Comparison of molecular testing modalities for detection of ROS1 rearrangements in a cohort of positive patient samples

Abstract

Introduction –

ROS1 gene fusions are a well-characterized class of oncogenic driver found in approximately 1-2% of non-small cell lung cancer patients. ROS1-directed therapy in these patients is more efficacious and is associated with fewer side effects compared to chemotherapy and is thus now considered standard-of-care for patients with advanced disease. Consequently, accurate detection of ROS1 rearrangements/fusions in clinical tumor samples is vital. In this study, we compared the performance of three common molecular testing approaches on a cohort of ROS1 rearrangement/fusion-positive patient samples.

Methods –

Twenty-three ROS1 rearrangement/fusion-positive clinical samples were assessed by at least two of the following molecular testing methodologies: break-apart fluorescence in situ hybridization (FISH), DNA-based hybrid capture library preparation followed by next-generation sequencing (NGS), and RNA-based anchored multiplex PCR library preparation followed by NGS.

Results –

None of the testing methodologies demonstrated 100% sensitivity in detection of ROS1 rearrangements/fusions. FISH results were negative in 2 out of 20 tested samples, the DNA-based NGS assay was negative in 4 out of 18 tested samples, and the RNA-based NGS assay was negative in 3 of 19 tested samples. For all three testing approaches, we identified assay characteristics that likely contributed to false-negative results. Additionally, we report that genomic breakpoints are an unreliable predictor of breakpoints at the transcript level, likely due to alternative splicing.

Conclusions –

ROS1 rearrangement/fusion detection in the clinical setting is complex and all methodologies have inherent limitations of which users must be aware in order to correctly interpret results.

INTRODUCTION

ROS1, a gene encoding for a receptor tyrosine kinase, is known to undergo rearrangement in many cancer types, including non-small cell lung cancer (NSCLC) [1]. Rearrangement commonly leads to the creation of chimeric fusion genes that promote an oncogenic phenotype through constitutive activation of ROS1 kinase activity [1]. Pre-clinical and clinical data has demonstrated that cancer cells harboring ROS1 fusions are sensitive to small molecule tyrosine kinase inhibitors (TKIs) with activity against ROS1, including crizotinib [2, 3]. These initial studies led to clinical trials in which crizotinib demonstrated impressive activity in ROS1 rearrangement-positive NSCLC and was associated with only mild side effects [4–6]. As a result, crizotinib is now approved by regulatory agencies worldwide for the treatment of ROS1 rearrangement/fusion positive NSCLC. Second generation inhibitors with activity against ROS1 are also currently being investigated in this setting [7–9].

Due to the superior efficacy and more tolerable side effect profile of ROS1 inhibition compared to standard-of-care chemotherapy in ROS1 rearrangement/fusion positive patients, it is vital that these events are accurately detected for clinical management. The gold-standard assay has historically been break-apart fluorescence in situ hybridization (FISH), in which separation of two FISH probes that hybridize to either side of the gene is indicative of ROS1 rearrangement [2–4]. Immunohistochemistry (IHC) and reverse-transcription polymerase chain reaction (RT-PCR) have also been successfully employed [10, 11]. However, a major limitation with the above assays is that they are limited in the degree to which they can be multiplexed. Consequently, simultaneous evaluation of ROS1 and other gene fusions via these approaches is inefficient.

Recent years have witnessed a large increase in the number of characterized actionable or potentially actionable fusion genes in lung cancer [12–18]. Many of these fusion genes can be targeted with approved therapies (on- or off-label) or therapies that are currently in clinical trials. Consequently, clinical molecular diagnostic laboratories are, or soon will be, faced with a scenario in which testing for multiple gene fusions is requested by physicians. Therefore, employing single-gene assays such as FISH, IHC, or RT-PCR exclusively is no longer sufficient. Next-generation sequencing (NGS) is a relatively new technology that allows for nucleic acid sequencing in massively parallel fashion, thus facilitating the assessment of multiple genes simultaneously. Consequently, NGS-based assays are now being employed in various clinical diagnostic settings to test for gene rearrangements/fusions. There are a multitude of different approaches for preparing sample-derived nucleic acid for NGS (often termed ‘library preparation’), but a primary distinguishing feature among these is whether DNA or RNA is used as the initial input material for the assay. If using DNA, it is imperative that introns are sequenced, as the majority of genomic breakpoints that lead to gene fusions occur in introns. Alternatively, if using RNA, sequencing can instead be focused on coding regions.

In this study, we compared the performance of break-apart FISH, a DNA-based NGS approach, and an RNA-based NGS approach in the ability to detect ROS1 rearrangements/fusions in clinical lung cancer samples. To accomplish this, specimens from twenty-three ROS1-positive NSCLC patients were evaluated, a relatively large cohort considering that these patients make up only 1-2% of the overall NSCLC patient population. Inclusion in the cohort required a positive ROS1 result via any of the approaches plus testing (regardless of result) by at least one of the other approaches. For each assay, we found instances where the ROS1 rearrangement or fusion was not correctly identified, and we provide data and/or rationales for these false negatives. In addition, we demonstrated that genomic breakpoint position does not always predict breakpoint at the transcript level, adding another level of complexity to result interpretation.

MATERIALS AND METHODS

Break-Apart Fluorescence In Situ Hybridization

FISH analysis was predominantly performed in the Department of Pathology (Colorado Molecular Correlates Laboratory or Colorado Genetics Laboratory) at the University of Colorado – Anschutz Medical Campus, with the exception of three cases tested in outside laboratories (AmeriPath, Clarient, Memorial Sloan Kettering Cancer Center). For internally tested cases, FISH was performed on 4μM (± 1μM) thick formalin-fixed, paraffin-embedded (FFPE) tumor sections. ROS1 FISH probes used for these cases included Vysis (Abbott Molecular, Abbott Park, Illinois), SureFISH (Agilent, Santa Clara, California), or custom designed probes. FISH assays were performed as previously described [19], or using the Vysis Paraffin Pretreatment IV and Post-Hybridization Wash Buffer Kit (Abbott) per the manufacturer’s instructions. Signals were evaluated in at least 50 tumor nuclei per specimen. Specimens were considered positive for ROS1 rearrangement if >15% of cells displayed a split 5’/3’ and/or single 3’ signal pattern. Separation between 5’ and 3’ ROS1 signals of >1 signal diameter was required to score as split. The specific ROS1 FISH probes used for each sample are listed in Table 1. In some cases that were tested by outside laboratories, information regarding the specific FISH probes was not available.

Table 1.

The Specific ROS1 FISH Probes Used for Each Sample

Pt. #:	FISH Probesa:	FISH Results:	RNA-Based NGS:	RNA-QC:	DNA-Based NGS:
1	internal design	positive - 90% single 3′	SDC4(ex2):ROS1(ex32)	PASS	SDC4(int2):ROS1(int31)
2	Abbott	positive - 84% split signals	CD74(ex6):ROS1(ex34)	FAIL	CD74(int6):ROS1(int32)
3		not tested	CD74(ex6):ROS1(ex34)	PASS	CD74(int6):ROS1(int33)
4	Abbott	negative	GOPC(ex4):ROS1(ex36)	PASS	GOPC(int4):ROS1(int34)
5	Abbot + Agilent	positive - 86% split signals	SDC4(ex2):ROS1(ex32)	FAIL	not tested
6	Abbott	positive - 68% single 3′	no fusion call	FAIL	no rearrangement detected
7	Abbott	positive - 75% single 3′	not tested		EZR(int10):ROS1(int33)
8	unavailable	positive - 74% split signals	no fusion call	FAIL	CD74(int6):ROS1(ex33)
9	Abbot + Agilent	positive - 42% split signals	CD74(ex6):ROS1(ex34)	PASS	CD74(int6):ROS1(int33)
10	Abbott	positive - 72% single 3′	no fusion call	FAIL	EZR(int10):ROS1(ex33)
11	Abbott	positive - 88% split signals	SLC34A2(ex13):ROS1(ex32)	FAIL	not tested
12	Cytocell	positive - 70% split signals	not tested		CD74(int6):ROS1(int33)
13	Abbot + Agilent	positive - 78% single 3′	CD74(ex6):ROS1(ex34)	FAIL	not tested
14	Abbot + Agilent	positive - 76% split signals	ZCCHC8(ex2):ROS1(ex36)	PASS	not tested
15	Abbot + Agilent	positive - 69% single 3′	SLC34A2(ex13):ROS1(ex32/34)	PASS	SLC34A2(int13):ROS1(int31)
16	unavailable	positive - 100% split signals	not tested		TFG(int4):ROS1(int34)
17	internal design	positive - 75% split signals	CD74(ex6):ROS1(ex34)	PASS	not tested
18	internal design	negative	SLC34A2(ex13):ROS1 (ex34)	PASS	SLC34A2(int13):ROS1(int33)
19	Abbot + Agilent	positive - 83% single 3′	CD74(ex6):ROS1(ex34)	PASS	CD74(int6):ROS1(int32)
20	Abbot + Agilent	positive - 81% single 3′	CD74(ex6):ROS1(ex34/35)	PASS	CD74(int6):ROS1(int33)
21	Abbott	positive - 66% split signals	not tested		no rearrangement detected
22		not tested	CD74(ex6):ROS1(ex34)	PASS	no rearrangement detected
23	Abbot + Agilent	FISH fail	SLC34A2(ex13):ROS1(ex32)	FAIL	no rearrangement detected

CD74, CD74 molecule; EZR, ezrin; FISH, fluorescent in situ hybridization; GOPC, golgi associated PDZ and coiled-coil motif containing EZR; NGS, next-generation sequencing; RNA-QC, RNA quality control; SDC4, syndecan 4; SLC34A2, solute carrier family 34 member 2; TFG, TRK-fused gene; ZCCHC8, zinc finger CCHC-type containing 8.

^aManufacturer information: Abbott Molecular, Abbott Park, Illinois; Agilent, Santa Clara, California; Cytocell, Cambridge, United Kingdom.

RNA-Based Next Generation Sequencing

This assay was validated and performed in the Colorado Molecular Correlates Laboratory in the Department of Pathology at the University of Colorado – Anschutz Medical Campus. Total nucleic acid (TNA) was extracted from FFPE processed material via the Agencourt FormaPure Kit (Beckman Coulter, Brea, CA). TNA was then processed via the Archer FusionPlex Solid Tumor library preparation kit (ArcherDx, Boulder, CO). The resulting libraries were sequenced on an Illumina MiSeq instrument using v3 chemistry (Illumina, San Diego, CA). Raw sequence data were then analyzed by using the Archer Analysis software package (version 4.1.1.7; ArcherDx). Bioinformatically identified fusions were verified to be in-frame by manual inspection of the breakpoints.

DNA-Based Next Generation Sequencing

This assay was designed by and performed in the laboratory of Dr. Robert C. Doebele in the Department of Medicine - Division of Medical Oncology at the University of Colorado – Anschutz Medical Campus. Total genomic DNA (gDNA) was isolated from tissue samples via the QIAamp DNA Mini Kit (Qiagen, Hilden, Germany) according to the manufacturer’s specifications. The gDNA was then sheered to 300bp using a Covaris S220 Focused-ultrasonicator (Covaris, Woburn, MA). DNA libraries with unique barcodes were constructed using the Kapa Hyper Prep Kit (Kapa Biosystems, Wilmington, MA). Samples were pooled and subjected to a hybridization-based targeted enrichment via a custom-designed pool of bait probes specific to selected exons and introns of 48 genes (NimbleGen SeqCap EZ Choice Library, Roche, Pleasanton, CA). Sequencing was performed on an Illumina NextSeq 550 System (Illumina). Raw sequence data were analyzed by the Genomic Short Read Nucleotide Alignment Program (GSNAP) and structural rearrangement analysis via the ‘Clipping REveals STructure’ (CREST) algorithms.

RESULTS

In this study, three commonly used approaches for rearrangement/fusion testing were evaluated (Table 1). Inclusion in the tested cohort required a positive ROS1 result via any of the approaches plus testing (regardless of result) by at least one of the other approaches. Much of this cohort was comprised of samples that initially tested positive by break-apart FISH and were then retrospectively assessed by the NGS-based assays (as initial FISH testing preceded availability of the NGS assays in many instances). Thus, this study should not be considered a prospective screening analysis.

Break-apart FISH was performed in several CAP-accredited/CLIA-approved settings. Since multiple laboratories were involved, several versions of the assay were employed. The specific FISH probes used for each sample are described in Table 1. For samples that tested positive by FISH, the percentage of positive cells and predominant staining pattern are also noted in Table 1. The RNA-based NGS assay was also performed in a CAP-accredited/CLIA-approved setting. Library preparation was achieved via the ArcherDx FusionPlex Solid Tumor kit, which employs anchored multiplex PCR chemistry [20]. For ROS1 fusion-positive cases, the exons of both ROS1 and the fusion partner that formed the breakpoint in the transcript are noted in Table 1. The DNA-based NGS assay was not validated in a CAP/CLIA setting. Library preparation for the assay was achieved via a hybrid-capture technique using a custom design panel through the Roche SeqCap system. For samples that tested positive for ROS1 rearrangement, the intron or exon in which the breakpoint was found for both ROS1 and the partner is noted in Table 1. Importantly, for all three assays performed at the University of Colorado, no false positive ROS1 results were observed during validation or routine application. Thus, specificity for all three assays is calculated to be 100%.

FISH was performed successfully on 20 of the patient samples. Results were negative in 2 of the cases, but in both cases a ROS1 fusion was clearly detected by both RNA-based and DNA-based NGS assays (and the fusion partner genes were concordant between these assays). In one of the false-negative cases, patient #4, the fusion partner predicted in both NGS assays was GOPC. GOPC resides very close to ROS1 on chromosome 6, and the fusion is created by a genomic deletion that results in a breakpoint between introns of the two genes. A schematic that approximates Abbott FISH probe binding regions and the genomic deletion detected in the sample (determined via the DNA-based NGS finding) is shown in Fig 1A. Importantly, this relatively small deletion leaves enough of the complimentary regions for hybridization of both of the Abbott FISH probes (which were used for testing this patient sample). Thus, when FISH is performed, the two probes still bind and are not separated, thus appearing like a rearrangement-negative sample (Fig 1B). For the other false negative, the result was a complex FISH pattern with numerous nuclei showing atypical doublet fusion signals and only a few nuclei with split signals (below the 15% cutoff) (Fig 1C). The case appeared abnormal but did not meet established criteria for a positive FISH score and was called negative.

Figure 1 – Deficiency of FISH to Detect GOPC-ROS1 Fusion.

A) Schematic depicting Abbott FISH probe hybridization regions and the genomic deletion detected in patient #4 (determined via detected genomic breakpoints in GOPC and ROS1). B) FISH results from patient #4 showing no evidence of rearrangement. C) FISH results from patient #18 demonstrating atypical fusion doublets along with a low percentage of nuclei with split signals.

The RNA-based NGS assay was used to test 19 of the samples. In three cases, a ROS1 fusion was not detected, even though FISH results were clearly positive in all three and the DNA-based NGS assay detected a fusion in two (Table 1). However, in all three of these cases, an important quality control (QC) metric in the RNA-based assay was not met. In anchored multiplex PCR, one end of each fragment in the library is variable due to the random nature of cDNA generation combined with the ‘anchored’ nature of the assay [20]. Assessment of the number of different fragment end sites (termed ‘start sites’ in the assay) is an informative measure of library complexity, because unique cDNA molecules put into the assay will generally have different start sites. The assay used in this study contains primers to 4 different housekeeping genes (genes expected to be expressed in every cell type in the body). The analysis algorithm determines the number of unique start sites in reads from each of these primers and then calculates an average. If this value is less than 10, then the sample is deemed to have poor quality RNA. In the laboratory that performed the assay in this study, samples in which this QC metric are not met are deemed uninformative and not reported as negative if no fusion is called (however a positive fusion call can still be reported if the QC metric is not met). Thus, the 3 samples in which no ROS1 fusion was detected were actually deemed uninformative and therefore were not true false negatives. Actual start site data from the control genes is demonstrated in Table 2. Values from two of the samples (from patients #6 and #10) that failed the QC metric are shown in comparison to a sample that passed the QC metric (patient #14). In addition, the QC status of every sample assessed by the assay is noted in Table 1 under ‘RNA-QC’. Importantly, several instances of a positive ROS1 fusion result occurred in a sample that failed QC. These positive results were deemed reportable (based upon additional assay metrics).

Table 2.

Start Site Data From the Control Genes

	Start Sites per Primer
Primer	Patient #6	Patient #10	Patient #14
CHMP2A exon 3	0	0	36
CHMP2A exon 4	0	3	83
GPIexon 15	1	3	93
GPI exon 16	1	11	132
RAB7A exon 3	0	3	52
RAB7A exon 4	0	11	137
VCP exon 14	0	4	56
VCP exon 15	0	9	67
Average	0.25	5.5	82.0
QC Status	FAIL	FAIL	PASS

CHMP2A, charged multivesicular body protein 2A; GPI, glucose-6-phosphate isomerase; QC, quality control; RAB7A, member RAS oncogene family; VCP, valosin containing protein.

The DNA-based NGS assay was performed on 18 of the samples. In 4 of these samples, no ROS1 fusion was detected despite clear positivity in the FISH or RNA-based assays. When examining sequencing coverage of the ROS1 introns in this assay, it became apparent that in certain regions coverage was less than complete. Although complete coverage of introns was requested during the design phase of the assay, the presence of repetitive DNA, namely two large long interspersed nuclear elements (LINEs), in intron 31 precluded bait coverage of all desired regions. Fig 2 directly demonstrates this deficiency. In this figure, sequencing coverage is denoted by stacks of gray and colored bars in the Integrated Genomics Viewer (IGV) screenshots, with approximate positions of LINEs represented by black bars beneath the screenshots. Fig 2A demonstrates sequencing coverage of patients #15 (top) and #1 (bottom) in ROS1 intron 31. Both patients were positive for rearrangement/fusion in all three assays, and the genomic breakpoints in ROS1 intron 31 are denoted by the blue arrows in Fig 2A. In these two cases, the breakpoints occurred in regions of intron 31 that were sufficiently covered. Fig 2B demonstrates intron 31 coverage from the assay performed on patient #23. This sample was positive in RNA-based NGS but negative in DNA-based NGS. The breakpoint at the transcript level from the RNA-based assay included ROS1 exon 32 (Table 1). While this does not unequivocally prove that the genomic breakpoint occurred in ROS1 intron 31 (see below), it is likely that this is the case. However, since much of the intron was not sequenced, it is highly plausible that the breakpoint occurred in a poorly-baited and thus unsequenced region. Fig 2C demonstrates that in ROS1 intron 34 complete sequencing of the intron was achieved (this figure also denotes a detected breakpoint for patient #4). Thus, the difficulty in achieving complete intronic sequencing coverage adversely affected the negative predictive value of the DNA-based assay, as it is possible that breakpoints occurred in unsequenced regions.

Figure 2 – DNA-Based NGS Sequence Coverage.

IGV screenshots denoting sequencing coverage from the assay are shown (gray and colored bars stacked as well as gray histograms above stacks denote areas of sequencing coverage). Blank areas in the introns suggest no sequencing coverage. Blue arrows indicate breakpoints of detected fusions. A) patients #15 (top) and #1 (bottom), B) patient #23, and C) patient #4. Exons are denoted by thick blue bars at the bottom of each IGV image. The black bars in A and B represent approximate positions of two large LINE elements, L1PA8A and L1PA10 (identified via the UCSC Genome Browser).

Repetitive intronic sequence also created complications for the fusion calling algorithm used in the DNA-based NGS assay (CREST). A parameter in this software tool called “max repetitive coverage” is designed to ignore regions in which sequencing reads are consistently misaligning. In this study, a region of ROS1 intron 33 near exon 33 frequently misaligned during mapping, and, consequently, reads in this region were not being analyzed properly by CREST using default settings. It was only by changing the max repetitive coverage setting from 500 to 2000 that this region was adequately assessed by the program. Consequently, for the samples from patients #3, #8, #10, and #20, all of which demonstrated genomic breakpoints in ROS1 in intron 33 or exon 33, the fusion was only detected upon this change in the settings.

A critical step when assessing potential pathogenicity of fusions detected via the RNA-based assay is ensuring that the ROS1 component remains in-frame in the detected transcript (fusions in which the ROS1 component is out-of-frame would not be expected to be oncogenic). In theory, this logic could be applied to DNA-based NGS results as well, by assuming that the nearest exons 3’ to the ROS1 breakpoint and 5’ to the partner breakpoint are spliced together. For many of the samples in our cohort that were positive in both RNA and DNA-based assays, this assumption was true, and the nearest exons to the genomic breakpoints were involved in the transcript (Table 1). However, in three of the samples (patients #2, #4, and #19) the ROS1 breakpoint at the transcript level did not match that predicted by the genomic breakpoint. Importantly, in these three cases, the predicted transcript based on genomic breakpoint would not have been in-frame, but the actual detected transcript from the RNA-based assay was in-frame. Therefore, it is likely that alternative splicing that removed an exon and resulted in an in-frame transcript was employed by the cells. Additionally, for two of the samples (patients #15 and #18) positive for the SLC34A2-ROS1 fusion, DNA-based NGS demonstrated a genomic breakpoint in the 3’ untranslated region for SLC34A2. Consequently, the predicted transcript would not be expected to produce a functional chimeric protein, either due to inclusion of the SLC34A2 stop codon or removal of the ROS1 component during polyadenylation. Inspection of the actual breakpoints (demonstrated via the RNA-based assay) in these transcripts though revealed mid-exon breakpoints in SLC34A2 exon 13. Thus, alternative splicing was likely utilized to create an in-frame fusion transcript that did not contain the SLC34A2 stop codon. Two additional SLC34A2-ROS1 fusions (in patients #11 and #23) in which exon 13 of SLC34A2 was involved also demonstrated mid-exon breakpoints in the transcript. In conclusion, genomic breakpoint is an unreliable predictor of breakpoint at the transcript level.

DISCUSSION

Accurate detection of ROS1 rearrangements/fusions in tumor samples from NSCLC patients with advanced disease is critical to ensure that these patients receive optimal therapy. There is currently a wide-variety of testing approaches and platforms available to perform ROS1 rearrangement/fusion testing, each with inherent strengths and weaknesses. In this study, we compared three of the most prevalent approaches and identified limitations of each assay that led to missed calls. While the sample size was not large enough to accurately identify the most sensitive approach, the RNA-based NGS assay proved to be the most reliable assay within this cohort when RNA quality was considered (there were no false negative results for samples that passed QC metrics).

Break-apart FISH has been the gold-standard assay for ROS1 rearrangement detection, and the assay played a vital role in early studies that determined prevalence of these events in lung cancer and in the initial clinical trials of crizotinib [2–4, 21]. However, several limitations to the assay have been described including the possibility of false-positives due to unproductive rearrangements, significant slide-to-slide heterogeneity, aberrant probe hybridization leading to false positives, high levels of background noise, and lack of ability to identify the fusion partner [22]. In this study, one false negative result appeared to be a consequence of the inability of certain FISH probes to detect rearrangements that result from small genomic deletions. A deletion on chromosome 6 in this sample left enough of the hybridization regions for both the 5’ and 3’ FISH probes to bind. Thus, despite being clearly positive for a GOPC-ROS1 fusion via both NGS-based assays, the sample appeared rearrangement negative by FISH (Fig 1). The potential for false negative FISH results in cases of GOPC-ROS1 fusion has also been noted previously [23, 24]. Another false-negative result occurred in a case of a complex staining pattern in which many atypical fusion doublets were noted but the percentage of cells with the typical split signals was below cutoff (15%). Thus, this case provides an example of why orthogonal methodologies should be employed for instances of atypical results.

Perhaps the most critical limitation of the break-apart FISH assay is the limited degree to which rearrangements in multiple genes can be assessed simultaneously. A growing list of actionable or potentially actionable gene fusions coupled with the fact that lung cancer biopsies tend to be small samplings has resulted in a scenario in which use of single-gene assays is not an efficient approach for clinical laboratories. Thus, break-apart FISH is becoming obsolete as a primary assay for ROS1 rearrangement detection. NGS, on the other hand, is ideally suited to query multiple genes simultaneously, and therefore NGS-based assays have been developed in many clinical settings to identify rearrangements/fusions. Though a wide variety of NGS-based approaches for fusion testing are available, a primary distinguishing factor among these is whether RNA or DNA is used as the starting input material.

Since most (but not all) genomic breakpoints that lead to gene fusions occur in introns, DNA-based NGS assays designed to detect rearrangements/fusions must sequence introns. However, introns are known to frequently contain repetitive sequences that are difficult to assess by NGS [25]. As a result, a problematic deficiency of DNA-based NGS is the possibility that genomic breakpoints may occur in intronic regions that cannot be properly sequenced. In the DNA-based NGS assay employed in this study, sequencing of ROS1 intron 31 was less than complete due to the presence of two large LINEs that contain repetitive DNA sequence, precluding complete bait coverage. Therefore, it was impossible to know whether or not breakpoints occurred in unsequenced regions of the intron. While this assay was not validated in a CAP/CLIA setting and therefore cannot be considered a clinical assay in its current form, other DNA-based NGS assays are likely to be at least somewhat hampered by this complication.

RNA-based NGS has an advantage over DNA-based NGS in that sequencing can be focused on coding sequences instead of introns. However, the primary drawback of this approach is the high reliance on RNA quality, which can be quite poor in clinical samples, particularly those that are FFPE processed. Therefore, any fusion assay that uses RNA as the input material must have some associated measure of RNA quality. In the assay employed in this study, a post-sequencing metric is calculated that indicates library complexity of four house-keeping genes. Failure of this metric to achieve a defined cutoff is indicative of poor quality RNA, and precludes interpretation of negative results. In this study, all 3 cases of failed ROS1 fusion detection were associated with failure to achieve this cutoff, thus these results were interpreted as uninformative and not true false-negatives.

Previous studies that have compared multiple different ROS1 testing paradigms have also come to the conclusion that disparate approaches are generally not 100% concordant. In a study by Shan et al., FISH, IHC, and RT-PCR were compared on a cohort of 60 lung adenocarcinoma samples [26]. While concordance between the assays was generally good, many discrepancies were observed. Notably, three samples that were ROS1 FISH-negative in this study were positive by both IHC and RT-PCR. No clear reason for the false-negative results could be determined. Wu et al., also compared these three methodologies and found that while ROS1 FISH and RT-PCR results were concordant, concordance of IHC was dependent on staining intensity and the percentage of cells that stained positive [27]. In another study by Rogers et al., FISH was compared to an RNA-based NGS approach (ThermoFisher AmpliSeq), an approach that utilizes targeted capture and reporter probes (NanoString Elements), and an expression imbalance MassARRAY approach (Agena LungFusion) [28]. Two ROS1 FISH positive samples in this study were also positive by the other assays; however, a positive result from the NGS assay and a positive result from the MassARRAY assay were not confirmed by the other approaches. Reasons for the discrepancies were unclear.

In conclusion, this study demonstrates that break-apart FISH, RNA-based NGS, and DNA-based NGS each have inherent deficiencies that can lead to false negative results in the testing for ROS1 rearrangements/fusions. It is imperative that interpretation of results from these assays take these limitations into consideration. Ideally, molecular testing laboratories should employ orthogonal assays for rearrangement/fusion testing. While it is perhaps unfeasible to test all samples by two or more different assays, testing schemas that take into consideration other information from the sample and/or patient can be employed. For example, since ROS1 fusions generally (though not universally) occur exclusively of other known oncogenic drivers, ROS1-negative samples found to be negative for other driver oncogenes should be tested by an orthogonal assay to ensure that false-negatives for ROS1 (and other rearrangements/fusions) are avoided [29]. It is through such mindful approaches that accurate molecular diagnoses of tumor samples are achieved.