Title: A Data Independent Acquisition Workflow Enables the Profiling of Thousands of Human Cancer Tissues for Precision Oncology
Assigned Poster Number: P04.07
Authors: Jakob Vowinckel1; Karel Novy1; Thomas Corwin2; Tobias Treiber1; Roland Bruderer1; Lukas Reiter1; Eike-Christin von Leitner2; Oliver Rinner1; Claudia Escher1 (1Biognosys AG, Schlieren, Switzerland; 2Indivumed GmbH, Hamburg, Germany)
Abstract: Precision oncology requires a deep understanding of the molecular mechanisms involved in cancer biology. The combined analysis of several types of omics data will generate knowledge that goes beyond purely genetic approaches on which precision medicine has relied on almost exclusively in the past. With the rise of data-independent acquisition (DIA) mass spectrometry and advances in chromatography, proteomics technology has come into close proximity to the scale and depth of next-generation sequencing required for multi-omics efforts.
Here we present a platform for the acquisition of true large-scale proteomics and phospho-proteomics datasets for a unique collection of biospecimens derived from the IndivuType cohort of Indivumed, Germany. Matching fresh-frozen tumor and adjacent normal tissue samples from thousands of patients were obtained from Indivumed’s global network of partner hospitals. These samples are of outstanding high quality as they are collected following strictly defined standard operating procedures minimizing variation from pre-analytical variables.
Sample processing from 5 mg of tissue was performed on 96-well plates with the help of a liquid handling platform. Phospho-peptide enrichment was carried out using a Kingfisher Flex device and MagReSyn Ti-IMAC magnetic beads. DIA LC-MS/MS was performed on multiple platforms each consisting of a Thermo Scientific Q Exactive HF-X mass spectrometer coupled to a Waters M-Class LC. Chromatography was operating at 5 µL/min, and separation achieved using 45 min (total proteome) and 60 min (phospho-proteome) gradients. Several thousand samples were analyzed using the platform to date. Quality of raw data is continuously validated using directDIA searching and runs passing QC thresholds are analyzed using Spectronaut software with comprehensive tissue-specific spectral libraries. The resulting high-quality proteomics data containing more than 9’000 proteins (6’000 proteins per run) and 20’000 phospho-peptides (14’000 phospho-peptides per run) is integrated into Indivumed’s IndivuType multi-omics database, supporting identification and validation of new molecular cancer drug targets and biomarkers.
Title: A Novel Way for the Data Analysis of Limited Proteolysis Coupled to Mass Spectrometry – Based on Machine Learning
Assigned Poster Number: P07.01
Authors: Roland Bruderer1; Nigel Beaton1; Ilaria Piazza1; Paula Picotti2; Lukas Reiter1 (1Biognosys AG, Schlieren, Switzerland; 2ETH, Zurich, Switzerland)
Abstract: High attrition rates in target-centric drug development approaches, as well as a limited number of targets, have shifted the focus of drug development back towards phenotypic screening. In parallel, novel proteomics-based target deconvolution approaches to drug target identification have gained popularity. Limited proteolysis coupled with mass spectrometry (LiP-MS) is a new target deconvolution technique that exploits protein structural alterations, as well as steric effects driven by drug binding.
Here, we present a data analysis strategy for LiP-MS experiments based on multiple features of the data and machine learning. This algorithm was designed for the analysis of LiP-MS discovery experiments in an unbiased way. We demonstrate this on controlled experiments with drugs and their known targets.
We developed an algorithm based on LDA using R. The features using for machine learning include dose-response correlation testing, t-testing, multiple peptide score and more. This algorithm enables the ranking of the peptides and proteins in the LiP-MS experiment with a combined LiP score. We performed dose-response experiments in HeLa using the drugs Calyculin A, Rapamycin, FK506, Selumetinib, Fostriecin, and Staurosporine. Using this data, we trained the machine learning framework using a subset of the experiments. Then we applied it to all controlled data sets. This resulted in candidate lists ranked by the LiP score. In the analyzed experiment, the known drug targets are consistently found in the top ten candidates. We demonstrate that the LiP score ranking outperforms individual scores. 3D visualization of the obtained LiP responding peptides clearly shows their proximity to drug binding sites.
Title: High Quantitative Accuracy and Sensitivity in PRM with Multiplexing and Asynchronous Fill-Time Correction
Assigned Poster Number: P25.07
Authors: Sebastian Müller; Tejas Gandhi; Lukas Reiter (Biognosys AG, Schlieren, Switzerland)
Abstract: Parallel reaction monitoring (PRM) is used to monitor peptide quantities robustly across large sample sets. The gold standard for quantification employs stable-isotope standard (SIS) peptides as references to identify endogenous peptides and directly derived absolute quantities. A drawback of this methodology is the doubling of peptides that need to be acquired per analyte, i.e. heavy and light.
One solution to address this is to multiplex the SIS and endogenous peptide acquisition into a single scan. The fill times can be the same length for heavy and light peptides, isochronous, or dependent on their abundance, non-isochronous. The latter requires post-acquisition correction of the raw intensities to obtain accurate quantification, as the instrument itself cannot distinguish which fragments are derived from the SIS or the endogenous peptide.
A plasma sample was spiked with PQ500 reference peptide mix, containing 804 SIS peptides representative of 581 proteins and data acquired on a Q Exactive HF-X, with standard or heavy-light-multiplexed PRM, with isochronous or non-isochronous filling. Scheduled method set-up as well as targeted data analysis, including automatic scan-wise non-isochronous fill-time correction, was performed with SpectroDive 9.
We show that non-multiplexed acquisition leads to higher endogenous peptide identifications (619 representatives of 419 plasma proteins) than standard multiplexing with isochronous filling (523 IDs; 359 proteins), while the latter shows higher quantitative accuracy with median quantitative CVs of 3.7 % compared to 10.4 % in standard PRM. When employing non-isochronous filling and correcting the intensities, the quantitative accuracy is kept (median CV of 3.2 %) and the sensitivity of the standard method nearly reached (589 IDs; 383 proteins).
We present a plasma proteomics workflow for targeted proteomics on a discovery scale, using PQ500 reference peptides and SpectroDive 9. This methodology will be specifically useful to novel targeted proteomics workflows, like MaxQuant.Live, employing optimized fill times for individual analyte groups.
Title: Quantifying More than 500 Human Plasma Proteins with SureQuant and PQ500 in a Single Run
Assigned Poster Number: P25.08
Authors: Tejas Gandhi1; Sebastian Müller1; Jan Muntel1; Sebastien Gallien2; Shannon Eliuk3; Oliver M. Bernhardt1; Lukas Reiter1 (1Biognosys, Schlieren, Switzerland; 2Thermo Fisher Scientific, Paris, France; 3Thermo Fisher Scientific, San Jose, CA)
Plasma proteome is one of the most challenging biological matrices because of its large dynamic range. A classical untargeted proteomics method suffers from this due to the oversampling of abundant proteins. Targeted workflows have a higher dynamic range but are difficult to set up and are not as comprehensive as discovery methods.
Here, we evaluate the use of a plasma protein assay panel containing SIS peptides measured with a new targeted acquisition method. This method uses the spiked-in SIS peptides to trigger the acquisition of the endogenous peptides. This greatly simplifies the classical targeted acquisition workflows by not needing to schedule peptide acquisition by their retention times. However, analyzing data acquired from such a pipeline poses new challenges. Towards this end, we adapted our analysis pipeline in SpectroDive software to optimally analyze SureQuant data.
A pooled non-depleted plasma sample was spiked with the PQ500 kit (Biognosys) containing 804 SIS peptides for 582 proteins. The sample was acquired in technical triplicates with a 50-minute gradient using a Thermo Scientific Exploris 480 in SureQuant acquisition mode. The data was analyzed with an adapted SpectroDive software with 1% peptide FDR. As a next step, we acquired plasma samples from a cohort of cancer and healthy patients using this newly established pipeline.
We could obtain absolute quantities for 740 endogenous peptides corresponding to 529 proteins in all 3 replicates with a median CV of 6.5%. Mapping the identified proteins to proteins with a known concentration in plasma [Anderson, 2002] shows we measured 8 orders of magnitude in dynamic range of plasma protein concentration.
This workflow is more comprehensive than a typical discovery workflow while providing the precision of a targeted workflow. The use of AAA quantified peptides allows us to estimate absolute protein quantities and comparison of quantities between laboratories and LC-MS setups.