large extracellular vesicles (LEVs) present in the PB collected from patients newly diagnosed with primary UTUC prior to surgical resection.
The third-generation high-definition single-cell assay (HDSCA3.0) workflow provides critical information regarding the molecular and morphological characteristics of rare events detected in the liquid biopsy with immunocytochemistry [16] coupled with downstream molecular characterization to deliver diagnostic pathology-quality data for clinical decision making [17][18][19][20][21][22]. Through enumeration, events in a sample, primarily CTCs and LEVs, can be discriminated and then quantified based on size, morphology, DNA load, and phenotype [23], providing a comprehensive summary of the disease state. The primary objective of this study was to investigate the prognostic relationship between CTCs and disease state in the PB samples collected from patients diagnosed with UTUC taken prior to RNU. In the assessment of this relationship, the presence of CTCs in reference to clinical data metrics will also be examined, namely, clinical and pathological staging as well as histological subtype. Overall, this study aims to provide evidence for the efficacy of the liquid biopsy in further understanding UTUC with the future intent to predict metastatic relapse post-surgery, ultimately improving patient outcomes.

Study Design
This was a single institution prospective study of patients diagnosed with primary UTUC in which PB samples were collected prior to surgery. Eligible patients underwent surgical removal of the primary tumor with curative intent. Samples were collected from the University of Southern California's Keck School of Medicine between May 2021 and March 2022. Prospectively collected clinical, radiologic, and pathologic data elements, as well as a limited amount of follow-up data, were also provided. Recurrence was defined as any clinical recurrence, the majority shown radiologically, either symptomatic or not. Patient recruitment took place according to institutional review board's approved protocols (IRB no. HS-19-00886 (liquid biopsy collection) and no. HS-17-00168 (clinical and demographic data collection)), and all study participants provided written informed consent. Here, we present the liquid biopsy analysis from a total of 20 patients diagnosed with UTUC. Additionally, 50 normal donor (ND) blood samples from individuals with no known pathology were procured from Epic Sciences (San Diego, CA, USA).

Blood Sample Processing
As previously described, PB blood samples were collected in 10 mL blood collection tubes (Cell-free DNA BCT, Streck, La Vista, NE, USA) within 24-48 h prior to processing by the Convergent Science Institute in Cancer (CSI-Cancer) at the University of Southern California [16]. Samples then underwent brief red blood cell lysis in isotonic ammonium chloride solution, after which the entire nucleated cell fraction was plated on custom glass slides (Marienfeld, Lauda, Baden-Württemberg, Germany) at approximately 3 million cells per slide. Slides were blocked with 7% BSA and dried prior to cryostorage at −80 • C and subsequent rare-event analysis.
Negative and positive controls were incorporated as contrived samples. The negative control was the ND blood, and the positive control was ND blood spiked with cell lines with a known expression profile, as previously described [16]. These controls were processed according to standard operating procedures and were used throughout the HDSCA workflow.

Rare-Event Detection and Classification
As previously reported [24,25], the detection and classification of rare events employed OCULAR (Outlier Clustering Unsupervised Learning Automated Report), a custom computational methodology designed to detect rare cell and LEV candidates [23]. A total of 761 morphometric parameters were provided for each event based on nuclear and cytoplasmic morphology, as well as biomarker expression (CK, Vim, CD45/CD31) from the 4-channel immunofluorescence assay (DAPI, AlexaFluor ® 488, AlexaFluor ® 555, AlexaFluor ® 647).
Manual reporting was completed on computationally identified events to confirm signal intensity and distribution, as well as morphology. Images of candidate rare events were presented to a hematopathologist-trained technical analyst for analysis and interpretation, in which rare events were classified into 12 categories (8 cellular, 4 LEV) based on the 4 fluorescent channels, creating a combination of biomarker expressions. Cells defined by a distinctly appearing nucleus by DAPI morphology presenting as CK-positive, Vim-negative, and CD45/CD31-negative were classified as epithelial-like CTCs (epi.CTCs). Epi.CTCs expressing Vim were further specified as mesenchymal-like CTCs (mes.CTCs).
To contextualize the number of rare events identified within PB samples, automatically determined white blood cell (WBC) counts (Medonic M-series Hematology Analyzer, Clinical Diagnostic Solutions Inc., Fort Lauderdale, FL, USA) were used in conjunction with the number of WBCs detected by the assay per slide to calculate the amount of blood analyzed per test. As such, results of rare-event detection and classification are presented as fractional values of events/mL.
The identification of LEVs closely followed the OCULAR methodology outlined above; candidates were CK-positive with variable Vim and CD45/CD31 expression. Additional consideration was given in manually classifying LEV candidates as either free-floating or in close proximity to other cells. Further corrections during the manual classification of LEVs included the exclusion of bubbles, halos, or light refractions resembling the round and membranous morphology of LEVs in the examination of CK imaging frames. A maximum threshold of three LEVs per frame was used to exclude CK-positive debris that may have been introduced onto the slide during processing and/or staining.

Statistical Analysis
To calculate the correlations between clinical and demographic data and the rare events detected in the liquid biopsy samples, the Spearman's rank correlation [26] and Mann-Whitney U test were applied, also known as the Wilcoxon rank-sum test [27,28]. For continuous clinical variables and variables with ordinal encoding, the Spearman's rank correlation was used. Furthermore, for all categorical variables, the Wilcoxon rank-sum test was used for every variable pair. Statistical significance was determined at a p-value ≤ 0.05. The two statistical tests used were non-parametric tests, which matched the conditions of the variables used. The analysis performed herein is similar to a sister study of bladder cancer patients [25]. All computational analyses and visualizations were performed in python (version 3.8.5, https://www.python.org/, accessed on 11 May 2022) utilizing the scipy (version 1.5.0, https://scipy.org/, accessed on 11 May 2022), scikit-learn (version 0.23.2, https://scikit-learn.org/stable/, accessed on 11 May 2022), and matplotlib (version 3.2.2, https://matplotlib.org/, accessed on 11 May 2022) packages.
For visualizing the variation of morphological parameters of detected cellular events and to uncover their variation within the channel-based classification, a two-dimensional tSNE (t-distributed stochastic neighbor embedding) was used [29]. Along with the tSNE, a clustering algorithm was used to identify cellular subgroups beyond the channel-type classifications. An agglomerative clustering algorithm, imported from the sklearn library version 0.23.2, https://scikit-learn.org/stable/ (accessed on 11 May 2022) [30], was applied utilizing Ward's linkage and a Euclidean distance metric [31].
Negative and positive controls were incorporated as contrived samples. The negative control was ND blood, and the positive control was the normal blood donor spiked with cell lines with a known expression profile (i.e., SKBR3 or HPAEC). These controls were processed according to standard operating procedures and included in the immunofluorescence staining, scanning, and rare-event detection.

Results
A total of 20 patients with primary UTUC were accrued for this study, each providing a single PB sample obtained prior to surgery. Demographic and clinical data metrics were collected and are provided in Table 1. The demographic information for the NDs was limited to the individuals' ages (median 57; range 45-82; mean 58.9 years).

Liquid Biopsy Analysis Prior to Surgery
The complete blood cell count was measured prior to blood processing at CSI-Cancer. For all samples in this study, the average cell count per frame was 1430.80 ± 343.85 (UTUC: 1326.20 ± 344.42; ND: 1472.64 ± 343.62). For the 20 UTUC samples included here, there was a median WBC count of 6.65 (range 2.8-9.8; mean 6.01) million cells/mL PB. Across all UTUC samples, total rare-event detection had a median of 178.23 events/mL (range 25.66-779.80; mean 234.88). For ND samples, total rare-event detection had a median of 38.50 events/mL (range 4.39-141.55; mean 47.86). A significant difference in the total rare-event (i.e., cells and LEVs) count/mL was observed between the patients diagnosed with UTUC and the NDs (p-value < 0.0001).

Rare-Cell Characterization
We identified eight cellular categories that were defined by a nuclear DAPI signal and depended on the different biomarker expression profiles. Figure 1 shows a gallery of CTCs, while the graphical representations of the enumeration and frequency of each rare event identified per test for each patient sample are shown in Figure 2, respectively. Total rare-cell detection for the UTUC samples had a median of 169.85 cells/mL (range 18.62-771.63; mean 219.83). The ND samples presented with a median rare-cell detection of 34.46 cells/mL (range 4.39-137.03; mean 43.21). A statistically significant difference in total rare-cell detection was observed between the UTUC patients and ND samples (p-value < 0.0001).    Epi.CTCs were detected with a median of 0 cells/mL (range 0-5.90; mean 1.30), while mes.CTCs were detected with a median of 0 cells/mL (range 0-17.00; mean 2.24) in the UTUC samples. There was no statistically significant difference between the UTUC and ND samples in epi.CTCs/mL or mes.CTCs/mL observed. Eleven patients (55%) did not present with epi.CTCs or with mes.CTCs. Only seven (35%) patients overlapped in presenting with 0 cells/mL for both epi.CTCs and mes.CTCs.
Additional  Figure 2C). The biological significance of these cellular populations has yet to be determined. Enumerations of all rare events in both UTUC and ND samples can be found in Supplementary Table S1.
The most prevalent cell types detected in the PB of patients with UTUC were Vim|CD45/ CD31 (median 28. The cellular morphometrics were used to further investigate the heterogeneity of the rare-cell populations detected in the PB of patients with UTUC. A dimensionality reduction algorithm and a clustering algorithm were applied to visualize the cellular heterogeneity in a two-dimensional plane. The morphometrics considered included the area and the eccentricity of the cells and the nucleus, respectively. Furthermore, the median intensity of the four channels was also used as input. The resulting cellular tSNE, depicted in Figure 3A,B, is color-coded according to channel-type classification and cluster number, respectively. Cellular area and eccentricity, along with the median channel intensities, are plotted in Figure 3C-G. For example, the cellular heterogeneity within classifications is evident for mes.CTCs, with the appearance of two peaks in the distributions for cell eccentricity, median CK, and Vim expression. The tSNE depicted in Figure 3 is a visualization of the morphological relationships between channel classifications, with each point depicting a single cellular event. The cells split into clusters, with morphological heterogeneity driving separation within channel classifications. For instance, the cells in the Vim|CD45/CD31 classification are divided between clusters 1, 2, 3, and 5. Cluster 1 contains the majority of Vim|CD45/CD31 cells, along with CK|Vim|CD45/CD31 cells and DAPI only cells. Cluster 2 includes the majority of Vim only cells and a subset of Vim|CD45/CD31 cells. In addition, cluster 3 contains a mixture of Vim only, DAPI only, and Vim|CD45/CD31 cells. Cluster 5 has cells across the DAPI only, Vim only, and Vim|CD45/CD31 categories. Finally, clusters 4 and 6 contain the remaining CK-positive cells.

LEV Detection
LEVs were classified by a lack of DAPI signal and positive CK signal, as well as morphology. Total LEV detection for the UTUC samples was a median of 8.33 LEVs/mL (range 2.97-54.94; mean 15.05; Figure 2). The ND samples presented with a median of 3.34 LEVs/mL (range 0-27.91; mean 4.65). There was a significant difference in LEVs/mL detected between the ND and UTUC samples (p-value < 0.0001). In the UTUC samples, LEVs were detected either alone (n = 118; 50.9%) or in close proximity to cells (n = 114; 49.1%). In the ND samples, these LEV populations totaled 85 (45.9%) and 100 (54.1%), respectively.
CK only LEVs were detected in 19 UTUC samples (95%), with a median of 4.34 LEVs/mL (range 0-54.94; mean 10.36). CK|CD45/CD31 LEVs were also detected in 11 patients (55%), with a cohort median of 1.04 LEVs/mL (range 0-25.08; mean 4.05). The most The tSNE depicted in Figure 3 is a visualization of the morphological relationships between channel classifications, with each point depicting a single cellular event. The cells split into clusters, with morphological heterogeneity driving separation within channel classifications. For instance, the cells in the Vim|CD45/CD31 classification are divided between clusters 1, 2, 3, and 5. Cluster 1 contains the majority of Vim|CD45/CD31 cells, along with CK|Vim|CD45/CD31 cells and DAPI only cells. Cluster 2 includes the majority of Vim only cells and a subset of Vim|CD45/CD31 cells. In addition, cluster 3 contains a mixture of Vim only, DAPI only, and Vim|CD45/CD31 cells. Cluster 5 has cells across the DAPI only, Vim only, and Vim|CD45/CD31 categories. Finally, clusters 4 and 6 contain the remaining CK-positive cells.

LEV Detection
LEVs were classified by a lack of DAPI signal and positive CK signal, as well as morphology. Total LEV detection for the UTUC samples was a median of 8.33 LEVs/mL (range 2.97-54.94; mean 15.05; Figure 2). The ND samples presented with a median of 3.34 LEVs/mL (range 0-27.91; mean 4.65). There was a significant difference in LEVs/mL detected between the ND and UTUC samples (p-value < 0.0001). In the UTUC samples, LEVs were detected either alone (n = 118; 50.9%) or in close proximity to cells (n = 114; 49.1%). In the ND samples, these LEV populations totaled 85 (45.9%) and 100 (54.1%), respectively.

Correlation with Clinical Data
Correlation analysis was used to determine the relationship between the clinical and demographic metrics collected for the patients with UTUC (n = 20) and the various liquid biopsy analytes detected. Here, we report only the significant correlations; for variables analyzed under the Wilcoxon signed-rank test, only those having four or more patients were reported. A complete table of all comparisons can be found in Supplementary File S1. A negative correlation was detected between height and weight and mes.CTCs/mL (Spearman coefficient = −0.59 and −0.57, respectively; p-value = 0.01 for both), as well as preoperative glomerular filtration rate (GFR) and CK|CD45/CD31 cells (Spearman coefficient = −0.55, p-value = 0.01). A positive correlation was found between age and For the 20 patients diagnosed with UTUC, recurrence data post-surgery were collected. However, the majority of the patients had short follow-up times, with a median follow-up of 111.5 days after surgery. Thus, to perform correlation analysis, the follow-up time after surgery was taken into account. To that end, we considered patients with recurrence and follow-up information of 6 months and 130 days, which limited the dataset to six and five patients, respectively. Within the two timeframes, three patients recurred within 6 months, and two patients recurred within 5 months. No statistically significant differences were found when performing the correlation analysis for recurrence. Additional time is needed for the data to mature in order to elucidate further correlations between recurrence status and liquid biopsy analytes.

Discussion
This study shows the detection of liquid biopsy analytes that allow for the stratification of NDs from patients diagnosed with UTUC prior to RNU. Combined with current clinical standards, the liquid biopsy has the potential to provide diagnostic confidence or prognostic utility around treatment response and the monitoring of minimal residual disease. We document several important findings from the liquid biopsy analysis of patients diagnosed with UTUC prior to surgery: (1) LEVs and CTCs are detectable in the PB. (2) The rare-event profile is highly heterogeneous. (3) Clinical data elements correlate with liquid biopsy analytes. As a non-invasive tool for the discovery of cancer-related biomarkers, the liquid biopsy has the potential to represent the complex process of tumorigenesis and metastasis.
The data presented here suggest that liquid biopsy analysis for CTCs and LEVs should be further investigated in a larger cohort for inclusion in UTUC patient care.
Total rare cells and LEVs were shown to be detected in UTUC samples at a significantly higher count than in ND samples (median 169.85 cells/mL vs. 34.46 cells/mL and 8.33 LEVs/mL vs. 3.34 LEVs/mL, respectively). Additionally, both positive and negative correlations were identified between multiple liquid biopsy analytes and clinical data metrics, including CCI status, tumor location, and previous bladder cancer status. A promising new analyte for UTUC patient care detected in an enrichment-free liquid biopsy approach are circulating LEVs. The four different LEV classifications detected in this study represent further tumor heterogeneity and were demonstrated to be present at significantly higher counts in patients with UTUC prior to RNU compared to the NDs. The presence of tumor-associated LEVs is most likely the result of the presence of an intact primary tumor in the patient. Exosomes contain various analytes (RNA, DNA, proteins, and metabolites) concordant with the parental cell properties, suggesting they are a promising alternative to circulating tumor DNA (ctDNA) or CTCs as disease biomarkers.
The liquid biopsy may be a source for detecting biomarkers of micrometastases indicative of early disease dissemination, and the assessment of these data prior to surgical intervention is critical. A meta-analysis of 2161 UC patients indicated that the presence of CTCs in the PB correlated with histological grade, tumor stage, and metastasis while being an independent predictive indicator of poor outcomes [14]. It is important to note that this analysis included 30 different published studies using various methods for CTC detection. Therefore, patients with low CTC counts prior to undergoing RNU are hypothesized to have a lower chance of recurrence and are therefore encouraged to receive surgical treatment. Using the HDSCA3.0 workflow, we detected both epi.CTCs and mes.CTCs in a subset of patient samples (45%), but no statistical difference was found compared to ND samples for these cell frequencies. Despite this, additional candidate CTCs were observed at higher frequencies, including CK|Vim|CD45/CD31 (median 27.30 cells/mL) and CK|CD45/CD31 (median 4.54 cells/mL). In metastatic prostate cancer, the rare-cell populations expressing CK|CD45/CD31 included platelet-coated epi.CTCs, which was determined to be a predictive biomarker for treatment response [24]. Understanding the molecular role of platelets in CTC biology will allow for the identification and characterization of clinically relevant CTCs for patient care [32]. Foerster et al. show that, in a large cohort of patients diagnosed with UTUC, mean platelet counts increased with advanced stage and tumor size [33]. Platelets may be indicative of more aggressive disease, and further proteomic characterization is warranted to understand the biological significance of this cellular population. This study highlights the promise of the liquid biopsy for the early risk stratification of patients with UTUC, prediction of disease outcome, and early detection of metastatic relapse.
Our analysis revealed a heterogeneous rare-cell population in the PB of patients with UTUC. Eight unique rare-cell categories were defined by their expression of CK, Vim, and CD45/CD31. Each cellular category included a mixture of cell types, which can be exemplified through the morphometric parameters, such as cellular eccentricity and size. The CCI status negatively correlated with various liquid biopsy rare-cell types; therefore, we conclude that the rare cells detected are representative of a range of comorbid conditions. Furthermore, the detection of Vim only cells in the PB correlated with the primary tumor location. This is evidence for the circulation of various rare cells as a measure of overall disease state. Further molecular analysis on the rare cells detected is needed and could help sub-stratify cells within each channel-type classification. For example, single-cell RNA sequencing could reveal the origin of the Vim only cells to further understand the association with specific tumor locations [34]. Here, we show that circulating rare cells are associated with a disease state and may be critical towards understanding a more comprehensive liquid biopsy profile beyond CTCs.
Molecular characterization of both circulating rare cells and LEVs detected in this study will elucidate their role in the processes associated with UTUC tumorigenesis and metastasis. Understanding the genomic architecture and proteome will reveal elements attributing to tumor growth, treatment resistance, and metastasis to inform clinical decision making for patient-centered treatment and the development of novel therapies. The HDSCA3.0 workflow provides the opportunity for a comprehensive liquid biopsy analysis via targeted multiplexed proteomics and single-cell genomic analysis [17,18,20,22,23,35], as well as ctDNA analysis [36][37][38]. Prior studies indicated that high ctDNA levels were associated with progression and metastatic disease [39][40][41]. With the use of a single platform, it is possible to characterize tumor heterogeneity through the combination of methodologies to provide a precision medicine approach for patients diagnosed with UTUC. Future research will focus on establishing evidence for the clinical utility of the liquid biopsy in UTUC to predict relapse post-RNU and to assist in clinical decision making to improve patient outcomes.

Conclusions
This study demonstrates initial evidence for the clinical utility of the liquid biopsy in UTUC patient care with the future objective of predicting recurrence post-surgery, enabling clinical intervention(s) that can improve patient outcomes. Here, we find that rare cells and LEVs can be used to distinguish patients with UTUC from NDs, with distinct populations indicating inter-and intra-patient heterogeneity through the liquid biopsy. While additional studies are warranted to uncover the power of liquid biopsy analytes to predict recurrence, the findings from this study support the liquid biopsy as a promising clinical tool for UTUC patient care.