Plasma and Urine Free Glycosaminoglycans as Monitoring Biomarkers in Nonmetastatic Renal Cell Carcinoma—A Prospective Cohort Study

Take Home Message No liquid biomarkers are approved in renal cell carcinoma (RCC), making early detection of recurrence events following surgery of nonmetastatic RCC difficult. Here we develop a noninvasive blood/urine test measuring free glycosaminoglycans that can detect RCC recurrence events.


Introduction
Renal cell carcinoma (RCC) is the most common kidney malignancy and accounts for about 174 000 deaths annually worldwide [1]. Nonmetastatic RCC (M0 RCC) is largely curable by surgical resection via partial or radical nephrectomy, the recommended treatment for the majority of patients [2]. However, 20-30% of all M0 RCC cases experience postsurgical recurrence within 5 year [3]. Current follow-up strategies aim at early detection of recurrence because symptomatic recurrences have poorer prognosis than recurrences detected within a follow-up protocol [4]. Despite the importance of follow-up, there is disagreement in the selection of the optimal strategy. Current protocols largely depend on radiological imaging, typically chest/abdomen computed tomography every 6-12 months up to [3][4][5] year depending on the recurrence risk after surgery [5]. There is evidence that imaging is currently offered too intensely with little survival benefit [6]. Conversely, longterm recurrences after 5 year from initial surgery occur in a non-negligible fraction of patients [7].
Noninvasive detection of recurrence using molecular biomarkers could radically improve follow-up. However, currently neither molecular nor liquid biomarkers are approved in the clinical management of RCC [8,9]. Liquid biopsies are an emerging class of cancer diagnostics based on biofluidic tumor-related biomarkers, for example, circulating free DNA (cfDNA) [9]. However, cfDNA has so far shown disappointing yields in RCC [10]. Recent developments using methylated cfDNA and metabolomics are promising [11,12], but these assays are complex and the availability of standardized kits remains years away.
Glycosaminoglycans (GAGs) are structurally diverse polysaccharides. Their complex profile of sulfation and epimerization patterns-also called the GAGome-is now viewed as a nongenetic code that controls a wide range of biological functions, with many of these being active in cancer [13,14]. We and others observed that genetic alterations specific to RCC correlated with significant reprogramming of cell metabolism [15][16][17][18]. Using a systems biology approach, we identified GAG biosynthesis as the most deregulated metabolic pathway in RCC [19]. In exploratory prospective and retrospective case-control studies, plasma and urine GAGomes were significantly altered in RCC patients compared with healthy individuals [15][16][17]. Successively, we verified the analytical performance of an ultra-high-performance liquid chromatography and triplequadrupole mass spectrometry (UHPLC-MS/MS) kit for GAGome analysis, which selectively measures the proteinfree fraction or free GAGomes [18]. As opposed to total GAGomes in previous studies [15,16], free GAGomes had stable and robust reference levels in healthy adults [19].
In this study, we sought to explore whether urine and plasma free GAGomes correlated with postsurgical recurrence in M0 RCC in a single-center prospective cohort study.

Study design
This study is reported in compliance with the STARD guidelines (Supple-

Correlation analyses
Each detectable feature in the plasma and urine free GAGome was corre-

Statistical analysis for the diagnostic performance of free CS RCC recurrence scores
We performed internal validation using bootstrap resampling and reported the predictive performance in terms of scaled Brier score (%) and Nagelkerke R 2 (%) with empirical 95% confidence intervals (CoIs).
We assessed the discrimination performance in terms of the area under the receiving operating characteristic curve (AUC) for the classification of RD versus NED according to each score versus the observed recurrence status. The goodness of fit and corresponding v 2 statistics were tested using the Hosmer-Lemeshow test. We used a decision curve analysis to determine a single test positivity cutoff for all free CS RCC recurrence scores, so that the net benefit over the base scenarios ''intervention for all'' (ie, radiological evaluation for all) and ''intervention for none'' (ie, radiological evaluation for none) was maximized [21]. We determined the clinical usefulness of each score by cross-tabulating RD versus NED as predicted by the free CS RCC recurrence score versus the recurrence status. We performed three distinct analyses, one per score (plasma, urine, and combined), to determine its sensitivity and specificity and the two-sided exact 95% CoI. The index test was considered valid if the sum of sensitivity and specificity point estimates was significantly higher than 1, that is, if the odds ratio (OR) from logistic regression where the dependent variable was the recurrence status and the independent variable was the index test result (positive or negative) was >1. A minimum of 30 patients were targeted for enrollment for the present study, which was deemed sufficient to estimate sensitivity with 15% marginal error, assuming an ideal index test with 80% sensitivity and 80% specificity to RD and assuming 25% prevalence of RD [22]. All statistical analyses were reported according to best practices in clinical prediction models [23], and further details are presented in the Supplementary material.

Survival analyses
We correlated recurrence-free survival (RFS) with each preoperative score in the subset of patients with pathological diagnosis of M0 RCC from cohort 1 (Supplementary material). A univariable survival analysis was performed by fitting a Cox proportional hazard model to estimate the OR per unit of change of each preoperative score (0, 1, and 1 missing datum for plasma, urine, and combined, respectively) and the 95% CoI. In addition, two other variables were considered for the univariable Cox regression: age and Leibovich risk score. A multivariable Cox model was prespecified using variables reaching statistical significance in the univariable analysis. The sample size was not powered for this analysis given that survival was not intended as a primary endpoint.

Patient characteristics
The study design and patient flow are illustrated in Figure 1. We prospectively enrolled a total of 127 participants-72 were elected for partial or radical nephrectomy for suspected RCC (cohort 1) and 55 were scheduled for first-line therapy for metastatic RCC (cohort 2). Of the 127 patients in the total population, the study population included 50 patients with M0 RCC having undergone curative intent surgery from cohort 1 and 12 patients with metachronous RCC recurrence from cohort 2, for a total of 62 patients. The sample collection visits per patient and time to recurrence are shown in Supplementary Figure 1. The median age was 63 year (interquartile range [IQR]: 56-70). The population was predominantly composed of males (65%, 40 males vs 22 females). The most common histological subtype was clear cell RCC (ccRCC; N = 51, 82%), followed by five papillary RCC (two type 1, two type 2, and one unspecified) and six chromophobe RCC cases. We listed other histopathological characteristics per cohort and across the study population in Table 1.
In the total population (N = 127), there were 122 RCC (106 ccRCC and 16 non-ccRCC) and five oncocytoma patients. We listed other population characteristics per cohort and across cohorts in the total population in Table 1.
No adverse events were recorded during sample collection or recurrence status evaluation.
Overall, we measured the free GAGome in a total of 378 (195 plasma and 183 urine) samples in a single blinded central laboratory using a standardized UHPLC-MS/MS kit. No samples were identified as outliers.

Correlation between free GAGomes and radiological recurrence
In the study population (N = 62), we explored the correlation between free GAGomes and recurrence status using samples obtained at the same postoperative visit. The median follow-up time from curative intent surgery were 3 months (IQR: 1.8-7.8 months, maximum: 55 months). Each patient contributed on average 1.2 visits for a total of 72 visits. Throughout these, we recorded 17 (24%) RD and 55 (76%) NED cases. Plasma and urine samples were obtained, respectively, in 71 (99%) and 67 (93%) visits.
We found that seven of 40 plasma and 20 of 40 urine free GAGome features were detectable (ie, measurable in the majority of samples). Of these, we determined that six (86%) of seven detectable features in plasma and eight (40%) of 20 detectable features in urine were compatible with RD versus NED ( Fig. 2A, Supplementary Fig. 2 Table 1). In general, RD tended to correlate with plasma and urine free GAGomes featuring a higher concentration of unsulfated CS (0S CS) in both absolute and relative levels, and conversely a relatively lower concentration of certain monosulfated CS (4S CS in plasma and 6S CS in urine). Overall, we observed that recurrence was compatible with alterations in 14 (52%) of detectable free GAGome features.

Correlation of preoperative free GAGome features with histopathological variables
In the subset of the total population with preoperative samples from cohort 1 (N = 67), we observed that 11 (79%) of 14 preoperative free GAGome features compatible with recurrence were also compatible with at least one histopathological variable and in the same direction of correlation (Fig. 2B  and Supplementary Table 2). In both plasma and urine, we observed compatible correlations with tumor size, Fuhrman nuclear grade, and TNM stage. We observed nominal correlations with malignant histology and clear cell histology, but these analyses appeared underpowered (only five benign and ten non-clear cell cases, respectively). Most of these correlations were with free GAGome features measured in relative concentrations (%). It was worth noting, however, that the absolute concentration of unsulfated CS in the urine was strongly compatible with large tumor size (>7 cm).

3.4.
Development of plasma, urine, and combined free CS RCC recurrence scores We developed three BART models to regress RD versus NED based on plasma or urine or combined free GAGome features. We manually curated the selection of two features for each model: for plasma, unsulfated CS (in lg/ml) and 4-sulfated CS (in %); and for urine, unsulfated CS (in lg/ ml) and 6-sulfated CS (in %). For the combined model, we used as inputs the outputs of the plasma and urine models. The predicted probability of each model was scaled between 0 and 100, and henceforth referred to as plasma, urine, or combined free CS RCC recurrence score, given that the models rely solely on free CS. For example, if the plasma free CS RCC recurrence score is >50, then the probability of RD is >50% according to the model that uses only plasma features as predictors.

3.5.
Clinical usefulness of plasma, urine, and combined free CS RCC recurrence scores We used a decision curve analysis to determine the net benefit of using each free CS RCC recurrence score to inform the diagnosis of RD at different cutoff values (Supplementary   5). We defined a single cutoff of >30 for all scores, indicating that a predicted probability of RD of >30% would trigger a radiological evaluation. In other words, the clinician would be willing to undergo about two unnecessary radiological evaluations to diagnose one case of RD. At this cutoff, the sensitivity and specificity to RD were 69% and 84% in plasma (OR = 1.7, p < 0.001), 81% and 80% in urine (OR = 2.0, p < 0.001), and 80% and 82% when combined (OR = 2.1, p < 0.001), respectively ( Table 2).
We examined the net benefit in two scenarios. In the first scenario, we imagined a follow-up visit in which there is no radiological evaluation of recurrence planned (''intervention in none'' scenario). For example, the European Association of Urology (EAU) guidelines recommend that radiological evaluations are curtailed after 3 year of postoperative follow-up with NED in those with a low risk of recurrence [2]. In this scenario, the net benefit was 10%, 13%, and 12% for the plasma, urine, and combined free CS RCC recurrence scores, respectively ( Table 2). As opposed to the strategy of performing no radiological evaluations in none of the patients, we estimated that performing the radiological evaluation based on a positive score was equivalent to a strategy that found ten, 13, and 12 cases of RD per hundred patients without any unnecessary imaging in plasma, urine, or combined scores, respectively.
In the second scenario, we imagined a follow-up visit in which there is a radiological evaluation of recurrence planned (''intervention in all'' scenario). For example, the National Comprehensive Cancer Network guidelines recommend a radiological evaluation after 9 and 15 months after surgery as opposed to the EAU guidelines, indicating different willingness to intervene [24]. In this scenario, the interventions avoided were 49%, 51%, and 54% for the plasma, urine, and combined free CS RCC recurrence scores, respectively ( Table 2). As opposed to the strategy of performing radiological evaluations in all patients, we estimated that performing imaging based on a positive score is equivalent to a strategy that avoids 49, 51, and 54 radiological evaluations per hundred patients without missing any RD in plasma, urine, or combined score, respectively.
In Table 2, we reported the clinical usefulness metrics also at a ''low'' cutoff (20%), which was more sensitive to RD and more acceptable in the first scenario when no radiological evaluations are routinely performed, and a ''high'' cutoff (40%), which was more specific to RD and more acceptable in the second scenario when radiological evaluations are performed in excess.

RFS and preoperative free CS RCC recurrence scores
In the subset of the total population with preoperative samples from cohort 1 and pathological diagnosis of M0 RCC (N = 51), there were seven (14%) recurrences at a median follow-up time of 11 months. In the univariable Cox regression analysis (

Discussion
In this single-center prospective study, we discovered that plasma and urine free GAGomes changed in association with RCC recurrence after surgery. Although previous work demonstrated that GAGomes were altered in RCC compared with normal healthy levels [15,16], their behavior at recurrence had not been investigated. The specific free GAGome features that altered with recurrence were overall consistent with changes previously associated with RCC compared with healthy controls both in urine [16] and in plasma [15], except for some inconsistencies in plasma sulfation patterns that we attribute to the fact that the present study focused on free rather than total GAGomes as in the previous study and total GAGomes include protein-bound GAGs.
We developed recurrence scores that incorporated the largest changes in plasma and/or urine free GAGomes. These included only CS features, such as an increase in unsulfated CS. This is consistent with a previously reported in vivo experiment using a syngeneic mouse model of RCC progression that caused the same increase in both plasma and urine [25]. Consistently, here unsulfated CS correlated with tumor size and grade. This suggests that the recurrence scores track a free GAGome feature that may causally be implicated with RCC. However, other factors may cause the here-observed aberrations in plasma and urine CS that may be specific or less to RCC, like inflammation-associated proteoglycans [26], glycocalyx degradation [27,28], or renal and hepatic catabolism [29]. It will be important to elucidate the role of these factors to both improve the performance of the scores and understand potential sources of false positives and negatives.
Despite intense research on liquid biomarkers, none of these are recommended in the follow-up of RCC owing to the fact that virtually none of these reported prospective evidence of diagnostic performance in this setting [8]. In one of such studies, Wan et al [30] investigated prospectively total plasma cfDNA levels for monitoring postoperative recurrence events, reporting promising sensitivity and specificity to RD. However, the cutoff value and other critical performance parameters were not reported, rendering the interpretation of results for clinical practice difficult. In this context, our study represents one of the first prospective evaluations of liquid biomarkers for surveillance for RCC recurrence.
The free CS RCC recurrence scores appeared clinically useful. At different cutoffs, the free CS RCC recurrence scores, which are equivalent to the predicted probability of recurrence, had substantial net benefit compared with the strategy of performing imaging in all or no patients. This suggests that recurrence scores can considerably improve follow-up protocols of RCC patients, which are currently imaging intense with weak clinical benefits [6]. For example, we showed that performing a radiological evaluation based on a positive score could detect more recurrences or avoid unnecessary imaging with no extra harm for the patients. However, we also note that at present the sensitivity, which determines the false negative rate, appears insufficient to completely forego routine radiological evaluations. Beyond scientific evidence, the integration of free CS RCC recurrence scores in the clinical routine would also require determining the cost effectiveness of this extra procedure and benefits for patients and clinicians. We speculate that intercepting a patient when the recurrence is still local or oligometastatic may not only extend the number of quality-adjusted life years, for example, by opening a window for salvage treatments, but may also decrease the need of pricey systemic therapies indicated for metastatic RCC. The addition of free CS RCC recurrence scores to current follow-up protocols for surveillance for recurrence may therefore have positive health economics with substantial benefits for patients.
Limitations of this study include the relatively small event size (ie, number of recurrences) and a lack of external cohorts for validation. These two factors compound the risk of selection and detection biases. We believe that the prospective consecutive series cohort design of the study and the double-blind analysis of biomarker versus radiological results properly controlled for these biases. However, other factors such as inclusion of different RCC subtypes and median follow-up of <1 year may confound and limit the generalizability of the findings. The small sample size for these subsets cannot rule out that free GAGomes may not be as significantly altered in conjunction with longterm recurrences or in non-clear cell subtypes. Therefore, the generalizability of the here-reported performance of free CS RCC recurrence scores in a clinical setting should wait until evidence of external validation is produced in a more narrowly defined and larger M0 RCC population. This is, for example, investigated in the clinical study AURORAX-0087A (NCT04006405).

Conclusions
We conclude that plasma and urine free GAGomes were altered in association with RCC recurrence and that free CS RCC recurrence scores could substantially improve the follow-up of RCC patients after surgery. External validation of the scores is warranted prior to a wider implementation in the clinical routine.
Author contributions: Jens Nielsen, Francesco Gatto and Sven Lundstam had collectively full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Drafting of the manuscript: Gatto.
Critical revision of the manuscript for important intellectual content: All authors.
Administrative, technical, or material support: None.