In total 150 (93.2%) of the urine samples and 11 (91.7%) of tissue samples passed the quality control for further testing (Figure 1). Overall, 107 males and 43 females were enrolled as subjects, with a median age of 60 (range from 18 to 88) years. Patients and tumor characteristics are described in Table 1. In 64 cases, UTUC was confirmed as a cause of hematuria while the cause of the remaining 86 patients was found to be non-malignant. Patients diagnosed with UTUC were significantly older compared to non-malignant patients (p<0.01, Table 2). FISH tests were only performed on 80% (n=51) of UTUC+ and 9% (n=8) of UTUC- patients. The sensitivity and specificity of FISH were 51% (26/51) and 100% (0/8) respectively.
The concordance profiling between urine ctDNA and matched tumor tissues
The consistency of mutations in urine samples and the corresponding tissue samples were evaluated to confirm the sources of these variants. As a result, a total of 12 matched tissue samples were effectively collected and 11 qualified DNA samples identified. As shown in Figure 2, 14 variants from 5 genes were detected from 9 UTUC+ samples where 13 of them were positive in both types of samples. In the case of RG180, AKT mutation was shown from urine other than tissue samples indicating the effectiveness of urine samples as a supplement for genetic analysis of UTUC tissue samples in instances where genetic heterogeneity is considered as an issue. In addition, no mutations were detected in the urine and tissue samples of the two UTUC- patients. In summary, the concordance rate of variant detection between urine and tissue samples was 93% (13/14).
Univariate logistic regression of significant features
Univariate analysis was performed for each of these variants as well as clinical factors to assess the strength of these factors in evaluating UTUC risk by calculating the odds ratios (Table 2). Mutated or methylated Gene including FGFR3, TERT, TP53, ONECUT2, and age older than 50 showed a significant impact in evaluating UTUC risks (P-value<0.01). And the superiority of the panel was witnessed in the integration of all these markers (≥ 1 of 17 gene mutated or ONECUT2 CpG methylated).
Gene mutations results of urine samples were consistent with characteristics of previous UTUC mutation and provided new clinical potential applications
To better understand how each variant contributes to this panel, a heatmap was drawn in Figure 3. Despite this panel covering hot-spots mutations of 17 genes, only variants from 9 genes showed positive mutations. TERT C228T, FGFR3 c.746C>G, c.1118A>G, and TERT C250T were on the top 4 of the list with a long tail of several other mutations from ERBB2, HRAS, KRAS, PIK3CA, TP53,U2AF1, and AKT1 (Supplementary table 1). This distribution pattern of driver genes corroborates with previous research in the sense that TERT, FGFR3, TP53, PIK3CA and RAS genes exhibited high frequencies in UTUC mutation landscape. FISH test results were shown along with the panel, notably, the sensitivity of the FISH test was low (about 51% (26/51), but with perfect specificity and no false-positive found from the 8 tested UTUC- samples (Table 1). With the mutation test, the sensitivity and specificity were 71.9% and 91.4% respectively (Table 3), which was close to the sensitivity of 75% in the UTUC diagnostic cohort in a previous study that solely used mutant genes panel . Therefore, these data confirm the limitation of sensitivity when gene mutation detection was used solely in the diagnosis of UTUC.
Additionally, a significantly higher frequency of TP53 mutations in high versus low-grade samples (31.9% vs. 0%; p=0.0065, Fisher’s Exact Test) was observed, and conversely, found disproportionately more FGFR3 mutations (47.1% vs. 17.0%; p=0.0223, Fisher’s Exact Test) and PIK3CA mutations (23.5% vs. 2.1%; p=0.0155, Fisher’s Exact Test) in low versus high-grade cases (Figure 4a). Likewise, a significantly higher frequency of TERT promoter mutations (72.7% vs. 25.6%; p=0.0005, Fisher’s Exact Test) and HRAS mutations (0% vs. 18.2%; p=0.014, Fisher’s Exact Test) was evident in non-muscle-invasive versus muscle-invasive samples for the first time in UTUC cohort (Figure 4b). This thus reflected the significance of adding detection of gene mutation to our test panel, which could provide evidence for the classification of UTUC+ patients.
ONECUT2 methylation exhibited a satisfactory performance as a diagnostic biomarker of UTUC
The analysis was performed to confirm the best cutoff of ONECUT2 methylation status (Figure 5). The Δct value of ONECUT2 in all urine samples is shown in Figure 5a. With a cutoff of 7.93, the ONECUT2 methylation detection ability was the largest, displaying the AUC of 0.93 (Figure 5b). With the singly use of ONECUT2 methylation test in this cutoff value, genetic abnormalities in 89.1% (57/64) urine of UTUC+ patients and 5.8% (5/86) of UTUC- group were detected resulting in a sensitivity of 89.1% (57/64), and a specificity of 94.2% (81/86) (Table 3). This performance of the ONECUT2 methylation test was better than the one reported previously (sensitivity of 82% and specificity of 62% with a panel of VIM, RASSF1A, GDF15, and TMEFF2 methylation in UTUC group) .
The performance of the test panel in UTUC detection
By combining ONECUT2 methylation and gene mutation results as a UTUC diagnostic test panel (≥1 of 17 genes mutated or ONECUT2 CpG methylated showed a positive result), the performance of the test improved, the sensitivity of this test panel rose to 92.2% (59/64), and the specificity was 91.9% (79/86). Simultaneously, the panel demonstrated a positive predictive value of 89.4% and a negative predictive value of 94.1% (Table 3). Moreover, by combining the detection results of gene mutations with ONECUT2 methylation, the sensitivity was further improved. It was worth noting that the double-positive result (≥ 1 of 17 genes mutated and ONECUT2 CpG methylated) potentially reveal a higher risk of UTUC. A regular follow-up of two patients RH645 and RG342 (Supplementary Table 1) with a double positive test result in UTUC- cohort was conducted, it was found that patient RG342 was diagnosed with ureteral cancer in May 2019. Notably, close follow-up of patient RH645 was still ongoing.
Comparison of multivariate logistic regression models
Out of the 150 samples, 108 were randomly selected as the training set, and the remaining 42 samples were the validation set. Based on the results of Univariate logistic regression, significant features were combined to construct 4 logistic regression models. From the ROC curve shown in Figure 6, the model constructed with the features of age and the mutation status of TERT promoter (mutation of at least one hotspot on TERT g.1295228C>T and g.1295250C>T) and ONECUT2 methylation level (Model D) had the largest area under the curve of 0.957, whereas the AUC of other three models were 0.947 (age and panel test results, Model C), 0.953 (age and ONECUT2 methylation, Model A) and 0.903 (age and 17 genes mutation test result, Model B). By selecting the optimal cutoff according to the highest Youden index in each of the four models, the model with age, mutation status of TERT promoter and ONECUT2 methylation level showed an optimal performance with a sensitivity of 94.0%, a specificity of 93.1%, a PPV of 92.2% and an NPV of 94.7% (Table 4). This model maximized sensitivity without a major reduction in specificity hence was considered optimal. And in the validation set, a prediction using the above features were completed and obtained an AUC of 0.962 (Supplementary Figure 1).