Abstract
The rising utilization of circulating tumor DNA (ctDNA) assays in Precision Oncology may incidentally detect genetic material from secondary sources. It is important that such findings are recognized and properly leveraged for both diagnosis and monitoring of response to treatment. Here, we report a patient in whom serial cell-free DNA (cfDNA) monitoring for his known prostate adenocarcinoma uncovered the emergence of an unexpected FGFR3-TACC3 gene fusion, a BRCA1 frameshift mutation, and other molecular abnormalities. Due to the rarity of FGFR3 fusions in prostate cancer, a workup for a second primary cancer was performed, leading to the diagnosis of an otherwise-asymptomatic urothelial carcinoma (UC). Once UC-directed treatment was initiated, the presence of these genetic abnormalities in cfDNA allowed for disease monitoring and early detection of resistance, well before radiographic progression. These findings also uncovered opportunities for targeted therapies against FGFR and BRCA1. Overall, this report highlights the multifaceted utility of longitudinal ctDNA monitoring in early cancer diagnosis, disease prognostication, therapeutic target identification, monitoring of treatment response, and early detection of emergence of resistance.
Introduction
Circulating tumor DNA (ctDNA) analysis has an established role in uncovering actionable driver mutations in solid tumors, while having potential future uses in cancer diagnosis and monitoring as well1,2,3. The concept of a “molecular response”, as defined by on-treatment reductions in ctDNA, correlates with traditional assessments such as radiographic response and survival4,5,6, but is not yet fully validated in solid malignancies. Recently, a ctDNA-guided approach to de-escalating adjuvant chemotherapy in resected stage 2 colorectal cancer resulted in reduced chemotherapy usage with no decrease in recurrence-free survival7. The use of ctDNA for early cancer detection is also under evaluation8,9,10. Moreover, with increasing use of ctDNA testing in clinical practice, clinicians will inevitably encounter unexpected incidental molecular findings derived from clonal hematopoiesis of indeterminate potential (CHIP) or previously undiagnosed malignancies11. Here we report a case where ctDNA monitoring in a patient with known prostate adenocarcinoma resulted in the expedited diagnosis of an asymptomatic urothelial carcinoma (UC) and was incorporated into the longitudinal assessment of response to treatment.
Results
Clinical history
A 73-year-old Hispanic male smoker with biochemically recurrent, non-metastatic castration-sensitive prostate cancer (PC), well-controlled on androgen deprivation therapy (ADT), was found on longitudinal ctDNA monitoring to have acquired somatic DNA variants that had not been reported in his PC tissue, including a FGFR3-TACC3 gene fusion, raising suspicion for a second malignancy.
Nine years earlier, the patient was diagnosed with prostate adenocarcinoma, Gleason 4 + 3, with serum prostate-specific antigen (PSA) 15.4 ng/mL. He was treated with intensity-modulated radiation therapy and 16 months of ADT with PSA nadir 0.08 ng/mL. Four years later, his PSA rose to 6.36 ng/mL; he declined salvage prostatectomy and was restarted on ADT. Since then, his serum PSA levels have remained minimally detectable and stable at <0.1 ng/mL.
Approximately 8 years after his initial presentation, the patient underwent next-generation sequencing (NGS) of his original PC biopsy tissue (Fig. 1A) and was started on longitudinal cell-free DNA (cfDNA) monitoring using the commercially available Tempus platform as part of a Precision Oncology initiative at our institution. Liquid biopsies for ctDNA monitoring were performed at the time of each follow-up visit, on average three months apart. Initially these reported no abnormalities, but approximately nine years from his initial PC presentation, the clinical reports of the liquid biopsies began to demonstrate multiple molecular abnormalities at rising variant allele fractions (VAFs), including TP53 p.G245D, BRCA1 p.N1521fs, MYC amplification and FGFR3ex18-TACC3ex12 fusion (Figs. 1A, B). Because none of these were present in the patient’s original PC tissue, and as FGFR3-TACC3 fusions are very rare in PC (Supplemental Table 1), we pursued workup for a second malignancy12. Imaging showed a left ureteral mass (with biopsy revealing high-grade UC, positive for GATA-3, P63, CK7, and CK20), a pathologic para-aortic node, and liver lesions in segments 2 and 6 (Figs. 1A, 2A–C, 3). Liver biopsy confirmed metastatic UC (mUC) (Fig. 3). NGS performed on the liver biopsy specimen demonstrated the FGFR3ex18-TACC3ex12 fusion, which, interestingly, was not detected in the primary UC biopsy (Figs. 1A, 4).
Systemic treatment for mUC was initiated with cisplatin and gemcitabine. After one cycle, all molecular abnormalities disappeared from the patient’s liquid biopsy clinical report (Fig. 1B). After three cycles of chemotherapy (out of six planned), there was concern for cisplatin nephrotoxicity. Repeat imaging showed response at all sites of disease (Fig. 2D–F). Given radiographic as well as molecular response with elimination of detectable ctDNA findings from the clinical NGS reports, it was decided to de-escalate treatment and switch to immunotherapy with pembrolizumab. Six months later, the clinical reports of cfDNA NGS began again to show abnormalities seen previously in the patient’s UC, including TP53 p.G245D as well as, eventually, the BRCA1 p.N1521fs and the FGFR3ex18-TACC3ex12 fusion (Fig. 1B). Restaging scans at that time did not show radiographic progression of disease (Fig. 2G–I).
Approximately four months after the reappearance of the UC-associated ctDNA abnormalities, restaging imaging showed a questionable new sclerotic lesion within the T9 vertebral body. All prior sites of disease, including the urothelial primary site and liver metastases, were stable in size. After discussion with Radiology, a decision was made to obtain a bone scan, which was performed one month later and was convincing for new bony metastatic disease, with sclerotic lesions showing radiotracer uptake at T7-T10, T3, sternum, right second rib, and left ninth rib (Fig. 2M, N). Repeat body imaging also demonstrated minimal enlargement of the UC primary, while the known liver metastases remained stable (Fig. 2J–L). During this time, the VAFs of the UC-associated molecular abnormalities in the cfDNA continued to rise (Fig. 1B), and, for the first time, additional gain-of-function (GOF) variants FGFR2 p.F276C and NRAS p.Q61L were reported in the clinical results by the NGS vendor. Meanwhile, the patient’s PSA remained minimally detectable at <0.1 ng/mL, and no PC-related mutations were found in the cfDNA. Thus, clinical suspicion was much higher for progressing UC rather than PC relapse. At this time, the patient was offered a therapy switch for presumed progressing UC, but ultimately it was decided to first pursue a confirmatory biopsy of the T9 bone lesion, which showed metastatic UC (Fig. 3). Unfortunately, this tissue was inadequate for NGS testing. Radiation treatment was administered to the T9 bone lesion, with a rapid reduction in ctDNA VAFs of UC-associated mutations (Fig. 1B, Supplemental Table 2). The patient was then initiated on systemic therapy with the FGFR inhibitor erdafitinib, which initially resulted in further improvement of the ctDNA VAFs of the UC-associated mutations. However, very quickly, the VAFs for TP53 p.G245D and BRCA1 p.N1521fs began to rise again (Fig. 1B, Supplemental Table 2), while the ctDNA presence of FGFR3ex18-TACC3ex12 and FGFR2 p.F276C declined dramatically (Supplemental Tables 2 and 3). At the same time, the NRAS p.Q61L VAF increased rapidly (Supplemental Table 2). Collectively, these observations suggest that FGFR3ex18-TACC3ex12 and FGFR2 p.F276C are sensitive to erdafitinib, while NRAS p.Q61L can serve as a mechanism of resistance to it.
Retrospective re-analysis of cfDNA sequencing data
Throughout the patient’s course detailed above, clinical-grade NGS results were provided by the clinical sequencing vendor to the treating physicians in real time. While NGS on the cfDNA platform is performed at 5,000x coverage, clinical reports are limited by reporting cutoffs set by the vendor, including VAF of 0.1% for missense variants and 0.5% for insertions/deletions (indels), as well as lack of reporting VAFs for fusion products. We, therefore, retrospectively sought to better understand our patient’s disease course by re-aligning the raw sequencing data, as described in the Methods section, for possible UC-derived variants that were present below the vendor’s reporting thresholds.
In cfDNA samples from the period preceding the diagnosis of mUC, our own tumor-informed re-alignment and re-analysis found the TP53 G245D and BRCA1 N1521fs mutations to be present 12 (at least) and 8 months, respectively, prior to their earliest appearance in the clinical reports, at VAFs of 0.24% and 0.04%, respectively. FGFR3ec18-TACC3ex12 was not detected at any new earlier time point compared to the clinical reports (Fig. 1B and Supplemental Table 2).
After initiation of cytotoxic chemotherapy for mUC, the clinical NGS reports were negative for any UC-associated cfDNA findings for approximately six months. However, our own tumor-informed re-alignment and re-analysis found the TP53 G245D to be detectable in the cfDNA throughout this entire period, while we detected the BRCA1 N1521fs in cfDNA samples collected 2 months prior to its re-appearance in the vendor’s clinical reports (Fig. 1B). Details on all analyzed variants are presented in Supplemental Table 2.
Collectively, our own tumor-informed re-alignment and re-analysis shows that UC-associated somatic mutations could be detected in the cfDNA, albeit at low VAFs, at least 12 months prior to their earliest appearance on the vendor’s clinical reports; at least 18 months prior to clinical UC diagnosis; and at least 12 months prior to definitive radiographic progression after first-line therapy.
Supplemental tissue genomic testing
The clinical sequencing vendor provides a homologous recombination deficiency (HRD) test, in which a score computed based on a proprietary algorithm on formalin-fixed tissue samples is designed to predict the probability that a tumor’s gene expression profile correlates with benchmarks of a HRD phenotype, such as those found in tumors with biallelic BRCA loss13. Testing of both the UC primary and the liver metastasis yielded a highly positive score of 88/100 and 96/100 by this assay, respectively (assay cutoff: >50/100 is considered HRD+)…
