MSACL 2016 EU Abstract

Translation of a Predictive Protein Response Signature to Lumretuzumab from a Fresh Frozen Xenograft Model to Formalin-fixed, Paraffin-embedded Human Tumors

Axel Ducret (Presenter)
Pharmaceutical Sciences

Bio: Axel Ducret is a protein biochemist by training who has been using mass spectrometry for the characterization and quantification of proteins in complex mixtures. Throughout his career first at Merck Frosst in Montreal (Canada), then at F. Hoffmann-La Roche in Basel (Switzerland), he has applied increasingly more complex proteomics strategies for the discovery and validation of biomarkers in tissue, biological fluids, or other complex biological systems. One of his current field of interest has been the accurate and specific quantification of proteins from formalin-fixed, paraffin-embedded tissue as an untapped sample collection for retrospective biomarker discovery and validation, and more generally the use of targeted mass spectrometric strategies as a bridging technology between discovery and early development in clinical samples.

Authorship: Axel Ducret(1), Ian James(2), Martina Feilke(3), Sabine Bader(3), Thomas Friess(2), Birgit Bossenmaier (2), Marlene Thomas (2), Maurizio Ceppi(2)
Roche Pharma Research and Early Development (pRED), (1) Pharmaceutical Sciences, Roche Innovation Center Basel, (2) Oncology Disease Therapeutic Area, Roche Innovation Center Münich, (3) Pharmaceutical Sciences, Roche Innovation Center Münich

Short Abstract

The overexpression of the receptor tyrosine-protein kinase erbB-3, also known as HER3, is typically correlated with poor prognosis. In a discovery proteomics approach, we identified a set of 5 proteins that predicted response/non response to the HER3 targeting monoclonal antibody (mAb) Lumretuzumab in preclinical xenograft models. We describe here the development of a targeted mass spectrometry assay in FFPE human tumors and the validation of the signature in samples collected from a breast cancer clinical trial. We believe that this approach provides a valid strategy to develop reliable, cost-effective clinical assays especially if validated antibodies should not be available.

Long Abstract

Introduction: The overexpression of the receptor tyrosine-protein kinase erbB-3, also known as HER3 (human epidermal growth factor receptor 3), is typically correlated with poor prognosis as HER3-driven tumors cells are believed to have escaped from the typical HER1/2-Tyrosine Kinase inhibition model. Using a discovery proteomics approach, we identified a set of 5 proteins that predicted response/non response to the HER3 targeting monoclonal antibody (mAb) Lumretuzumab in preclinical xenograft models. The abundance of these 5 proteins, DPYSL2 (Dihydropyrimidinase-related protein 2), OAT, (Ornithine aminotransferase, mitochondrial), CLIC3 (Chloride intracellular channel protein 3), GM2A (Ganglioside GM2 activator), and PADI3 (Peptidylarginine deiminase III), was verified using targeted mass spectrometry. The protein signature predicted response with a positive predictive value of 91% and sensitivity of 100%. However, for practical use in a clinical setting, the protein signature required translation from fresh-frozen (FF) tumor tissue to formalin-fixed, paraffin-embedded (FFPE) tumor tissue. In this study, we describe the development of a targeted mass spectrometry assay in FFPE human tumors, the translation of the response prediction signature from FF to FFPE tumors, and the validation of the signature in samples collected from a breast cancer (BC) clinical trial.

Methods: Peptide candidates were identified by analyzing cell lines fixed with formalin and commercially available FFPE breast and lung tumor tissue. Peptides were extracted from formalin-fixed material using the Expression Pathology’s Liquid Tissue kit. A multiplexed selected reaction monitoring (SRM) assay, based on isotope dilution with internal reference standards, was developed for four to five best performing peptides of each of the target protein. Each target peptide was further evaluated in terms of linearity, reproducibility and lower limit of quantification. Translation of the 5-proteins signature from fresh frozen to FFPE material was achieved by re-validating the response prediction in a similar set of FFPE xenograft tissue. The abundance levels of the 5 proteins were measured from tumor biopsies obtained from patients with HR+ HER2-low (defined as HER2 IHC 1+ or IHC 2+) metastatic BC participating in the Lumretuzumab combination trial BP28752, after obtaining informed consent. 10 µm tissue slices were sectioned from each bloc and the tumoral cells were excised by laser microdissection. Response prediction status was correlated against the patient RECIST score after treatment.

Results: As FF- and FFPE-extracted peptides often differ in yield, selection of the best candidate peptides for the SRM assay was performed in various formalin-fixed cell lines, after which the multiplexed SRM assay was tested against in primary human lung and breast cancer digests. Interestingly, OAT, DPYSL2, and GM2A were quantifiable both in lung and breast tumors while CLIC3 was significantly less abundant and PADI3 was not detected in breast, indicating that xenografts and primary tumors were not equivalent with regards to protein distribution.

The response prediction algorithm was re-validated using a similar set of xenograft tissue samples (from responder and non-responder cell lines) that was used in the discovery/targeted proteomics approach using fresh frozen material, except that tumors were immediately formalin-fixed after isolation. Overall, the differential protein abundance levels between responder and non-responder cell lines were maintained both in direction and magnitude. However sensitivity was reduced to 80% at a specificity of 84%.

Finally, tumor material from 17 patients was collected and submitted for MS analysis to corroborate whether the xenograft-derived response prediction algorithm could be validated using clinical samples from patients with known clinical responses. As previously observed, all patient samples had quantifiable amounts of the OAT, DPYSL2, and GM2A proteins while CLIC3 and PADI3 had a considerable number of missing values (below limit of detection, 47 and 81%, respectively). Overall, using the RECIST score as a treatment response, the 5 protein biomarkers signature showed no clear separation of responders from non-responders. In particular, the absolute abundance levels of OAT and DPYSL2 were inverted compared to the original response prediction model, indicating that the response prediction signature to Lumretuzumab is not likely to hold in primary human tissue, even though only a small number of samples were analyzed.

Conclusion: We show here a process for the rapid method development and validation of a multiplexed SRM assay illustrated by the measurement of a 5-protein response prediction signature in human clinical samples. We demonstrate the ability to translate a research prototype assay into the clinics with the precision and the accuracy given by the use of internal standards and target mass spectrometric measurements. In particular, the concordant measurement of several peptides for each protein of interest was critical to demonstrate the robustness of the assay and provide an objective measure for the signature’s performance. We believe that this approach provides a valid strategy to develop rapidly and reliably cost-effective clinical assays especially if validated antibodies should not be available or need to be developed.


References & Acknowledgements:


Financial Disclosure

DescriptionY/NSource
Grantsno
SalaryyesF. Hoffmann-La Roche Ltd
Board Memberno
Stockno
Expensesno

IP Royalty: no

Planning to mention or discuss specific products or technology of the company(ies) listed above:

no