Christina Stangl

The goal however, would be to increase the number of clinical intervention studies based on the fusion-status of the patients. Currently, clinicians are limited in their freedom to treat these patients based on the observed genetic abnormality, as indication labels of cancer drugs are not only cancer-type specific but also mutation specific, e.g. vemurafenib for BRAF V600E mutations or trastuzumab for HER2 overexpressing tumors. Off-label cohort studies like the Drug Rediscovery Protocol (DRUP study) are excellent opportunities to increase the number of cancer patients that are treated according to their genomic or molecular profile, e.g. fusion status. These studies will provide evidence on the feasibility to include fusion gene status as an indication for certain pathway inhibiting cancer drugs.

Fusion-positive patients – treatment guidance through fundamental research
Fusion gene interpretation mostly remains a challenge for clinicians. Apart from singled-out examples such as the NTRK fusions in solid cancers and highly recurrent and investigated fusions in hematological cancers (e.g. BCR-ABL1 in CML), there are no clear guidelines on how to treat patients with gene fusions, especially not if tumor types are factored in. As discussed earlier, the main reason for this is a lack of (pre)clinical data to substantiate potential treatment choices. It will require time and more fundamental research to gather this data.

However, lessons can be learned from in vitro studies of recurrent fusion gene partners, which are known to have mutually exclusive equivalent SNV events (e.g. a BRAF fusion vs. a BRAF V600E mutation) (Figure 1). Pathways involved in differentiation and proliferation, such as the AKT, MAPK and PLC pathways, are commonly deregulated in cancer. This is facilitated through mutations in key gene players within these pathways, which are often kinases that get constitutively activated or overactivated. These molecular changes offer opportunities to target these acquired abnormalities e.g. through specific kinase or pathway inhibitors or to refrain from treatment due to known resistance mechanisms. In the case of metastatic CRC patients for example, EGFR-inhibition is avoided if mutations in MAPK key genes such as KRAS or NRAS are detected, as these are known to induce resistance to the targeted intervention. In Chapter 3 we show that all BRAF fusions, similar to mutually exclusive SNV events in the MAPK pathway, conveyed resistance to targeted monotherapy against the EGF receptor family in vitro. This suggests that BRAF fusions should be screened alongside other SNV-based MAPK pathway alterations to identify metastatic CRC patients to exclude from anti-EGFR targeted treatment. The fact that most of the recurrent fusion events are mutually exclusive with other mutations in the same pathway indicates that there is a substitutional role of the SV for the SNV event. As a consequence, concordant biological effects like pathway deregulation are expected, which likewise would suggest a similar treatment approach.

To pave the way to standardized treatment protocols for fusion-positive patients, a prospective clinical trial could be envisioned. To reach sufficient numbers, this would preferable be a multi-center basket trial including fusion-positive metastatic patients in a tumor agnostic manner with no available suitable standard-treatment options left, similar to the DRUP study. In this context, fusion-positive patients could be treated beyond the EMA and FDA approved indications based on a combination of parallels to mutually exclusive SNV events and preclinical evidence (Figure 1).

Figure 1. Guided treatment of fusion-positive patients based on equivalent SNV events or preclinical evidence
To identify if a patient presenting with a fusion event can potentially benefit from an off-label drug treatment, several steps need to be taken. (A) First, mutual exclusivity to an oncogenic SNV equivalent must be confirmed. (B) Secondly, data on pathway activation of the fusion gene must be in concordance with the pathway activation and/or response to targeted therapies of the SNV equivalent. (C) If the fusion gene is mutually exclusive with the SNV event and activates the same pathway, treatment from SNV-positive patients could be adopted to fusion-positive patients. (D) If no treatment protocols from equivalent SNV events or no equivalent SNV events are available, treatment choices could be extrapolated from preclinical research results such as pathway activations and drug screenings.

In a first instance, it should be investigated if a patient harbors a fusion gene that has a mutually exclusive known oncogenic SNV equivalent (Figure 1A). In a next step, preclinical information of the fusion event such as effects on oncogenic pathway activations or in vitro or in vivo evidence of drug sensitivities or resistances should be collected (Figure 1B). More specifically, if a patient presents with a fusion gene that has a mutual exclusive SNV equivalent, e.g. a BRAF fusion vs. a BRAF V600E mutation, and preclinical data is available that BRAF fusions, similar to the BRAF V600E mutation, activate the MAPK pathway and/or are sensitive to MAPK pathway-inhibitors, this should give clinicians - in the context of a clinical trial - an incentive to treat the patient as if the patient would present with the SNV equivalent (Figure 1C). It has to be noted that inhibitors that specifically act on or bind point-mutations such as vemurafenib (specific to BRAF V600E mutations) cannot be adopted for respective fusion events, but pathway-specific drugs need to be employed instead. The existence of a mutually exclusive SNV event for which clinical data is available provides extra confidence to initiate a certain treatment regimen, however, treatments could also be administered based solely on pathway activations or drug screening data (Figure 1D).

Taken together, while it is still too early to make definite conclusions on how fusion partners influence drug sensitivity, we can extrapolate information from mutually exclusive SNV equivalents and molecular observations such as activation of oncogenic pathways to guide treatment choices and design clinical trials to broaden our understanding of how to provide best care to fusion-positive cancer patients.

SVs - BIOMARKERS FOR MRD TRACING
Somatic SVs are mostly unique molecular events which can be used to specifically identify tumor cells or lineages. As discussed in Chapter 4, Chapter 5, and Chapter 6, the specific breakpoint junctions of SVs are optimal biomarkers to trace tumor dynamics within e.g. blood. These SV events may be functionally relevant driver events for tumorigenesis such as fusion genes (Chapter 4), biological relevant SVs such as Ig and TCR rearrangements within a subset of immune cells (Chapter 5), or passenger SV events that have no obvious impact on the functionality of a cell (Chapter 6). The unique genomic breakpoint junctions within fusion genes or Ig/TCR rearrangements serve as tumor-specific biomarkers which can be utilized for MRD tracing and the applicability of SVs as biomarkers for PCR-based tracing methods has been extensively shown. The bigger challenge however remains, until today, to identify suitable SV events from tumor material in a clinically relevant time-frame.

Nanopore sequencing is ideally suited for SV detection and offers the opportunity to perform real-time sequencing, and hence, could be a solution to provide SV biomarkers in a timely manner. Furthermore, hands-on time to prepare the sequencing libraries is minimal (between 10 mins to 3 hours) and the actual sequencing process takes less than 48 hours. As sequencing can be performed through relatively small devices, each lab could potentially perform sequencing in-house, as opposed to outsourcing to sequencing or core facilities. However, few labs, especially diagnostic labs, have implemented nanopore sequencing so far. This reservation stems from the fact that it is a relatively young technology that is in constant development and Oxford Nanopore Technologies (ONT) has not yet produced a stable product line which can be used for diagnostic accreditation. Nonetheless, we here show in Chapter 4, Chapter 5 and Chapter 6 the immense potential of nanopore sequencing to provide adequate SV-based genomic MRD biomarkers within less than 72 hours.

Bioinformatic challenges
In Chapter 4 and Chapter 6, we provide the bioinformatic pipelines to achieve the respective underlying goal: NanoFG to detect fusion genes from targeted nanopore sequencing data or SHARC to detect somatic SVs from low coverage nanopore sequencing of tumor only material. While these tools work well within their given parameters, there are still improvements to be made to make them more versatile. NanoFG (Chapter 4) for example, only reports fusion genes if the breakpoint in both fusion partners falls within annotated gene regions (including promoter regions, 5’UTR and 3’UTR). Other common rearrangements which do not follow this criterion, such as translocations between enhancer regions and genes (e.g. IGH/MYC translocations), will not be detected by NanoFG. To overcome this, NanoFG could be further developed so that it additionally identifies i) fusions where only one break falls within an annotated gene region (e.g. SIL-TAL1 fusions) and ii) fusions between gene segments and regulatory elements such as enhancers by linking NanoFG to known enhancer or regulatory sequence databases. SHARC (Chapter 6) efficiently identifies a subset of somatic SV events from a tumor sample. To enrich for somatic SVs without sequencing a corresponding germline sample, we here collected a number of germline SVs that we detected in in-house sequencing efforts, and combined these in a filtering database. However, this database contains only a limited number of samples. Building and expanding a comprehensive germline SV database could serve as a more advanced filtering base for somatic events but could also, with increased numbers, aid to unravel biological germline processes and patterns.

In Chapter 5 the bioinformatic challenges and shortcoming of nanopore sequencing with respect to read accuracy became more apparent. Possible rearrangements of the V(D)J segments within the Ig and TCR loci are extensive and highly variable. In a first instance we have opted to report all SV events within these loci - irrespective of the underlying V(D)J configuration - with the SV caller Sniffles. While this approach worked well and we could confirm 89% of the diagnostically provided MRD targets as well as additional potential targets, it would be desirable for downstream diagnostic efforts, such as gene segment-specific primer utilization, to not only identify SV events but immediately call the specific V(D)J configuration. One approach could be the extraction of SV-covering reads and using these to BLAST against a comprehensive V(D)J repertoire database such as IMGT. However, this will likely result in inconclusive results as the read accuracy of nanopore sequencing at the single nucleotide level is suboptimal and some V, D and J segments exhibit extensive homology, sometimes with only one base difference. To not rely on individual reads and overcome this hurdle, consensus calling could be performed, collapsing overlapping SV covering reads and therefore providing the most likely accurate sequence before performing a BLAST against a V(D)J database. Future experiments will have to evaluate if the high levels of homology between the Ig and TCR gene segments combined with the lower read accuracy of nanopore sequencing will actually cause problems, and if so, if consensus calling, consensus sequencing or subsequent Sanger sequencing can effectively resolve this issue.

Taken together, nanopore-based SV detection has multiple characteristics optimally suited for diagnostic implementation, however, there is room for improvement to comprehensively detect all types of fusion genes and genomic rearrangements from nanopore sequencing data. Advancing our current bioinformatic solutions or utilizing more accompanying databases to refine the analysis will most likely increase the accuracy of somatic SV detection. However, our current bioinformatic approaches already show great promise as we were able, in all cases, to provide the sequence of genomic biomarkers for subsequent MRD tracing within less than three days.

Optimal characteristics of biomarkers for MRD tracing
Key aspects of an optimal biomarker for MRD tracing is not only the ability to detect it in a clinically relevant time-frame but an adequate biomarker should additionally be i) tumor-specific, ii) present in the majority of the tumor cells, iii) stable, and iv) deliver specific and clear PCR results (Figure 2).

Figure 2. Characteristics of an optimal biomarker for MRD tracing
An optimal MRD biomarker should (A) be readily available, (B) be tissue-specific, (C) reflect a major clone of the tumor, (D) be stable, and (E) produce unambiguous PCR results. * highlights biomarker positive cells.

The assays we developed in Chapter 4, Chapter 5 and Chapter 6 fulfill most of the above mentioned requirements, while improvements could still be achieved. We provide genomic sequences of potential SV-based MRD biomarkers in a rapid manner, between 15-72 hours after initiation of the sequencing library preparation (Figure 2A). The tumor-specificity of a MRD biomarker can be established by performing SV-specific PCRs on the tumor material and germline controls (e.g. corresponding leukocytes or healthy donor leukocytes with polyclonal background in the case of Ig and TCR rearrangements) (Figure 2B). In Chapter 6 we validated 50% of the tested SV biomarkers as tumor-specific, despite only having sequenced and analyzed the tumor sample. Determining the clonality of a biomarker in a given sample (Figure 2C) is desirable, however not yet possible with our current sequencing approaches. To be able to estimate the clonality of an MRD event, information on the tumor purity of the input sample is required. Furthermore, this would allow, at least in Chapter 6, to make assumptions about the heterozygosity of an observed SV event. In Chapter 4 and Chapter 5 these extrapolations get complicated through the use of crRNAs to enrich for targeted genomic regions. We observe differences in the efficiency of crRNAs, influencing the amount of molecules that will be factually sequenced. Therefore, extrapolations on the sensitivity of the assays and the clonality of MRD events without making assumptions are impossible to make. Additionally, the optimal biomarker should be stable and not be influenced by expression levels or short-lived (Figure 2D). Hence, the biomarker should preferably be DNA as RNA molecules are easily degraded and influenced by the expression levels of the promoter and therefore do not accurately reflect the amount of tumor cells present in a sample. In Chapter 4 and Chapter 5, we detect genomic breakpoints of commonly used MRD targets (e.g. BCR-ABL1) of which current diagnostic efforts are only able to detect the transcriptomic breakpoints. This advantage can only be realized through the long-read sequencing capabilities of nanopore sequencing. Most diagnostic MRD techniques rely on PCR-based detection and quantification methods. Hence, to be able to immediately utilize an identified SV biomarker as a MRD target, successful and unambiguous PCR results are a necessity (Figure 2E). In Chapter 4 and Chapter 6 we extracted breakpoint-spanning primers provided by the developed bioinformatic pipelines and validated a large number of predicted SV events with definite PCR results. Furthermore, in Chapter 6 we show that nanopore sequencing provides breakpoint sequence information down to nucleotide resolution, allowing the subsequent design of breakpoint-specific probes for highly sensitive approaches such as digital PCR. In cases where the exact sequence around the breakpoint junction is unclear, Sanger sequencing of the PCR fragment may be performed. Due to the rapidness of identifying the possible MRD targets, ample time to perform extra validations or breakpoint-probe designs is provided.

Taken together we show that our Nanopore sequencing assays provide SV-based biomarkers which meet the criteria of diagnostically important characteristics. We identify relevant biomarkers within 15-72 hours that can be immediately utilized for downstream diagnostic applications such as MRD assay designs. The identification of genomic biomarkers within this short time-frame does not only enhance the chances that each patient is provided with an appropriate and personalized set of MRD targets, but also reduces the hands-on time and time-pressure put on diagnostic labs to provide these biomarkers within a clinically relevant time-frame.

SVS- CAUSE FOR CARE
Chapter 2 and Chapter 3, together with existing literature, clearly highlight involvement and functional influence of fusion genes on tumorigenesis and progression. Fusion genes are mutually exclusive with other driver mutations in oncogenic pathways (e.g. MAPK pathway), they are able to activate these pathways and their expression affects sensitivity of in vitro cell systems such as organoids to targeted therapies. Furthermore, their functionality may be influenced by the respective partner gene that they are fused to. Fusion genes occur at a relatively low frequency when compared to driver SNV events, but nonetheless affect a substantial amount of patients worldwide. Oncogenic fusion genes, of which we know or can predict the functional impact by deducting information from the involved partner genes or because they are recurrent events, make up a small number of all somatic SV events occurring throughout cancer types.

Any tumor accumulates somatic SV events throughout its development, of relevant or irrelevant nature for the tumor cell. In Chapter 4, Chapter 5 and Chapter 6 we leverage the occurrence of somatic SVs, including fusion genes, by exploiting their character as a unique genomic biomarker for tumor cells. We developed diagnostic assays to help improve patient care by having these biomarkers available in a clinical appropriate time-frame. Furthermore, in the case of fusion genes, accurate detection, including the fusion partner, may allow personalized and targeted therapies for these patients.

Thus far, preclinical, diagnostic and clinical efforts are only scratching the surface of the vast potential that SVs offer for personalized medicine approaches. Future endeavors aimed to increase our knowledge about fusion gene biology and systematic indexing of their occurrence and impact on disease course and treatment regimens, will broaden the spectrum of fusion events that can be exploited in the patient’s interest. Furthermore, the routine implementation of SV-based MRD biomarkers across cancer types will allow intervention therapies matched to actual clinical responses and positively affect the effect/side-effect ratio per patient. Ultimately, integrated pipelines to simultaneously detect SNVs and SVs in an unbiased and clinical relevant time-frame will offer a comprehensive molecular dashboard to diagnose and treat each patient according to the best standards.

In conclusion, the work in this thesis has shown that SVs may be harmful events which can drive tumorigenesis and induce resistance to targeted therapies. However, we also succeeded in taking these tumor-intrinsic events and turned them into tools to improve and provide personalized diagnostics for cancer patients.