Quantification and localization of oncogenic receptor tyrosine kinase variant transcripts using molecular inversion probes

Oncogenic membrane receptor tyrosine kinases such as MET and EGFR, or auto-active variants thereof, are important targets for cancer precision therapy. Targeted inhibition of these oncogenic receptors however invariably leads to resistance, resulting from acquisition of resistance-inducing mutations or from selective outgrowth of a priori resistant tumour cells. Most applied molecular protocols cannot distinguish between intracellular and intercellular heterogeneity of oncogene (variant) expression, which may lead to misinterpretation of the molecular make-up of a cancer and suboptimal application of targeted therapies. We here combined two related techniques to allow semiquantitative and localized in situ detection of specific transcript splice variants using single molecule molecular inversion probe (smMIP)-based next generation sequencing and padlock probe-based rolling circle amplification, respectively. We show highly specific padlock probe-based multiplex detection of MET, METΔ7-8 and METΔ14 transcripts, lacking exons 7–8 and exon 14 respectively, and of EGFR and the auto-active EGFRvIII, lacking exons 2–7. The combination of quantitative transcript variant detection with smMIPs and transcript localization using padlock probes can be used for detection of oncogenic transcripts on the single-cell level, allowing study of tumour heterogeneity. Visualization of tumour heterogeneity can shed light on the biology underlying drug resistance and potentially improve targeted therapeutics.

A second example of oncogene aberration is EGFRvIII. Whereas EGFR is amplified in 50% of glioblastomas, half of these events is accompanied by an intragenic deletion, resulting in EGFR transcripts lacking exons 2-7 9,17 . Protein products of this transcript lack a large part of the extracellular domain and are auto-active. Because the EGFRvIII protein contains a neo-epitope at the exon 1-8 junction, it is tumour specific and therefore of considerable interest for targeted therapies 18,19 . In glioblastoma, expression of this variant emerges after initial EGFR amplification and is heterogeneous, being more prominent in diffuse infiltrating areas of the brain 17 . Importantly, this spatial heterogeneity cannot be inferred from genetic analysis. This highlights the need for expression analysis on the single-cell level.
In this study, we combine single molecule molecular inversion probe (smMIP) targeted transcriptome sequencing 20,21 (Fig. 1) with in situ detection of transcript splice variants using padlock probe-based rolling circle amplification [22][23][24][25][26] (Fig. 2). We show that we are able to specifically and semiquantitatively detect transcript variants in RNA samples from both cell lines and xenografts using smMIPs, and we demonstrate how in situ padlock probe rolling circle amplification adds essential information on transcript localization and tumour heterogeneity.

Results
To investigate whether padlock assays can be used for specific in situ detection of oncogene transcripts, we here examined expression of MET and EGFR and auto-active splice variants of these tyrosine kinases, in a cohort of cell lines and xenografts with defined EGFR and MET expression. To set up the technique we used cell lines E98 and its corresponding xenograft (E98-FM, MET Δ7-8 amplification) 8 , cell lines Hs746T (MET Δ14 , high expression), Figure 1. Quantitative detection of transcript splice variants using smMIPs. (a) Schematic overview of the method. mRNA is reverse transcribed into cDNA using random primers. SmMIP probes hybridize to the target region of interest, leaving a gap of ~112 nucleotides. The gap is filled by primer extension and ligation, whereafter all remaining linear smMIPs, RNA and cDNA are removed by exonuclease treatment. All circular smMIPs are PCR amplified using a unique barcoded reverse primer for each sample. Resulting reads are mapped against reference transcriptomes, normalized to the total read count within a sample (FPM), and averaged per transcript variant. (b) Design of smMIP probes. To determine the total expression of a gene, smMIPs were designed against shared parts of the transcript of interest (upper panel). For specific detection of splice variants smMIPs were designed to target variant-specific exon-exon junctions (for example the exon6-9 junction for MET Δ7-8 ), either covering the junction with the gap or the ligation/extension probe (lower panel left and right, respectively). Note that for graphical representation, the figure is not to scale. H596 (MET Δ14 , low expression), U87-EV (EGFR wt , low expression), U87-EGFRvIII (over-expression) 27 , and xenograft E468 (EGFR amplification) 28 . Relative protein expression levels of MET and EGFR in the different cell lines and xenografts are shown in Fig. 3a. By PCR analysis we confirmed the MET transcript status of E98, Hs746T and H596, and EGFR transcript mutation status of E468, U87-EV and U87-EGFRvIII (Fig. 3b). We semiquantitatively determined transcript variant-specific expression levels of MET and EGFR in all cell lines and xenografts using smMIPs (Fig. 3c).
We first tested the specificity of the MET-targeting padlock probes on cell lines with known MET status. Single-probe assays with padlock probe MET 7,8 resulted in fluorescent signal in Hs746T, but not in E98 cells. Conversely, single-probe assays with padlock probe MET 6-9 resulted in fluorescent signal in E98, but not in Hs746T cells (Fig. 4a). These results are in line with the fact that E98 cells exclusively express MET Δ7-8 , while Hs746T cells only express METexon7-8 wt transcripts, confirming the specificity of MET 7,8 and MET [6][7][8][9] probes. Similarly, we performed single-probe padlock assays with the MET 14 and MET 13-15 probes on E98 cells in which no MET Δ14 is expressed, and Hs746T and H596 cell lines, expressing MET Δ14 . As expected, in E98 cells METexon14 wt transcripts were readily detected with probe MET 14 , while the assay was negative when performed with the MET 13-15 probe (Fig. 4b upper panel). Conversely, Hs746T and H596 cells were positive when assayed with the padlock MET [13][14][15] probe, and negative with the MET 14 probe (Fig. 4b middle and lower panels). Numbers of fluorescent spots in H596 cells were lower than in Hs746T cells, in agreement with the lower levels of MET protein that we detected by western blot in these cell lines (Fig. 3a) and lower levels of MET transcript as detected by smMIP-sequencing (Fig. 3c).
Having confirmed the specificity of these probes, we subjected E98 and Hs746T cells to a combination of MET 7,8 and MET 6-9 probes, and MET 14 and MET 13-15 probes, respectively. This duo-probe padlock assay allows In situ detection of whole-exon deletion splice variants using padlock probes. (a) Schematic overview of the method. mRNA is converted into localized cDNA molecules, which can be targeted through padlock probe hybridization and rolling circle amplification. Rolling circle products (RCPs) can then be detected with fluorescently labeled probes. Nucleic acid bridges in the BNA primer are depicted in green, padlock probe target sites are depicted in blue and padlock probe detection sites are depicted in red. (b) Design of padlock probes targeting METexon7-8 wt , MET Δ7-8 , METexon14 wt , MET Δ14 , EGFR wt and EGFRvIII. For specific detection of transcript splice variants, padlock probes were designed to target the mutation-specific exon-exon junction (detected with magenta fluorescence). For distinctive detection of their associated wild-type variants, padlock probes were targeted at the exon(s) missing in the exon deletion transcripts (detected with green fluorescence). Note that for graphical representation, the figure is not to scale.
SCIeNtIfIC RepoRts | (2018) 8:7072 | DOI:10.1038/s41598-018-25328-5 duplex detection of wild-type (detected in green) and mutant variants (detected in magenta) of the same RNA transcript simultaneously. The selectivity of the padlock probes was preserved when mutant-specific probes were applied together with their corresponding wild-type probe, and also the efficiency was not notably affected (Fig. 5).
After successful and specific detection of MET splice variants in cell lines, we tested our duo-probe padlock assay on FFPE xenograft material. We subjected serial E98-FM FFPE tissue slides, carrying the MET Δ7-8 mutation but with a wild type exon 14 allele, to duo-probe padlock assays using a combination of probes MET 6-9 and MET 7,8 (Fig. 6b), MET [13][14][15] and MET 14 (Fig. 6c), or MET 14 and MET [6][7][8][9] (Fig. 6d). H&E sections serial to the padlock assay tissues are depicted in Fig. 6a. Both MET Δ7-8 and METexon14 wt transcripts were specifically detected in E98 xenograft tissue, without detection spots for METexon7-8 wt or MET Δ14 transcripts, respectively (Fig. 6b,c). We could visualize different tumour areas with high and low expression of the targeted MET transcripts (Fig. 6b,c, left and middle panels, respectively). Note that normal tissue (right part of right panels) was completely negative. A double-positive padlock assay targeting METexon14 wt and MET Δ7-8 on E98 xenograft tissue showed specific detection of both transcripts simultaneously (Fig. 6d). Probe specificity is again highlighted at the tumour border ( Fig. 6d, right panel). Although both padlock probes were readily identified in these sections, co-localization was only occasionally observed (see Supplementary Fig. 2, arrows). This is possibly due to the low sensitivity of the padlock probe assay. Of note, signals were slightly decreased in the double-positive assay as compared to the single-positive padlock assays. Although we could not quantify this, we attribute this to competition of padlock probes on the same transcript. Similar to the padlock assays targeting MET, we performed padlock assays targeting EGFR wt and EGFRvIII transcripts. We subjected U87-EV (EGFR wt ) and U87-EGFRvIII overexpressing cell lines to a duplex assay to detect EGFR wt and EGFRvIII transcripts, using padlock probes EGFR 2,3 (green) and EGFR 1-8 (magenta), respectively. In U87-EV cells this assay resulted in fluorescent signal for EGFR wt , while EGFRvIII specific transcripts were abundantly detected in U87-EGFRvIII overexpressing cells (Fig. 7a). In U87-EGFRvIII cells no EGFR wt signals were detected in the duplex assay, probably because of the low abundance of EGFR wt transcripts in these cells, compared to high overexpression of EGFRvIII transcripts. Note that the number of EGFR wt fluorescent spots in U87-EV cells is much lower than the number of EGFRvIII spots in U87-EGFRvIII cells, which corresponds to the difference in EGFR expression (Fig. 3a,c). We also subjected EGFR wt expressing E468 xenograft tissue to a duplex EGFR padlock assay. Figure 7b shows specific detection of EGFR wt transcripts.

Discussion
To be able to investigate tumour heterogeneity with respect to oncogene expression, we applied padlock probe rolling circle amplification to indentify EGFR, MET and the whole-exon deletion splice variants of these oncogenes that play a role in, among others, glioma and NSCLC 1,5,16,[29][30][31] . Simultaneous in situ detection of tyrosine kinase splice variants with their associated wild-type variants enables visualization of tumour heterogeneity in FFPE tissue sections, material which is routinely used in diagnostics and can be preserved for years.
In recent years, articles have reported on the use of padlock probes for in situ analysis of single-nucleotide transcript variants 22,24 and fusion transcripts 23 . Here, we adapted these protocols to allow in situ identification of whole-exon deletion splice variants. Probe specificity was validated in single-probe assays on cell lines with known MET or EGFR status, and was preserved in duplex padlock assays. As for now, a drawback is low sensitivity of padlock probe assays, possibly explaining the lack of co-localization of MET 14 and MET 6-9 probes in Fig. 6d. The distinction between intercellular and intracellular tumoural heterogeneity, with different oncogene variants being present within the same cell or with subclones of cells carrying different oncogene variants, respectively 2,3,32 , can therefore not be made. This distinction is important since cells co-expressing different tyrosine kinases that initiate similar signalling pathways, will respond differently to drugs than cells expressing only the targeted kinase. Low sensitivity of padlock probe assays has been described before 22,24,25 . However, the reason is not clear and may be related to any of the complex steps in the procedure, including fixation method/RNA quality, RNA secondary structures that hamper effective in situ reverse transcription, and the presence of single nucleotide polymorphisms (SNPs) impeding hybridisation of the BNA probe and the padlock probe. Improvement of sensitivity of the method is necessary to detect co-expression of different transcripts in one cancer cell. However, the high specificity of the padlock assay will allow the investigation of intratumoural, intercellular heterogeneity of splice variant transcript expression. To allow successful padlock assays on clinical tissue, the essentiality of RNA integrity probably requires that tumour tissue is fixated shortly after surgery or biopsy in proper preservatives.
For smMIP-based targeted RNA sequencing we have observed a variance in efficacy between different smMIPs targeting the same transcript. We have reported that the average of five smMIPs targeting one transcript gives a reliable estimate of the expression level of a transcript 21 . Because padlock assays rely on the hybridisation of one individual probe for each transcript variant, these are not suitable for quantification of transcript (variant) expression levels. In situ rolling circle amplification using padlock probes does allow visualization of intercellular heterogeneity within tumour tissues. In contrast, smMIP-based targeted RNA sequencing allows quantitative and specific detection of splice variants, but does not enable visualization of tumour heterogeneity and localization of specific transcripts. The combination of both techniques possibly allows differentiation between intrinsic and acquired transcript expression (induced tumour heterogeneity), and between intrinsic and acquired treatment resistance development. This is an interesting aspect for future studies and may give more insight into tumour  resistance biology. This potentially has implications for targeted therapy, because patients with pre-existent resistant tumour clones may benefit significantly from combination therapies at start.
In conclusion, we here show the high value of combining two molecular inversion probe techniques. Using smMIP targeted transcriptomics we can quantitatively measure transcript variant-specific gene expression. Additionally, padlock probe rolling circle amplification enables specific and local visualization of transcript variants in situ, however with low efficacy. The technique of smMIP-based RNA splice variant sequencing can be easily expanded with smMIP probes to detect other transcript variants, and can serve as an important tool for detection of molecular heterogeneity and investigation of the role of splice variants in therapy resistance.

Materials and Methods
Cell lines and xenografts. Cell lines Hs746T and H596 were obtained from ATCC (Manassas, VA). Cell lines E98 28 , Hs746T, H596, U87-EV and U87-EGFRvIII (obtained from dr. Web Cavenee, Ludwig Cancer Inst., USA 27 ) were cultured at 37 °C in the presence of 5% CO 2 in Dulbecco's Modified Eagle's Medium (DMEM) containing 4.5 g/L glucose and 4 mM L-glutamin (Lonza, Basel, Switserland), 10% fetal calf serum (FCS, Gibco, Waltham, MA, USA) and 40 µg/ml gentamycin (Centrafarm, Ettenleur, The Netherlands). The MET Δ7-8expressing astrocytoma cell line E98 and the EGFR-expressing astrocytoma line E468, as well as the generation of orthotopic xenografts thereof, has been described before 8,28 .  PCR. RNA was isolated from cell lines and snap-frozen xenograft tissue using TRIzol reagent (Life Technologies, ThermoFisher Scientific, Waltham, MA, USA) and reverse transcribed with Superscript II (Invitrogen, ThermoFisher Scientific, Waltham, MA, USA) using random hexamer primers, according to the manufacturer's instructions. RT-PCR was performed using Amplitaq Gold 360 mastermix (Applied Biosystems, Life Technologies), which for PCR on MET was supplemented with 1 mM MgCl 2 and 20 µg/ml BSA (both New England Biolabs, Ipswich, MA, USA). To distinguish MET Δ7-8 transcripts from their wild-type variant (METexon7-8 wt ), primers MET1997Fw (5′-CTCCTTGGAAATGAGAGCTG-3′, forward, located in exon 6) and MET2393Rv (5′-AGATGCTTGTCTCTCGGTTG-3′, reverse, located in exon 9) were used. This PCR results in a 397-bp amplicon for METexon7-8 wt and a 157-bp product for MET Δ7-8 . To distinguish MET Δ14 transcripts and their wild-type variant (METexon14 wt ) primers MET2982Fw (5′-CAGGATTGATTGCTGGTGTTGTCTC-3′, forward, located in exon 13) and MET3252Rv (5′-CGGCATGAACCGTTCTGAGATG-3′, reverse, located in exon 15) were used, resulting in a PCR product of 271 bp for METexon14 wt and 130 bp for MET Δ14 . To detect EGFR wt transcripts primers EGFR819Fw (5′-GATATCACCATGCGACCCTCCGGG-3′, forward, located in exon 5) and EGFR1567Rv (5′-CGACTGCAAGAGAAAACTGA-3′, reverse, located in exon 12) were used, resulting in a 749-bp product for EGFR wt . For detection of EGFRvIII transcripts in U87-EGFRvIII cells, primers EGFR258Fw (5′-GATATCACCATGCGACCCTCCGGG-3′, forward, located in exon 1) and EGFR1567Rv (5′-CGACTGCAAGAGAAAACTGA-3′, reverse, located in exon 12) were used, resulting in a 529-bp product for EGFRvIII. PCR conditions were: initial denaturation of 3 min at 95 °C, followed by 36 cycles of denaturation at 95 °C, 30 s; annealing at 60 °C, 30 s; elongation at 72 °C, 30 s, and a final elongation step of 5 min at 72 °C. smMIP targeted RNA sequencing. RNA from cell lines and snap-frozen xenograft tissue was isolated and converted to cDNA as described under 'PCR' . The smMIP method and its application for targeted transcriptome sequencing has been described before 20,21 and is depicted in Fig. 1a. SmMIPs were designed against target regions of interest (UCSC human genome assembly hg19, and variant-specific FASTA sequences) with ligation and extension probes localized on different exons, depending on the splice variant to be detected, and leaving a gap of maximum 112 nt. SmMIPs were designed based on the MIPgen algorithm as described by Boyle et al. 33 and added to a large panel of previously designed smMIPs 21 . The panel of phosphorylated smMIPs was hybridized to cDNA after which the gap was filled by primer extension and ligation. Non-reacted smMIPs and remaining RNA and cDNA were removed by exonuclease treatment, followed by PCR amplification of the circularized smMIP library using a unique barcoded reverse primer for each sample. After library pooling and purification using AMPureXP beads (Beckman Coulter Genomics, High Wycombe, UK) smMIP-PCR libraries were sequenced on the Illumina Nextseq platform (Illumina, San Diego, CA) at the Radboudumc sequencing facility. Reads were mapped against the reference transcripts (UCSC human genome assembly hg19 and variant-specific FASTA sequences) using the SeqNext module of JSI SequencePilot version 4.2.2 build 502 (JSI Medical Systems, Ettenheim, Germany). The random 8xN nucleotide tag flanking the ligation probe was used to reduce PCR amplicates to one consensus read originating from the same smMIP (unique read). Read counts for each smMIP were normalized to the total read count within a sample and multiplied by 10 6 (Fragments per Million, FPM). Individual transcript levels were expressed as mean FPM of all smMIPs targeting that transcript.
SmMIPs were designed against transcript variants of MET (total MET, MET Δ7-8 , MET Δ14 ) and EGFR (total EGFR, EGFRvIII). SmMIP sequences are depicted in Table 1. An example of smMIP design for total MET (transcript variant 2) and MET Δ7-8 is depicted in Fig. 1b. To determine the cumulated expression of all MET transcript variants, smMIPs were designed against shared parts of the transcripts (total MET). To enable detection of truncated transcripts, smMIPs were designed against splice variant-specific exon-exon junctions, either with the maximum 112 nt gap or the ligation/extension probe covering the exon-exon junction (exon 6-9 junction for MET Δ7-8) . SmMIPs were designed in a similar way for MET Δ14 , total EGFR and EGFRvIII.
Padlock probe design. The protocol for padlock probe rolling circle amplification and the design of padlock probes with associated bridged nucleic acid (BNA) primers was adapted from Weibrecht et al. 26 and is summarized in Fig. 2a,b. Wild-type transcripts were distinctively detected with padlock probes targeting 'normal' exon-exon junctions that are missing in the exon-deletion transcript variants. For specific detection of splice variants, padlock probes were designed to target the splice variant-specific MET exon-exon junctions 6-9 (for detection of MET Δ7-8 ) and 13-15 (for detection of MET Δ14 ) and EGFR exon-exon junction 1-8 (for detection of EGFRvIII, Fig. 2b). To allow dual detection of splice variants together with their associated wild-type transcripts, specific sequences were included in the backbone of the padlock probe allowing detection via hybridization of complementary fluorescent probes. Bridged Nucleic Acid (BNA) primer, padlock probe and detection probe nucleotide sequences, as well as the nomenclature of probes and transcript variants used in this paper, are outlined in Table 2. Before use, padlock probes (IDT, Leuven, Belgium) were 5′-phosphorylated using 0. (DEPC)-treated PBS. Tissue sections were permeabilized with pepsin (2 mg/ml in 0.1 M HCl) and dehydrated through an ethanol series (70%, 85% and 99.5%). Secure-seals (Grace Bio-Labs inc., Bend, Oregon, USA) with a volume of 250 µl (Ø 20 mm, 0.8 mm deep) were mounted and cells were rehydrated with DEPC-PBS-Tween. All following incubation steps were performed within the secure seal chambers, incubated in a hybridizer (Dakocytomation, Glostrup, Denmark).
In situ reverse transcription was performed using BNA primers positioned within ≤1 nt from the 5′ end of the padlock probe target site (see Table 2 Table 2. BNA primer, padlock probe and detection probe nucleotide sequences. BNA primer Bridged Nucleic Acids are depicted in bold. Padlock probe target-specific sequences are depicted in italic. Detection probe complementary sequences in the padlock probe are coloured bold italic (wild-type) or italic underline (mutant).