Targeted MSxPRM proteome profiling performance in hybrid-PRM/DIA

Hybrid-PRM/DIA was co-developed with Thermo Fisher Scientific to enhance the sensitivity of detection of selected predetermined peptides, while simultaneously producing exploratory-type data. The capability to obtain both types of data within a single measurement is especially crucial for clinical samples, where sample quantities are often limited. In the developed hybrid-PRM/DIA acquisition scheme, multiplexed MSxPRM MS/MS scans are triggered and integrated with Data Independent Acquisition (DIA) data (Fig. 1). The triggering of MSxPRM scans relies on identifying isotope-labeled heavy reference peptides. Successful identification of these labeled peptides enables the accurate quantification of their corresponding endogenous counterparts utilizing narrower isolation windows and optimized ion injection times for the different species. The hybrid-PRM/DIA acquisition strategy generates a comprehensive data matrix of DIA data points complemented by targeted peptide monitoring. This data matrix is particularly useful for proteotyping clinical biospecimens because it does not have any measurement gaps related to clinically relevant pre-selected markers. This is important for their evaluation in molecular tumor boards.

Fig. 1

The hybrid-PRM/DIA acquisition scheme. Compared to conventional Parallel Reaction Monitoring (PRM) and Data-Independent Acquisition (DIA) methods, the hybrid-PRM/DIA approach utilizes rapid and simultaneous multiplexing of PRM MS/MS scans (MSxPRM) triggered by the detection of isotope-labeled heavy reference peptides. Successful detection of the isotope-labeled peptide triggers the high-quality measurement of the corresponding endogenous counter-peptide multiplexed with the isotope-labeled peptide by MSxPRM MS/MS scans. These scans are acquired with a narrower isolation window and maximized ion injection time for each species resulting in a higher sensitivity. Hybrid-PRM/DIA provides a data matrix with no missing data points of clinically relevant markers (heatmap gray: missing value; heatmap white: below LOD)

We initially evaluated the capacity of the hybrid-PRM/DIA technology using a set of 252 proteotypic peptides for a tumor-associated antigen (TAA) panel obtained from 65 annotated human proteins (Additional file 2: Table S1). These TAA peptide mixtures were available in both, their light and isotopically labeled heavy versions. We used the isotopically labeled peptides for triggering the MSxPRM scans in hybrid-PRM/DIA and the light peptides to simulate the endogenous peptide abundance of a potential sample. The heavy internal standard peptides were kept constant in all samples, maintaining a level of approximately 100 femtomoles (injected on the column). In contrast, the light counterparts underwent a dilution series ranging from roughly 100 femtomoles to 10 attomoles (Fig. 2A), reflecting the intensity range of endogenous peptides in different samples. After benchmarking the performance of our peptide mixes using MS/MS, we targeted a subset of high-quality 185 peptides associated with 64 proteins using MSxPRM, while simultaneously recording their DIA traces in hybrid-PRM/DIA. For DIA quantification we used our recently published MS1-based HRMS1-DIA approach [2]. On a high-resolution Orbitrap instrument, this approach results in an increased number of recorded MS1 scans to maximize peptide detection efficiency for more accurate quantification (three MS1 scans per cycle).

Fig. 2

MSxPRM performance of hybrid-PRM/DIA on 185 TAA tumor-associated antigen peptides of 64 proteins. A Dilution series of the light TAA panel (approx. 0.01–100 fmol) in the heavy TAA reference (approx. 100 fmol, constant), which was used to trigger MSxPRM, and a HeLa digest as background matrix. B Number of identified protein groups in hybrid-PRM/DIA MSxPRM mode, DIA, and PRM for the different dilutions. Samples were measured as triplicates. C Hybrid-PRM/DIA MSxPRM and DIA measurements of the tyrosine-protein kinase Lck peptide DFDQNQGEVVK for 0.01 and 0.1 fmol. Shown are the intensities of the heavy and light transition peaks over three technical replicates

When comparing the identifications of light TAA peptides in hybrid-PRM/DIA MSxPRM mode to those of DIA and MSxPRM alone (multiplexed measurements of light and isotopically labeled counter-peptides) we observed that MSxPRM in hybrid-PRM/DIA generally yields superior signal-to-noise ratio and a lower limit of detection as compared to DIA alone (Fig. 2B, C with HeLa lysate background matrix, Additional file 1: Figs. S1A, S1B without HeLa lysate background matrix). This is particularly true within the lower abundance mass range of approximately 0.1 fmol and below. In the abundance range of about 10 attomoles, hybrid-PRM/DIA technology surpasses standard DIA by identifying an average of 52 protein groups versus 34 in DIA (Fig. 2B). Furthermore, the intelligent acquisition offers almost the same protein identification coverage and data completeness as the scheduled stand-alone PRM with an average of 55 protein groups (Fig. 2B). The performance of hybrid-PRM/DIA is further illustrated in Fig. 2C and Additional file 1: Fig. S1B, which show the detectability of the tyrosine-protein kinase Lck (LCK) peptide DFDQNQGEVVK for 0.01 and 0.1 fmol (Fig. 2B) and that of the melanoma-associated antigen 3 (MAGEA3) peptide ISGGPHISYPPLHEWVLR for 0.1 and 1 fmol (Additional file 1: Fig. S1B), comparing hybrid-PRM/DIA with DIA. The narrow isolation window and maximized ion injection time of the light peptides in the MSxPRM scan of hybrid-PRM/DIA helped improve the selectivity and sensitivity of quantification, as well as the detection reliability, specifically when the background noise was high.

We also evaluated the hybrid-PRM/DIA globally for reproducibility. We determined the inter-injection median CVs over the three technical replicates measured at different concentrations. We found that the MSxPRM of the hybrid-PRM/DIA outperformed DIA in terms of CVs smaller than 20% only in the low abundance range (Additional file 1: Figs. S1C-F, S2A). However, the reproducibility of the measurements in scheduled stand-alone PRM and also in DIA on more abundant targets was better in terms of inter-injection median CVs. This is partly due to the thresholding of the MSxPRM event trigger, which sometimes led to a delay in signal acquisition and, thus, truncation of peaks for quantification. In addition, the number of MSxPRM scan events triggered in parallel also contributed to the rather high CV values we observed globally, caused by a considerable amount of time spent on the fast MSxPRM validation/triggering as well as the actual quantification MSxPRM scans (see also Fig. 3).

Fig. 3

Benchmarking of DIA performance in hybrid-PRM/DIA with an increasing number of triggered MSxPRM scans. A Protein groups identified through DIA in hybrid-PRM/DIA with a predefined list of  TAA target peptides (179, 120, and 60 peptides) compared to standard DIA. The HeLa digest was spiked with a mixture of crude TAA heavy triggering peptides and TAA light peptides, both at an amount of  100 fmol per injection. The lower panel of Fig. A shows the median CV values for all proteins quantified in DIA and hybrid-PRM/DIA. B Identified peptides, identified peptides with CVs below 20% or 10% respectively are shown for the different numbers of hybrid-PRM/DIA peptide targets in MSxPRM mode

Global DIA proteome profiling performance of hybrid-PRM/DIA

The global profiling performance of DIA in hybrid-PRM/DIA MS was investigated and compared with the standard stand-alone, high-resolution MS1-based DIA-MS method described above [2]. The analysis was conducted on a HeLa cell lysate digest spiked with the mixture of crude TAA panel light and heavy peptides at an approximate concentration of 100 fmol each. We evaluated the total protein group identifications for 0.5 μg of injected HeLa lysate by DIA (hybrid-PRM/DIA and stand-alone DIA) while simultaneously triggering scheduled MSxPRM scans of 60, 120, and 179 target peptides of the TAA panel in hybrid-PRM/DIA (Additional file 2: Tables S6-S8). Approximately 4600-4800 protein groups were identified at 1% FDR in a 120-min gradient on an ES903 50 cm C18 column in all DIA and hybrid-PRM/DIA conditions, regardless of the number of triggering events (Fig. 3A). This is in the same range as for the TAA dilution experiment on the μPAC™ Neo HPLC column (Additional file 1: Fig. S2B). Even when the number of target peptides was increased to 179, i.e. up to 36 parallel MSxPRM events (Additional file 1: Fig. S3), DIA data acquisition in hybrid-PRM/DIA demonstrated consistent and competitive proteome profiling capabilities. In addition to the similar number of identified protein groups, DIA in hybrid-PRM/DIA showed good quantification precision of proteins and protein groups with median CVs between 10% and 16% (Fig. 3A, bottom panel). In terms of target peptide identification, MSxPRM in hybrid-PRM/DIA was able to identify 57, 108, and 167 peptides, respectively, with a 1% FDR in SpectroDive (Fig. 3B). It can be seen that the coefficient of variation for peptide quantification increased with the number of targeted peptides: while 58% of the 60 peptides monitored had CVs ≤ 20%, this number decreased to only 18% when 179 peptides were monitored. This can be explained by the increasing number of parallel scheduling events and the increased time needed for the fast MSxPRM validation/triggering, as well as the actual quantification MSxPRM scans.

Evaluation of hybrid-PRM/DIA on clinical samples using a melanoma diagnostic marker panel

Hybrid-PRM/DIA was applied to a set of de-identified tumor specimens from melanoma patients to obtain molecularly actionable data from limited biological samples while digitizing the proteotype for future research studies. Initially, the global proteotype of the cohort of 95 samples was determined by standalone DIA. Subsequently, the Hybrid-PRM/DIA technique was used on a subset of 30 of these samples to monitor clinically relevant level-1 and level-2 marker proteins more reliably. Level-1 proteins serve as diagnostic markers with immediate clinical relevance [22]. Level-2 protein information provides additional details on potential drug targets or pathway nodes. Furthermore, DIA-MS complemented the data matrix with level-3 information on the global proteotype. This enabled a comprehensive characterization of the clinical phenotype and may lead to the discovery of novel biomarkers in the future.

Our comprehensive level-1 and level-2 melanoma marker list consisted of 65 protein groups relevant to melanoma disease diagnosis and treatment decision-making (Additional file 2: Table S10) [23]. Of these 65 protein groups, 43 were detectable in our melanoma patient cohort using standard DIA (Additional file 1: Fig. S4, Additional file 2: Table S11) [2]. The extracted level-1 and level-2 marker proteins exhibited varying degrees of missing data, with an average of 34% missingness across the entire patient cohort and extracted protein data matrix. The highest degrees of missingness were observed for the cellular tumor antigen p53 (TP53) and the G1/S-specific cyclin-D3 (CCND3) (Additional file 1: Fig. S4). We then synthesized 30 AQUA peptides to specifically monitor 28 proteins from our panel in MSxPRM using hybrid-PRM/DIA (Additional file 2: Table S3). The list of AQUA peptides included peptides for the melanocyte protein PMEL as a diagnostic marker for melanocytic tumors [22] and the drug target cyclin-dependent kinases 4 CDK4 [24]. Melanoma is an attractive target for CDK4/6 inhibitors because the p16INK4a/Cyclin D1-CDK4/6/RB pathway is dysregulated in the majority of melanomas [25]. Also, the neurofibromin protein NF1 as a tumor suppressor was monitored. Targeting NF1-regulated pathways such as RAS/MAPK or PI3K/mTOR offers potential therapeutic options for patients with melanoma [26, 27]. Our panel also included a peptide of the tyrosine-protein kinase receptor UFO, which is encoded by the AXL gene. AXL expression has been suggested to be associated with epithelial-mesenchymal transition (EMT) in melanoma, which contributes to both metastatic spreading and therapy resistance in cancer [28]. In addition, receptor tyrosine kinase inhibition has recently been shown to improve BRAF-targeted therapy [29,30,31]. The degree of missingness for UFO in our standard DIA protein data matrix is quite high (80%), so targeting UFO is an ideal test case scenario to detect peptide amounts close to the limit of detection.

We benchmarked the performance of hybrid-RPM/DIA versus DIA by measuring technical replicates on the subset of 30 patient samples where enough material was available. The 30 AQUA peptides were used as triggers for MSxPRM. DIA in hybrid-PRM/DIA led to an overall identification of close to 6500 protein groups, which is in the same range as for DIA alone (Fig. 4A). The measured samples clustered nicely by technical replicate and acquisition scheme (Fig. 4B). When we extracted the peptide traces for the 30 AQUA peptides, we found that the data completeness was 70.1% for the monitored peptides in DIA, compared to 84.4% in hybrid-PRM/DIA (Fig. 5A). In hybrid-PRM/DIA, the spike-in reference allowed us to clearly assign the 15.6% of missing values as below the limit of detection, excluding technical artifact as a reason for data missingness. The narrow isolation window and maximized ion injection time of the hybrid-PRM/DIA MSxPRM scans improved the selectivity and sensitivity of quantification, as shown for CDK4 in patient sample M150506 and for UFO in patient sample M090924 (Fig. 5B). The MS2 traces of the NF1 peptide LFDLVDGFAESTK in M180213 showed a more reproducible extraction in hybrid-PRM/DIA than in DIA (Fig. 5B). Specifically, we observed that in hybrid-PRM/DIA detection and quantification reliability increases in areas with a high background. Hybrid-PRM/DIA also allows for determining the lower limit of detection of endogenous peptides based on the spiked heavy-labeled reference. For instance, in the case of PMEL, we clearly showed that the endogenous peptide was not detectable in patient sample M040418 because it was below the detection limit (< 0.0101 fmol/μl) as estimated from the heavy reference calibration curve (Additional file 1: Fig. S5). This helps to avoid biases in downstream data processing, such as those introduced by imputation algorithms, which could lead to misinterpretation of the results. In conclusion, we have demonstrated that the new hybrid-PRM/DIA strategy has the potential to monitor clinical marker peptides with greater reliability, sensitivity and specificity than DIA alone, and after formal assay development and qualification, even allows absolute quantification of the monitored endogenous peptides. Additionally, the global proteotype can be monitored equally well in hybrid-RPM/DIA as in DIA.

Fig. 4

Identified protein groups by DIA or by DIA in hybrid-PRM/DIA in thirty melanoma patient samples. A The graph displays the number of identified protein groups per measurement for thirty melanoma patient samples. The samples were measured as technical replicates using DIA and hybrid-PRM/DIA. In B, the protein group measurements of DIA and hybrid-PRM/DIA are presented in a heatmap, clustered by hierarchical clustering (blue: DIA, red: hybrid-PRM/DIA)

Fig. 5

Monitoring of 27 melanoma-associated protein groups in thirty patient samples by DIA or Hybrid-PRM/DIA. A Heatmap of 27 protein groups that were detected and quantified by DIA or by MSxPRM of hybrid-PRM/DIA in 30 melanoma patient samples. Heavy reference peptides were used for triggering and peak integration. All samples were measured in duplicate for each acquisition scheme. Missing data points in DIA are shown in gray, values below the LOD in MSxPRM are shown in white. The four proteins CDK4, PMEL, NF1, and UFO shown in panel B are highlighted in bold. B Four examples illustrating the benefits of MSxPRM measurements in hybrid-PRM/DIA. MSxPRM helped to increase the specificity, reproducibility, and sensitivity of peptide detection. Peptide levels below the limit of detection (LOD) could be unambiguously assigned (target: light endogenous peptide, ref: heavy reference peptide)