Custom Gene Capture and Next Generation Sequencing to Resolve Discordant ALK Status by FISH and IHC in Lung Adenocarcinoma

Abstract

Background

We performed a genomic study in lung adenocarcinoma cases with discordant anaplastic lymphoma kinase (ALK) status by fluorescent in situ hybridization (FISH) and immunohistochemistry (IHC).

Methods

DNA from formalin-fixed paraffin embedded (FFPE) tissues of 16 discordant (4 FISH+/IHC− and 12 FISH−/IHC+) cases by Vysis ALK Break Apart FISH and ALK1 IHC testing were subjected to whole gene capture and next generation sequencing (NGS) of nine genes including ALK, EML4, KIF5B, SND1, BRAF, RET, EZR, ROS1, and TERT. All discordant cases (except one FISH−/IHC+ without sufficient tissue) underwent IHC by D5F3 antibody. In one case with fresh frozen (FF) tissue, whole transcriptome sequencing (RNAseq) was also performed. Twenty-six concordant (16 FISH+/IHC+ and 10 FISH−/IHC−) cases were included as controls.

Results

In four ALK FISH+/IHC− cases, no EML4-ALK fusion gene was observed by NGS but in one case using FF tissue, we identified EML4-BIRC6 and AAK1-ALK fusion genes. RNAseq revealed a highly expressed EML4-BIRC6 fusion transcript and a minimally expressed AAK1-ALK transcript. Among the 12 FISH−/IHC+ cases, no evidence of ALK gene rearrangement was detected by NGS. Eleven of 12 FISH−/IHC+ cases by ALK1 clone were concordant by a repeat ALK IHC with D5F3 antibody (i.e. FISH−/IHC− by D5F3 clone). Among the 16 ALK FISH+/IHC+ positive controls, whole gene-capture identified ALK gene fusion in 15 cases including one case with HIP1-ALK. No ALK fusion gene was observed in all 10 FISH−/IHC− cases. Other fusion genes involving ROS1, EZR, BRAF and SND1 were also found.

Conclusions

ALK FISH results appeared to be false positive in 3 of 4 FISH+/IHC− cases, while no false negative ALK FISH case was identified among 12 ALK FISH−/IHC+ cases by ALK1 clone, in keeping with the concordant FISH−/IHC− status by D5F3 clone. Our targeted whole gene capture approach using FFPE samples was effective for detecting rearrangements involving ALK and other actionable oncogenes.

Introduction

Fluorescent in situ hybridization (FISH) using break-apart probe for the anaplastic lymphoma kinase (ALK) gene has been generally regarded as the gold standard for determining ALK status of lung cancers for using ALK inhibitors. However, the true sensitivity and specificity of the FISH method for detecting ALK gene rearrangement are not well known. Various genomic abnormalities involving the ALK gene could be associated with abnormal signal patterns seen in ALK FISH and with the abnormal protein expression detected by ALK immunohistochemistry (IHC). Neither FISH nor IHC method is perfect and a third method using other molecular techniques may be required to determine the true ALK fusion status in some cases. Clinical experience has indicated that ALK status determined by FISH may not always correlate with responsiveness to ALK inhibitors and that a testing algorithm with combination of FISH, IHC, and reverse transcriptase-polymerase chain reaction (RT-PCR) might be a better approach to select the patients for ALK inhibitor therapy.¹

ALK analysis by FISH using the Vysis LSI ALK break apart probe (Abbott Molecular, Abbott Park, IL) has shown that ALK FISH signal patterns can be quite complicated. It is especially difficult to prove the positive ALK FISH result as false positive when a case is positive for FISH and negative for IHC or RT-PCR, due to the limited sensitivity of RT-PCR and IHC; ALK RT-PCR would not detect previously unidentified fusion transcript(s) and ALK protein expression in lung cancers might be low. Thus, it is usually ruled that the negative RT-PCR/IHC result as false negative in those with a positive ALK FISH result. On the other hand, a few cases of false-negative FISH results have been documented in the literature.^2–5

There has been an increasing need to develop clinically applicable multiplex assays that could maximize the yield of molecular tests on FFPE tissues from small lung biopsies. Although multiplexed genomic assays have been applied to determine ALK fusions at the DNA and RNA levels, published reports are still sparse in the literature.^6,7 The rapid development of technologies for large-scale sequencing (such as NGS) has facilitated high-throughput molecular analysis for large numbers of genes in a single test for a simultaneous detection of deletions, insertions, copy number alterations, rearrangements, and single base substitutions (hot-spot mutations) in all known cancer-related genes.⁸ Currently, NGS platforms, including whole genome, whole exome, and targeted gene sequencing, represent emerging diagnostic methodologies for the detection of oncogene fusions and mutations in FFPE as well as in FF tumor tissue specimens.^9,10

The primary purpose of this study was to examine the molecular basis for cases with discordant ALK status by FISH and IHC. We also wished to establish a scalable method to detect all ALK-rearranged and other targetable fusion events in patients who may benefit from targeted gene specific inhibitor therapies. Herein, we report the use of a NGS-based whole gene capture approach using DNA from FFPE samples and comprehensive bioinformatics analyses to successfully detect rearrangements involving ALK and other oncogenes in lung cancers as a promising diagnostic method applicable to routine clinical practice.

Materials and Methods

Cases and ALK assessment

This study was approved by Institutional Review Board of Mayo Foundation. All discordant cases were identified from our historical cohort of resected lung adenocarcinoma cases. Selected concordant cases were included as the positive and negative controls for NGS. A total of 45 cases were initially were analyzed further for fusions involving ALK and other targeted oncogenes. Three of these 45 cases were excluded due to failed gene capture and the remaining 42 cases were included in the present report. ALK IHC was performed with ALK1 antibody (Dako, 1:100 dilution) and scored as 0, 1, 2 or 3 as previously described.¹¹ ALK IHC score 0 and 1 were regarded as negative (IHC−) and ALK IHC score 2 and 3 as positive (IHC+) as in our previous study.¹¹ Interphase molecular cytogenetic studies using Vysis Break Apart ALK Probe Kit (Abbott Molecular) were performed on FFPE sections and scored as either positive (≥15%) or negative (<15%) as previously described.¹¹ Hundred tumor cells were counted in each case.

Discordant cases underwent repeat IHC with a different ALK antibody (clone D5F3, Cell Signaling Technology, Danvers, MA). Four micron sections cut from formalin-fixed, paraffin-embedded blocks were placed on charged slides; slides were then dried and melted in a 62°C oven for twenty minutes. Immunohistochemical staining was performed as follows: slides were placed on a Ventana BenchMark XT (Ventana Medical Systems Inc., Tucson, AZ) for staining. The staining protocol included on-line deparaffinization, HIER (Heat Induced Epitope Retrieval) with Ventana Cell Conditioning 1 for 32 minutes, primary ALK antibody incubation for 32 minutes at 37°C (clone D5F3, a rabbit monoclonal antibody, 1:100 dilution, Cell Signaling Technology, Danvers, MA). Antigen-antibody reactions were visualized using Ventana Optiview Universal DAB Detection Kit. Counterstaining was performed on the Ventana BenchMark XT using Ventana Hematoxylin II for eight minutes, followed by bluing reagent for four minutes. The positive control was a known ALK-rearranged lung adenocarcinoma and the negative control was a mouse IgG1 serum substitution for the primary ALK antibody.

Whole gene capture for NGS

A custom capture kit (Agilent) was designed to span the entire genomic sequence of 9 genes (based upon hg 19) including ALK, EML4, KIF5B, SND1, BRAF, RET, EZR, ROS1, and TERT for target-specific gene capture and fusion identification by NGS. The total captured region was approximately 1.6 Mb. DNA was extracted from lung adenocarcinoma using FFPE tissues where two to three 10μ-thick sections per case with tumor purity of 40–95% were used to extract the genomic DNA using the Qiagen AllPrep DNA/RNA FFPE kit (Quiagen). Paired-end indexed libraries were prepared using the NEBNext Ultra Library prep protocol (NEB). Briefly, up to 500 ng of FFPE genomic DNA in 50 μl TE buffer was fragmented using the Covaris LE210 sonicator to generate double-stranded DNA fragments with blunt or sticky ends at a fragment size mode of between 150–200bp. The ends were repaired using the NEB End-prep enzyme mix after which looped Multiplex NEBNext Adaptors for Illumina were added using the NEB Blunt/TA Ligase Master Mix and linearized with USER enzyme. Adapter-ligated DNA fragments were size-selected to enrich for 200 bp inserts (~320 bp total library size) using a double SPRI bead purification method followed by 6 cycles of PCR using NEB Universal PCR Primer and NEB Index Primer. The concentration and size distribution of the libraries were determined using Agilent Bioanalyzer DNA 1,000 chips.

Custom capture was carried out using the Agilent Bravo liquid handler according to a modified protocol for Agilent’s SureSelect XT. Four samples were pooled equally for a total of 750 ng of the prepped libraries. The pools were incubated with the custom biotinylated capture baits supplied in the kit for 24 hours at 65°C. The captured hybrids were recovered using Dynabeads MyOne Streptavidin T1 from Dynal. The DNA was eluted from the beads and purified using Ampure XP beads from Agencourt. The purified capture products were then amplified using the SureSelect Post-Capture Indexing forward and Index PCR reverse primers (Agilent) for 12 cycles. Libraries were quantified and sequenced as 101 × 2 paired end reads on an Illumina HiSeq 2000 using TruSeq SBS sequencing kit version 3 and HiSeq data collection version 2.0.12.0 software. On average, 73 million reads were generated for each sample, the minimal unique reads per sample ranged between 10 to 80 million with most samples having 20–40 million reads each.

Whole transcriptome sequencing using fresh frozen tissue

We used 100ng of RNAs from both tumor and matched adjacent normal tissue to generate libraries using TruSeq™ RNA sample preparation V2 protocol (Illumina, San Diego). Sequencing was performed on a HiSeq 2000 instrument with a 51 cycles pair-end read. FASTQ formatted raw files were mapped and aligned to hg19. Fusion transcripts were identified using SnowShoes-FTD version 2.0.¹² A fusion was called when there was at least two reads that span the fusion junction and three reads that encompassing the two fusion transcripts.

Data Analysis and Fusion Detection in Targeted Sequencing

A. CREST – identification of fusions utilizing split reads

Alignment and pre-processing

We aligned the reads to the hg19 reference genome using two different alignment programs, Burrows Wheeler Aligner (BWA,v 0.5.9)¹³ and NovoAlign (VN:V2.08.01).¹⁴ The default options were used for BWA. NovoAlign penalties were set for structural variation (120), gap opening (40), and gap extend (5). NovoAlign fragment length mean (425) and standard deviation (80) were set. Duplicate reads were marked by Picard.¹⁵

Detection and filtering of Structural Variants

Structural variants were detected using a localized search option implemented in CREST.¹⁶ The CREST algorithm detects structural variation by identifying split reads (reads that can be split into two segments that align to disjoint genomic locations) from soft-clipped reads. For each alignment and each gene, we ran the CREST program with the localized search interval set according to gene definition. Duplicate reads, events with coverage greater than 1,000,000 reads, and redundant structural variation predictions were removed. To identify the top candidates from among the structural variants detected by CREST, we considered the quantity and robustness of support, requiring that either there was at least 1 supporting soft-clipped read for each breakpoint in one alignment or that the event was detected in both the BWA and NovoAlign alignments. Further, we reviewed each of the events satisfying these criteria by applying the BLAT (BLAST-like alignment tool) to the consensus sequences determined by CREST.

Annotation, Summarization, and Prioritization of Structural Variants

The CREST program indicates the type of structural variant detected (e.g. translocation, inversion, insertion, duplication, deletion) as well as orientation, breakpoints, and consensus sequence across the breakpoint. We used the UCSC refFlat gene annotation table to further annotate the breakpoint regions according to the transcriptional boundaries of the RefSeq genes and the RefSeq gene region (e.g. intron, exon, intergenic, etc).¹⁷ For structural variant predictions involving two breakpoints that are within the transcript regions of separate genes, a fusion transcript could potentially result from the structural variant. From the consensus sequence provided by CREST, we determined the orientation of the fusion transcript. Essentially, when two segments of genomic DNA are joined, with each segment from a gene region, there are four possibilities with respect to the two gene segments. For genes A and B, the first case is that the 3′ end of gene A segment is joined to the 5′ end of the gene B segment. We use the notation A->B in this case. The second case is that the 5′ end of gene A segment is joined to the 3′ end of the gene B segment. This is represented by B->A. The third case is that the 5′ end of gene A segment is joined to the antisense strand of the 5′ end of the gene B segment. This is represented in our notation as <-A:B->. This is essentially equivalent to <-B:A-> in that if the DNA SV supports <-B:A-> it also supports <-A:B->. The fourth case is that the 3′ end of gene A segment is joined to the antisense strand of the 3′ end of the gene B segment. This is represented in our notation as A-><-B. This is equivalent to B-><-A, ie if the DNA SV supports B-><-A it also supports A-><-B. The orientations of the detected DNA SV’s in this work are provided in Supplemental Table 1. For all detected DNA SV’s involving two genes, we report an orientation class as defined by these 4 cases described above. In the case of a detected DNA SV’s for which one breakpoint is in a gene region and the other is in an intergenic region, there is no orientation class to assign, but we do report these events as well.

B. Structural Variant Analysis tools (SVAtools) identification of fusions utilizing both discordant read pairs and split reads

Alignment and pre-processing

Using a binary indexing mapping algorithm¹⁸ (BIMA), we mapped all read-pairs to the reference genome GRCh38 using a concordant insert size limit of 5000 bps. BIMA transforms the reference genome to a binary index allowing for rapid mapping. High accuracy is achieved by simultaneous mapping of both reads in a read-pair.¹⁸ BIMA is tuned to detect and report read-pairs that map to two discontinuous genomic areas, including when a single read crosses a breakpoint (split read) or a biotin-junction (common in NGS mate-pair library preparation).

Detection and filtering of Structural Variants

Structural variants were detected using a suite of algorithms, SVAtools, designed specifically for detecting structural variants in read-pair data mapped by BIMA.^19,20 SVAtools is an in-house R package, developed by the Biomarker Discovery Lab at Mayo Clinic. Briefly, all read-pairs are sorted by mapped chromosome and position for rapid indexing. Replicate read-pairs are removed. A customized rapid clustering algorithm detects read-pairs and split reads supporting a common breakpoint. A minimum of three read-pairs is required to form a cluster. These clusters must pass a system of masks and filters. The mask eliminates normal structural variants not present in the reference genome, eliminates mapping artifacts due to repeat or un-sequenced genomic regions and eliminates artifacts due to NGS library preparation. The filters use cluster size and BIMA mapping scores to identify poorly qualified breakpoints and false positives. Default settings for the filters were applied except as follows: –protocol pairedEnd –radius 3000 –mismatch_lim 10 – –homCountmin 20 –homf 1.75. Only clusters with at least 8 read-pairs were considered for structural variant detection, and only breakpoints involving genes, with at least one target gene were considered for this study. Genomic regions of suspected structural variation can be inspected and visualized via junction plots (illustrations of all read-pairs mapping within and between two genomic regions). Junction plots served as a useful tool to double-check for false-negatives, to reveal clusters of read-pairs that were incorrectly removed during masking or filtering. All structural variants detected via these methods are listed in supplement table 2. The interpretation of each rearrangement follows the notation described above in the CREST annotation.

Results

Fusion Identification by Whole Gene Capture

Discordant cases were comprised of 4 cases that were positive by FISH but negative by IHC using ALK1 clone (all IHC score 1) and 12 cases that were negative for FISH but positive for IHC using ALK1 clone (all IHC score 2) (Table 1). ALK IHC using D5F3 was also negative in all 4 FISH+/IHC− cases as using ALK1, while the ALK IHC+ cases by ALK1 clone in 11 of 12 FISH−/IHC+ cases tested by D5F3 clone were IHC− (i.e. the 11 tested cases were concordant by D5F3 clone as FISH−/IHC−) (Table 2). The percentages of cells with abnormal signals in FISH were 15–27% in the 4 FISH+/IHC− cases and 0–7% in the 12 FISH−/IHC+ cases (Table 2). Concordant cases (16 FISH+/IHC+ and 10 FISH−/IHC−) were used as controls (Table 1). No fusion gene involving either ALK or EML4 was identified in the 3 FISH+/IHC− cases using FFPE DNA for NGS while 1 case using FF tissue showed AAK1-ALK and EML4-BIRC6 fusion genes by both fusion detection algorithms (Figure 1). Whole gene capture and sequencing analysis with alignment algorithms revealed no ALK fusion in any of the 12 FISH−/IHC+ cases. Of note, one of 12 FISH−/IHC+ cases showed CD74−>ROS1 (Table 2). Of the 16 FISH+/IHC+ cases, 13 cases showed DNA structural variant support for EML4->ALK fusions by whole gene capture, while 1 case showed support for the reciprocal fusion ALK->EML4 fusion. One other case had a fusion involving Huntington-interacting protein-1 (HIP1) and ALK genes. The remaining one FISH+/IHC+ case did not show any evidence for a fusion by gene capture; this case had a borderline FISH positivity (15%) and IHC was also weakly positive (score 2). Repeated IHC with D5F3 clone was negative. Thus, both the original FISH and IHC results appeared to be a false positive based on the absence of identifiable ALK fusion by targeted gene capture and NGS. There was one FISH+/IHC+ case that revealed a novel SND1-BRAF fusion gene. No DNA support was found for a fusion involving the ALK gene in any of the 10 ALK FISH−/IHC− cases. However, we detected an SND1->BRAF fusion gene involving exons in two and EZR->ROS1 in one of these 10 cases. All identified fusions are listed in Table 2.

Table 1.

Summary of ALK IHC and FISH

	Score 3 IHC+	Score 2 IHC+	Score 1 IHC−	Score 0 IHC−	Total
FISH +	11	5	4	0	20
FISH −	0	12	6	4	22
Total	11	17	10	4	42

Values in bold indicate samples with discordant FISH/IHC scores.

Table 2.

Fusion identification by CREST and SVAtools.

Sample ID	Tumor Percent	ALK FISH (%)	ALK IHC (ALK1)	ALK IHC (D5F3)	CREST*	SVAtools**
Y31	N/A	− (1)	1	N.T	–	–
Y32	N/A	− (6)	1	N.T	–	EZR-><-RAB3C
Y33	N/A	− (2)	0	N.T	–	–
Y34	70%	− (N/A)	0	N.T	–	–
Y35	N/A	− (9)	1	N.T	EZR->ROS1 (10)	EZR->ROS1 (4)
Y38	80%	− (2)	0	N.T	–	–
Y39	90%	− (N/A)	1	N.T	–	–
Y65	40%	− (N/A)	1	N.T	–	–
Y68	95%	− (N/A)	1	N.T	SND1->BRAF not detected by BWA-Novoalign/CREST; detected by BWAMEM/CREST (17)	SND1-> BRAF (11)
Y69	90%	− (N/A)	0	N.T	SND1->BRAF (9)	SND1-> BRAF (21)
Y56	60%	− (N/A)	2	N.T	–	–
Y63	50%	− (0)	2	0	–	–
Y72	60%	− (3)	2	0	–	–
Y73	80%	− (1)	2	0	CD74->ROS1 (40)	CD74-> ROS1 (79); ROS1->CD74 (11)
Y74	90%	− (7)	2	0	–	–
Y75	50%	− (5)	2	0	–	–
Y76	70%	− (0)	2	0	–	–
Y77	90%	− (5)	2	0	–	–
Y78	90%	− (1)	2	0	–	–
Y79	70%	− (N/A)	2	0	–	–
Y80	80%	− (1)	2	0	–	–
Y81	70%	− (5)	2	0	–	–
891A	N/A	+ (27)	1	0	AAK1->ALK (94); EML4->BIRC6 (160)	AAK1->ALK (152); EML4->BIRC6 (187)
Y20	N/A	+ (15)	1	0	–	–
Y53	70%	+ (15)	1	0	–	–
Y60	60%	+ (18)	1	0	–	–
Y02	90%	+ (N/A)	2	N.T	HIP1->ALK (20)	HIP1->ALK (33)
Y04	95%	+ (29)	2	N.T	EML4->ALK (42)	EML4->ALK (94)
Y05	90%	+ (60)	3	N.T	ALK->EML4 (29);<-ALK:EML4-> (24)	<-ALK EML4->(25); ALK->EML4 (21)
Y06	80%	+ (25)	2	N.T	EML4->ALK (13);ALK->EML4 (60)	EML4->ALK (43); ALK->EML4 (48)
Y07	60%	+ (35)	2	N.T	EML4->ALK (56)	EML4->ALK (99)
Y08	90%	+ (N/A)	3	N.T	EML4->ALK (25);ALK->EML4 (25)	EML4->ALK (46); ALK->EML4 (31)
Y12	N/A	+ (52)	3	N.T	EML4->ALK (13)	EML4->ALK (13)
Y15	60%	+ (N/A)	3	N.T	EML4->ALK (13)	EML4->ALK (9)
Y18	60%	+ (N/A)	3	N.T	EML4->ALK (87); ALK->EML4 (58)	EML4->ALK (98); ALK->EML4 (52)
Y22	50%	+ (70)	3	N.T	EML4->ALK (19)	EML4->ALK (42)
Y26	N/A	+ (72)	3	N.T	EML4->ALK (43)	EML4->ALK (49)
Y27	70%	+ (68)	3	N.T	EML4->ALK (23)	EML4->ALK (36)
Y52	N/A	+ (N/A)	3	N.T	EML4->ALK (28);ALK->EML4 (41)	EML4->ALK (76); ALK->EML4 (49)
Y54	70%	+ (15)	2	0	–	–
Y59	80%	+ (N/A)	3	N.T	EML4->ALK (30); SND1->BRAF not detected by BWA-Novoalign/CREST; detected by BWAMEM/CREST (17)	EML4->ALK (13); SND1-> BRAF(13)
Y61	80%	+ (N/A)	3	N.T	EML4->ALK (79)	EML4->ALK (56)

^*Number of supporting split reads is shown in parentheses.

^**Number of supporting read-pairs and split reads is shown in parentheses.

“+” FISH Positive ≥ 15%, “ − ” FISH or Fusion negative, N.T; not tested, N/A; not available

Figure 1.

Junction plots of identified fusion genes by SVAtools. EML4-ALK (A), balanced EML4-ALK (B), HIP1-ALK (C), CD74-ROS1 (D), EML4-BIRC6 (E), and AAK1-ALK (F). Junctions plots are a 2-panel representation of the junction of both breakpoints in a chromosomal rearrangement. Reads from paired-ends mapping to breakpoint A are plotted on the top panel, connected by a line to the corresponding paired-end read mapping to breakpoint B in the bottom panel. Genes and reads mapping to the (−) DNA strand are shown in red, while genes and reads mapping to the (+) DNA strand are shown in blue. The color and pattern of discordant read-pairs can be used to identify different types of rearrangements, including balanced or unbalanced translocations, deletions, and inversions.

Molecular Rearrangements Identified by RNAseq Using Fresh Frozen Tissue in a FISH+/IHC− Case

Whole transcriptome profiling (RNAseq) of the mRNA revealed a strong expression of the EML4-BIRC6 fusion gene (Figure 2A), in support of the results of the DNA-based whole gene capture. In contrast, RNAseq showed a low level AAK1-ALK fusion gene expression (Figure 2B), which would explain positive ALK FISH test by a break-apart probe as well as little or no ALK protein expression by ALK IHC (score 1 by ALK1; score 0 by D5F3). Both RNAseq and the gene capture did not identify the EML4-ALK fusion in this case. As was observed in our previous study on this case,²¹ evidence of EGFR mutation was observed in the expressed transcripts by RNAseq (Figure 2C).

Figure 2.

Fusion structures of EML4-BIRC6 (A) and AAK1-ALK (B) genes by RNASeq and custom capture DNA sequencing. EGFR gene mutation is shown in (C).

Genomic Rearrangements Accompanying Oncogene Fusions

By whole gene capture of 9 genes, we were able to observe associated DNA rearrangements that accompanied the functional oncogenic fusions (Supplemental tables 1–3). Several junction plots were also illustrated (Figure 1). At least 4 cases had detectable reciprocal DNA rearrangements from the same genetic event that likely led to the EML4->ALK fusion while others involved ALK->EML4 or a fusion involving 5′ ALK and another gene on either chromosome 2 or another chromosome. Details of genetic events resulting in the HIP1->ALK fusion gene found in a FISH+/IHC+ positive control (case Y2) are listed in the supplemental Tables 2 and 3.

Several other tumors had DNA rearrangements involving one or more targeted oncogene but appeared to not generate an active oncogene fusion of the target genes. As described earlier, one of the 16 FISH+/IHC+ positive cases did not demonstrate any evidence of EML4 ->ALK but there was strong evidence of ALK->EML4 fusion indicating the presence of genomic rearrangements in the region leading to the ALK rearrangement and protein expression (supplemental Tables 1–3).

Discussion

A French multicentric study reported that discordances in ALK status between FISH and IHC were mainly found in the cases with FISH results ranging from 10% to 20% of rearranged cells.²² Three of our 4 FISH+/IHC− cases also fell in this range (15–18%), in keeping with the results in that study. A recent study showed that approximately 6% of ALK FISH cases belong to a borderline group for which ALK FISH evaluation has of limited reliability possibly due to sampling effects.²³ Thus, those cases should be considered equivocal and therapy decisions should include additional tests and clinical considerations.²³

A recent multicenter ALK IHC testing of non-small cell cancer showed that a high concordance after standardization of techniques and interpretation criteria.²⁴ ALK IHC using ALK1 clone might not be reliable, especially without using an appropriate detection system, but ALK IHC using D5F3 or other clones can provide reliable and reproducible results as shown in the previous studies.^11,25–30 There are many different clones of ALK antibodies for IHC.

Discordant cases in the present study were based on the cohort of cases stained with ALK1 clone in our database predating D5F3 clone. We repeated immunostaining on all discordant cases (except one case without available tissue) with D5F3 antibody, which has been shown to correlate better with ALK FISH.^27,31 In all four FISH+/IHC− cases, the negative IHC result by ALK1 antibody was confirmed by D5F3 antibody. On the other hand, the positive IHC result (all score 2) by ALK1 antibody reversed as negative by D5F3 antibody in 11 of 11 FISH−/IHC+ cases, where sufficient tissue was available for repeat staining. Thus, this FISH−/IHC+ group most likely have been concordantly ALK negative (i.e. FISH−/IHC−), which would explain the absence of detectable ALK gene rearrangements by NGS in this group.

Three of four FISH+/IHC− cases were negative for any ALK gene rearrangement by the genomic approach, which challenged the positive ALK FISH result in these cases. As mentioned earlier, the positive ALK FISH results in these 3 cases were borderline (15–18%), which might explain the discordances as observed in the recent French study.^22,23 ALK FISH in the remaining 1 FISH+/IHC− case showed break apart probes in 27% of tumor cell nuclei; this case showed BIRC6->EML4 and AAK->ALK fusions by both NGS of DNA and RNAseq. In our previous study²¹, we found EML4–ALK fusion by RT-PCR followed by sequencing in this case using FFPE. However, RNAseq of the mRNA on the FF tissue in the current study did not detect any EML4-ALK gene rearrangement and only demonstrated a minimal AAK-ALK mRNA expression. Rather, we identified a strong mRNA expression of the EML4-BIRC6 fusion gene (Figure 2A), which was not available at that time. As was shown in our previous study²¹, RNAseq in this study analyses supported EGFR del L747-S752 in exon 19 (Figure 2C). Given the questionable state of EML4-ALK gene rearrangement as well as the lack of ALK protein expression (i.e. IHC-by both ALK1 and D5F3 clones), EGFR mutation was most likely the driver mutation for the adenocarcinoma in this case.

As controls, we examined 26 concordant (16 FISH+/IHC+ and 10 FISH−/IHC−) cases by targeted gene capture-based NGS approach. The results of concordant ALK positive cases (FISH+/IHC+) were confirmed by NGS except one case that did not show any fusion by gene capture. This case had a borderline FISH positivity (15%). The original ALK IHC staining with ALK1 clone was also weakly positive (score 2) and a repeat IHC with D5F3 clone was negative. Thus, the original results of both FISH and IHC were likely false positive in this case, which would explain the absence of identifiable ALK fusion gene on NGS. Moreover, this case has been tested for KRAS gene and showed mutation in G12D (data not shown), which further supports the negative NGS result as ALK gene rearrangement and KRAS mutation are known to be mutually exclusive. NGS confirmed the results of all concordant ALK negative control cases (FISH−/IHC−).

Our study demonstrated the potential of NGS using archival FFPE samples as a diagnostic tool in detecting rearrangement involving multiple genes. We used two bioinformatics approaches by CREST and SVAtools to detect rearrangement(s) involving multiple genes with more detailed and comprehensive information as to the identity and the orientation of gene rearrangement. While independently analyzed, both methods were highly concordant and provided us with high confidence on discordant as well as concordant cases for ALK status by FISH and IHC.

However, the gene capture methods may still have false calls or missed rearrangements that are present at low frequencies and could not be distinguished from potential fusion artifacts, low level contamination, or background noise associated with NGS library prep and passing the filters set forth by the bioinformatics calling algorithms as might have been the case in one of our FISH+/IHC− case (case 891A); on examination of IGV of RNAseq, there might be some expressions in the exon 20 of ALK fused with EML4 (in addition to the exon 16 of ALK fused with AAK), which both bioinformatics in the present study might have failed to detect. The heterogeneity of mutation status within the same tumor sample may also present challenges to genomic based analysis³².

Finally, it is worth noting that our gene-capture approach identified fusions associated with genomic rearrangements for the genes of interest, regardless of their ability to give rise to an overexpressed oncogene. While this approach requires further evaluation in a much larger cohort of samples, the accompanying genomic rearrangements we identified in this study provided additional clues to enable more sensitive and specific assessment of the fusion status in a clinical sample. Among the cases with genomic evidence supporting the EML4->ALK gene fusions, we observed the reciprocal fusion ALK->EML4 fusion in four cases (Table 2). In one ALK FISH+/IHC+ case, however, gene capture only observed fusions involving ALK-> EML4 and/or <-ALKEML4-> but not the EML4->ALK fusion; the genomic evidence from the other four cases with both EML4->ALK and the reciprocal fusion ALK->EML4 fusions suggests that this case is probably also ALK positive. Unfortunately, the patients in our historic cohort have not been treated with ALK inhibitors, which did not allow us to evaluate its biologic implication as an oncogenic driver or relation to responsiveness to the therapy.

By NGS approach, we found several other oncogene fusions that have not yet been reported or only recently identified. We found 1 of 16 ALK FISH+/IHC+ tumors harboring the Huntington-interacting protein-1 (HIP1)-ALK fusion gene, which has been recently identified as a new fusion partner of ALK in non-small cell lung cancer^33–35, adding to the four currently known fusion partners (EML-4, KIF5B, TFG, KLC1). Both CREST and SVAtools identified a CD74-ROS1 fusion among the 12 ALK FISH−/IHC+ group and an EZR-ROS1 fusion among the 10 FISH−/IHC− negative group. Finally, we observed an SND1-BRAF fusion in three cases, one of which also showed an EML4-ALK fusion gene. A previous study also has reported an SND1-BRAF fusion along with other driver gene mutations including EGFR and ERBB2.³⁶

Taken together, our genomic study demonstrated that our 3 ALK FISH+ (all at borderline 15–18%)/ IHC− cases were likely false positive by FISH. Both gene capture with NGS and RNAseq methods demonstrated that the remaining FISH+/IHC− case harbored AAK1-ALK and EML4-BIRC6 fusions, both of which have not been documented in the literature. While requiring further evaluation using a much larger cohort of samples, our DNA based genomic approach using hybrid gene capture using FFPE tissues could be a powerful tool to detect ALK gene rearrangement as well as other molecular alterations of multiple genes as a single test.