NRG1 fusions, a (PDX) model challenge

With the advent of the ‘precision oncology’ movement, the success of biomarker-driven targeted therapies has fundamentally changed our approach to drug development and treatment selection. The results are clearly reflected throughout the infrastructure of our industry, from the rise of the clinical diagnostics to the fundamental role that patient selection now plays in clinical trial design.

The precision oncology movement is currently undergoing a further evolution. As access to and adoption of cancer genomic testing grows, population-level analysis of the accumulating data provides us with an increasingly sophisticated understanding of the biomarker landscape. We now understand that the different kinds of genomic alterations also fall on a spectrum of oncogenic potential. Notably, we can better characterize biomarkers as either driver alterations (which give cancer cells a growth advantage) or passenger alterations (which don’t directly confer a growth advantage).

Powered by this refined understanding and the steady identification of novel, druggable driver alterations, our industry is actively exploring a paradigm shift towards a “driver-alteration first” model. The recent tumor-agnostic approvals of NTRK inhibitors suggest when a true driver alteration can be identified, targeting it may be an effective therapeutic approach regardless of tumor type. The key lies in the unique biology defined by a driver alteration, which may define distinctive (though often rare) patient populations outside of the traditional silos of organ system of origin, histology, and stage.

Elevation Oncology is constantly re-thinking the traditional tools and processes to meet the new challenge of developing drugs from a “driver-alteration first” perspective. It is our hope to share our learnings and insights along the way, as fully realizing the potential of this approach will require a community-wide effort. We start here with sharing the challenges we faced in preclinical investigation and the opportunities for innovation we hope to continue to explore.

A PDX case study: NRG1 fusions

At Elevation Oncology, our first target is the NRG1 fusion, a rare genomic alteration now understood to drive tumorigenesis and survival through activation of HER3 and downstream pathways. Like NTRK fusions, NRG1 fusions are complex structural variants with high oncogenic potential. Importantly, NRG1 fusions rarely occur alongside other known driver alterations.

Today, preclinical efficacy studies using patient-derived xenograft (PDX) models are prioritized due to the ability of PDX models to retain features of the tumor microenvironment and better represent the complexities of a human tumor. However, identifying PDX models that captured the unique biology of a tumor driven by an NRG1 fusion posed several challenges:

  1. NRG1 fusions are rare, present in an estimated 0.2% of all solid tumors. As a result, they have not been well characterized in existing PDX models and it was not efficient for us to systematically test existing models in the hope of finding one that was NRG1 fusion positive.
  2. To create a new PDX model, you first need to identify a patient with the relevant tumor type and obtain fresh tumor tissue. This poses a particular development bottleneck when studying rare genomic drivers such as NRG1 fusions.
  3. NRG1 fusions have been identified in over 10 solid tumor types, making it a strong candidate for a tumor-agnostic development approach. However, to support this hypothesis, a single PDX model is insufficient. Instead, it was important for us to develop a library of models representing different tumor types.
  4. By nature, gene fusions are heterogeneous with regard to fusion partner and breakpoint. As we learn more about the clinical relevance of these variations, additional sub-models for each driver alteration may be needed.

Our first step was to search published literature where we learned about two available models in differing tumor types available through major contract research organizations (CROs). On further inquiry with these companies and other common vendors, we only managed to identify one additional model in a third tumor type. Based on our experience, it appears that systematic establishment of new models characterized by emerging rare drivers like NRG1 fusions is understandably limited and there are few, if any, additional models currently available.

Following discussion with our scientific advisors, we were made aware of primary derived PDX models and cell lines being developed at Memorial Sloan Kettering Cancer Center (MSKCC). As a leading academic research institution, MSKCC is constantly and proactively looking for tumor tissue from patients to generate new NRG1 fusion cell lines, organoids and PDX models.  Our research collaboration has so far generated initial data in a novel SLC3A2-NRG1 fusion lung PDX model and a CLU-NRG1 fusion ovarian PDX model obtained through WuXi, which were recently reported by Dr. Odintsov at ENA 2020 [1].  

Even with a research partnership in place, generating new PDX models is still extremely challenging. The biggest hurdle remains the ability to first locate fresh tumor tissue from NRG1 fusion positive patients. Then, once the tumor tissue is obtained, it typically takes 6-12 months to generate the model. However, by drawing on our collaborations with diagnostic providers for nationwide patient identification and our strong relationships with clinical investigators, we hope to accelerate future NRG1 fusion model development. We have multiple novel models in development and look forward to sharing additional data in 2021 as well as building additional partnerships to further advance our research.

Challenges and opportunities for PDX innovation

Our experience highlights two key areas of challenge and opportunity in evolving preclinical tools to accurately reflect the unique biological impact of rare genomic driver alterations.

Challenge #1: Patient identification.

Druggable driver alterations of interest can be very rare, accounting for less than 2% of all tumors, meaning relevant patients are uncommon. This is problematic as we need a comprehensive set of patients accounting for various tumor types of origin, genomic alteration variants, and treatment histories. A significant patient identification effort is required for each new alteration of interest.


A collaborative, industry-wide approach is needed to systematically build the libraries of models that are needed to advance development for cancers driven by rare genomic alterations. Only through the intentional integration of drug developers, clinical research institutions, diagnostic providers, and academic researchers are we likely to efficiently identify patients and develop models at scale. Our experience has found these integrated relationships to be mutually beneficial in advancing individual research goals as well as refining both therapeutic and diagnostic product development. With everyone working together, we can develop the models in real time to investigate pertinent clinical questions around novel driver alterations like NRG1 fusions.

It is worth noting, however, that even with this close integration, there will likely be times where identifying a patient with exactly the right set of tumor characteristics is not feasible. Therefore, alternatives to true primary derived PDX models may still provide important surrogates in order to create specific models of interest. Technologies, such as CRISPR, may be leveraged to introduce specific genomic alterations into patient-derived cancer cells, but it is not yet clear whether this could provide an equivalent to a true PDX model from a patient.

Challenge #2: Parallel investigation.

We are just starting to learn how to best translate the wealth of insights from genomic testing into development of actionable therapeutics. In particular, there are clearly limitations to our current working terminology of “driver vs. passenger” alterations, “tumor-agnostic” alterations, and “mutual exclusivity”, which are in constant refinement with each new patient case report. As the inevitable “exceptions to the rule” appear, the ability to characterize these nuances and their likely clinical impact will be an important area of research.


While PDX models remain critical tools for predicting likelihood of clinical efficacy of a therapeutic, moving forward they can also serve as critical tools for helping us refine our understanding of the biology of driver alterations.

We initially sought new NRG1 fusion PDX models based on our need to accurately reflect the driver alteration hypotheses, where a single genomic alteration drives the growth of a tumor. The byproduct is that we are now working with partners to establish model libraries that are, in a sense, very “well controlled”. The models should harbor no other known driver alterations while still retaining the complexities that are characteristic of a naturally occurring tumor, differentiating them from models which have been engineered to express only one genomic alteration in a controlled environment.

If we are able to develop collaborative relationships and networks for the efficient generation of PDX model libraries that are genomically well characterized by driver alterations, we anticipate that much more sophisticated and clinically relevant experiments become possible:

  • For example, 1:1 PDX clinical outcomes comparisons become more feasible than ever, and results may become more meaningful in a driver alteration-selected population.
  • In cases where multiple driver alterations are identified in a single patient, PDX models will be invaluable tools for characterizing the relative importance of each driver alteration. In addition, these models can help us investigate the potential for personalized combination therapy development for patients whose tumors harbor more than one driver alteration.
  • Finally, developing serial models over time from a single patient can help us understand if tumors are characterized by different drivers at tumorigenesis versus at metastasis or the time of emerging resistance to an ongoing therapy. Results from these experiments have the potential to help us better understand mechanisms of resistance as well as inform more sophisticated guidance for repeat genomic testing.


A 2017 study by Izumchenko et al. showed a significant association between drug responses in patients and their corresponding PDXs in 87% (112/129) of the studied therapeutic outcomes [2]. While this strong 1:1 correlation highlights the value of PDX models as a tool, there is still too much variability from patient to patient for individual results to reliably translate into predictions for broader patient outcomes. Indeed, only 5% of the anti-cancer drugs that have anti-cancer activity in preclinical studies are approved for clinical application by the United States Food and Drug Administration (FDA) [3].

We believe that our evolving understanding of the natural history of tumors, and what is truly driving the tumorigenesis and growth of a cancer, can help us as an industry close the gap on this discrepancy.

Creating libraries of PDX models that isolate individual oncogenic driver alterations is necessary to refine our ability to develop precision therapies and answer emerging questions around the biology of driver alterations. The challenges are complex, but the opportunity is great. More than ever, it is imperative that we work together as an industry to evolve our tools and processes and empower more precise drug development. Ultimately, this is our best chance of making genomic testing actionable for as many patients as possible to bring about meaningful results for cancer patients.

This article was originally published on LinkedIn. Follow us on our journey to #ElevatePrecisionMedicine.


[1] Odintsov I et al., The anti-HER3 monoclonal antibody seribantumab effectively inhibits growth of patient-derived and isogenic cell line and xenograft models with NRG1 rearrangements. EORTC-NCI-AACR 2020, Presentation# PD-031.

[2] Izumchenko et al., Patient-derived xenografts effectively capture responses to oncology therapy in a heterogeneous cohort of patients with solid tumors. Ann Oncol. 2017 Oct; 28(10): 2595–2605. <>

[3] Hutchinson L and Kirk R., High drug attrition rates–where are we going wrong?. Nat Rev Clin Oncol. 2011 Mar 30;8(4):189-90. <>