 Hello everyone. My name is Sandra Götzer from the Clinical Prototype Analysis Center at ETH Zirig. And today I would like to provide you with our perspective on how we think mass spectrometry can generally be applied to accelerate the development and translation of biomarkers into clinical essays. Molecular biomarkers derived from, for instance, genetic variants, transcripts, proteins, and or metabolites play key roles in, for instance, clinical oncology. Their molecular signatures can help predict the likelihood of cancer development or progression and they have the potential to detect the disease at an early stage. They can also support treatment decision-making and predict treatment responsiveness. On the level of the protein, numerous research grade biomarker signatures have been known for years, however, as striking discrepancy exists between the efforts to develop new protein biomarkers and the numbers that is eventually clinically approved. One major reason for this discrepancy is that testing of protein biomarker candidates in a large-scale validation study poses an enormous organizational and financial challenge that hampers translation into a clinical grade essay. To overcome limitations in translational biomarker research, we think that various technologies should be combined in a time and cost-efficient manner. On the one hand, there are ELISAs, which are broadly established and routinely used for clinical protein quantification. ELISAs are robust and accurate and can be produced at low cost. Also, automation provides high throughput. However, an ELISAs success relies on the availability of highly specific and sensitive monoclonal antibodies and the multiplexing capability of ELISAs is somewhat limited. This makes the technology unsuitable for large-scale biomarker screening. Modern mass spectrometry-based proteomics, on the other hand, enables multiple protein marker measurements in parallel with adequate sensitivity and without the need for highly specific antibodies. However, mass spectrometry is costly, often semi-quantitative, time-consuming and technically challenging for routine diagnostic point-of-care applications. Here, we would like to suggest a mass spectrometry-guided immuno-SA development strategy, short MS-GUIDE, which combines the strength of MS-analyzed throughput with measurement throughput of the ELISAs, so kind of combining the best of the two worlds. MS-GUIDE consists of a two-step strategy wherein a first step A, pre-qualified literature-harvested potential biomarker candidates, can be evaluated for which no high-quality antibodies exist. We call this first step the verification phase. Targeted mass spectrometry using parallel reaction monitoring is used to measure potential protein markers in a highly multiplexed fashion in a medium-sized training cohort. ELISAs can then be developed in a second step of phase B for a small subset of pre-qualified biomarker candidates, which have shown clinical potential and then used for significance testing and validation in a larger patient cohort. So again, it is the combination of the best of the two worlds, going from analyte throughput to sample throughput, which is in the end required for the eventual SA approval. We employed and benchmarked our MS-GUIDE strategy on a prostate cancer patient cohort. The goal there was to identify a prognostic protein stratification panel for men with localized prostate cancer for a better risk assessment of disease aggressiveness. Clinical routine and prostate cancer diagnosis comprises the determination of prostate specific antigen levels, short PSA in blood. If the PSA level is elevated, a tissue biopsy is taken to further confirm the cancer diagnosis. The tissue biopsy gleason score is then used to classify the grade of cancer aggressiveness. Clinically localized prostate cancer can be controlled by curative radical prostatectomy. Still around 40% of all surgically treated men will experience a detectable serum PSA increase as a clear indicator of cancer progression or metastasis. So there is the need to identify novel markers that are specifically indicating the presence of a biologically aggressive prostate cancer to improve treatment outcome. Blood or serum is an attractive source of biomarker discovery as it is easily accessible and the prostate cancer secreted proteins have been prioritized as biomarker candidates because even localized prostate cancer is genetically highly variable and biopsy-based sampling is usually random, making accurate stratification of disease based on tissue biopsies alone quite difficult. So our goal here was to develop a serum risk stratification test to improve prediction outcome. As first part of step A of MS-GUIDE, we pre-selected potential biomarker proteins related to cancer aggressiveness. For that we mainly focused on secreted serum proteins which we had previously identified using a P10 knockout mouse model. P10 inhibits the PI3K act pathway and loss of this tumor suppressor is one of the best characterized genomic events in prostate cancer. The loss of P10 is strongly associated with unfavorable oncologic outcomes making P10 and its downstream targets potentially useful markers for distinguishing indolent from aggressive disease. Also we already knew that human orthologs from secreted proteins affected by the P10 knockout and mice are equally detectable in serum from prostate cancer patients. In total we pre-qualified 48 biomarker candidates by parallel reaction monitoring mass spectrometry together with four unrelated controls. Their potential role in prostate cancer development and progression is illustrated based on hallmarks of cancer as can be seen on the right side of this slide. As next step of the verification phase A, ZARA from 78 individuals collected before undergoing prostatectomy were measured using mass spectrometry and the respective proteins were then quantified. Proteins identified in all samples with no more than one third missing values were then selected as candidates for ELISA development. This resulted in a list of 21 reproducibly detectable potential biomarker candidates indicated in green in the violin plot. The list of proteins was subjected to machine learning using a random forest classifier algorithm for dedicated feature selection by predicting risk groups. Random forest classification is particularly well suited in this context as this classification approach does not assume that the data are linearly separable. For further details on the classification approach please go ahead and check out our manuscript. The models derived by random forest classification repeatedly identified fibronectin and vitronectin to be among the top drivers of survival prognosis prediction. We therefore selected these two proteins for our platform change and switched to part B of MS Guide, the validation phase. For that we developed sandwich ELISAS and the corresponding monoclonal antibodies were generated by immunization of mice with a native protein. Our newly established ELISAS were tested and validated on an independent patient cohort. There we could show that the combination of the two protein signatures together with PSA alone or with PSA plus biopsyglesin score significantly outperforms the state of the art measures for prostate cancer aggressiveness. The combination of all parameters yielded an area under the curve of 0.66. This means that our derived protein signature together with PSA stratifies patients with localized prostate cancer 10% better than current gold standards of PCA diagnostics. So in summary we could show that the mass spectrometry driven two-step screening approach can be successfully applied to develop and translate new biomarker panels. Using MS Guide we derived a new prognostic protein signature for localized prostate cancer. This signature may be helpful in the future in the preoperative setting to stratify between men with indolent and lethal disease and lead to the identification of men with lethal disease who eventually require more intense treatment. The MS Guide strategy is generally applicable also in the context of other disease settings. In the last decade it became apparent that in the context of precision diagnostics there is the need for better and more reliable biomarkers to provide actionable information to guide treatment decisions. Each patient's individual molecular makeup should be the basis for guiding such medical decisions. The development of patient tailored treatment regimes requires the unbiased broad digitization of clinical specimen using various technologies. This is what we are currently pursuing further within SMOC, the Swiss Multiaumic Centre. So in case you are interested in what we are doing and offering there please check out our webpage. With that I am at the end of my short in silico talk and of course I would like to acknowledge the people that contributed to this work specifically Peter Schüffler who is responsible for the modeling part and also proteomics and Ralf Schiers as our industrial implementation partner as well as the CTI for the majority of our funding. Thank you for your interest and for listening in.