Verification of cerebrospinal fluid (CSF) biomarkers for multiple sclerosis and other neurological diseases is a major challenge due to a large number of candidates, limited sample material availability, disease and biological heterogeneity, and the lack of standardized assays. Furthermore, verification studies are often based on a low number of proteins from a single discovery experiment in medium-sized cohorts, where antibodies and surrogate peptides may differ, thus only providing an indication of proteins affected by the disease and not revealing the bigger picture or concluding on the validity of the markers. We here present a standard approach for locating promising biomarker candidates based on existing knowledge, resulting in high-quality assays covering the main biological processes affected by multiple sclerosis for comparable measurements over time.
Biomarker candidates were located in CSF-PR (proteomics.uib.no/csf-pr), and further filtered based on estimated concentration in CSF and biological function. Peptide surrogates for internal standards were selected according to relevant criteria, parallel reaction monitoring (PRM) assays created, and extensive assay quality testing performed, i.e. intra- and inter-day variation, trypsin digestion status over time, and whether the peptides were able to separate multiple sclerosis patients and controls.
Assays were developed for 25 proteins, represented by 72 peptides selected according to relevant guidelines and available literature and tested for assay peptide suitability. Stability testing revealed 64 peptides with low intra- and inter-day variations, with 44 also being stably digested after 16 h of trypsin digestion, and 37 furthermore showing a significant difference between multiple sclerosis and controls, thereby confirming literature findings. Calibration curves and the linear area of measurement have, so far, been determined for 17 of these peptides.
We present 37 high-quality PRM assays across 21 CSF-proteins found to be affected by multiple sclerosis, along with a recommended workflow for future development of new assays. The assays can directly be used by others, thus enabling better comparison between studies. Finally, the assays can robustly and stably monitor biological processes in multiple sclerosis patients over time, thus potentially aiding in diagnosis and prognosis, and ultimately in treatment decisions.
There are currently only a few biomarkers for multiple sclerosis (MS) in clinical use, including MRI (T2-weighted lesions), oligoclonal bands and IgG ratio, JC viral antibody titers and neurofilament light, as summarized in a recent review . However, both in-house discovery studies and available literature suggest numerous additional biomarkers representing several of the processes and pathways active in MS, such as inflammation and neurodegeneration [2,3,4,5]. Such findings however require further verification via robust targeted assays, e.g. through parallel reaction monitoring (PRM) using high-quality stable isotope labelled heavy peptides [6,7,8].
The process of developing robust targeted assays in turn requires the consideration of a multitude of factors in order to ensure the quality and relevance of the assays, especially when the goal is to provide absolute protein measurements that would allow the consecutive analyses of proteins both across different labs and over time. This will make it possible to monitor specific pathological processes occurring in individual MS patients and thereby gain a deeper insight into the processes active at the various stages of the disease, which in turn would be a valuable tool in diagnosis, prognosis and treatment decisions.
Cerebrospinal fluid (CSF) is a commonly used body fluid in studies of neurological diseases, such as MS. Although not as easily accessible as serum/plasma, it is likely to better reflect ongoing neurological processes as it is in direct contact with the central nervous system . However, large scale biomarker verification of discovery experiments has proven difficult in CSF and results are rarely consistent between studies. Likely reasons for this are methodological differences, large individual variations in total CSF protein concentrations  as well as significant heterogeneity in neurological diseases [11,12,13]. As a consequence, the quantitative data from individual biomarker discovery and verification studies do not always overlap and cannot directly and easily be compared and combined . The large dynamic range of proteins in CSF also leads to challenges when measuring small disease-related changes in low abundant proteins [13, 15], especially vulnerable to small methodological variations and inaccuracies. Combined with relatively low patient numbers in most studies, it becomes almost impossible to conclude regarding a biomarker’s potential, and thus move from the biomarker discovery phase to clinically useful biomarkers. It is therefore crucial to create robust targeted assays for accurate measurements of biomarker candidates.
Here we describe a suggested standard approach for the selection of candidate biomarkers in CSF for MS, and detail the required validation of PRM assays for absolute quantification of 25 proposed biomarker candidates. The validation includes (i) intra- and inter-day variation, (ii) the effect of trypsin digestion time, and (iii) verification of the separation capability between MS and controls observed from the literature . Additionally, the linearity around the typical concentration of target peptides was determined and corresponding response curves determined. The validated assays are ready to be used in large-scale analysis of patient samples and the presented standard approach for PRM assay development can also be applied for other neurological diseases.
Results—from biomarker candidates to high-quality PRM assays
The following sections describes the steps from potential literature-based biomarker candidates from CSF-PR (proteomics.uib.no/csf-pr), to the list of the most promising proteins and peptides to include in robust high-quality PRM biomarker assays for MS (Fig. 1).
Selection of proteins and peptides
Literature curation using CSF-PR
The recently published CSF-PR 2.0  contains structured and searchable quantitative data for thousands of CSF-proteins from close to 100 datasets related to MS, Alzheimer’s disease (AD) and Parkinson’s disease (PD). The data in CSF-PR comes from mass spectrometry studies that have passed certain filters notably related to (i) methodology (bottom-up shotgun or targeted proteomics for main experiment; ELISA for verification), (ii) number of patients (n ≥ 5 in each disease group; if using pools, ≥ 3 pools for each disease group and n ≥ 20 total), and (iii) sample type (CSF from living subjects).
The biomarker selection was conducted by merging relevant datasets from MS and control subcategories in CSF-PR and extracting the proteins found to be significantly different in abundance in the majority of studies, according to certain criteria. See “Materials and methods” section for more details. The CSF-PR data extraction resulted in an initial list of 133 proteins (Additional file 1: Table S1), representing promising biomarker candidates for MS quantified in several experiments where various degree of fractionation had been used. Separately, we also collected a list of proteins that were changed between MS and control, but quantified in only one study in CSF-PR (Additional file 2: Table S2). Most of the latter were proteins from our recent discovery study , where both depletion and extensive peptide fractionation was performed, and they are therefore likely the lowest abundant proteins possible to quantify by current mass spectrometry proteomics technology.
Identifying proteins within the suitable dynamic range of CSF
An important condition for assay development is to be able to perform PRM quantitation in crude CSF without high-abundant protein depletion or peptide fractionation. In order to find the proteins most likely fulfilling this condition, a CSF fractionation test was carried out, where the trypsin digested CSF-proteome was separated into eleven fractions and analysed by data-dependent acquisition (DDA) MS/MS, resulting in 1194 proteins. We estimate that PRM can be ten times more sensitive than what can be identified in a regular DDA shotgun experiment . Therefore, the identification of a protein in a 20 µg un-depleted sample, fractionated into 11 fractions (first fraction usually does not contain peptides), indicates that the protein can be quantified by PRM in crude CSF. A total of 120 of the 132 proteins extracted from CSF-PR were identified in the DDA analysis (Additional file 3: Table S3) and passed on to the next steps in the assay development. All of the proteins identified in the DDA analysis can be found in Additional file 4: Table S4.
Biological processes and categories
A closer inspection of the 120 proteins revealed several groups of related proteins with similar names, functions and abundance relationship between MS and control, e.g. immunoglobulin proteins, cadherin proteins, receptor-type tyrosine-protein phosphatases, and SLIT and NTRK-like proteins. These proteins are likely to be involved in the same biological processes, and therefore developing individual PRM assays for all these proteins is probably not necessary, as recent studies indicate that such related proteins are most-often affected in the same manner [4, 5].
A representative selection of the 120 proteins from Additional file 3: Table S3 was made based on available information from CSF-PR, i.e. the number and proportion of studies where changes between MS and Non-MS was observed. Additionally, the large network of interacting and significantly changed proteins between MS and other neurological diseases (OND) generated in our recent publication  was utilized to select one or two proteins as representatives of the various biological processes likely to be affected in MS. These processes include e.g. (i) inflammation—a hallmark of the MS disease, (ii) extracellular matrix organization proteins—providing structure and support for developing neurons (e.g. collagens and proteoglycans), (iii) ephrin proteins—involved in neuron development, myelination and axonal guidance, and (iv) cadherins—cell adhesion proteins known to be involved in de- and re-myelination. Additional proteins found especially interesting based on keywords in UniProt  or our own previous experiments were also included. All the steps in the protein selection process is outlined in Fig. 2, and the 25 selected proteins are shown in Table 1.
Selecting peptides to represent the proteins under investigation, so-called surrogate or signature peptides, is a crucial step in the development of a targeted proteomics assay, and a number of criteria determines if a peptide is suitable [19, 20]. Here, the initial peptide selection was done mainly based on peptide data available from CSF-PR combined with general guidelines for selecting peptides for targeted proteomics [6, 19,20,21]. As a rule, three- to four peptides were selected per protein and the corresponding isotopically heavy labelled versions were ordered. However, not all peptides were found with an acceptable signal in the MS/MS analysis, hence, some of the proteins are only represented by a single peptide. For one protein (chitinase-3-like protein 1), more than three heavy peptides were ordered given that this protein has been reported as particularly interesting in relation to MS [22,23,24,25,26,27,28], and we had previously experienced that this protein could be challenging to quantify (data not shown). In total, 72 peptides were selected to represent the 25 proteins (Table 2). Further testing was performed to determine whether they were truly suitable as protein surrogates, as outlined below.
Peptide stability testing
PRM assays ought to have low intra- and inter-day variation in order to allow comparable quantitative measurements over time. To test this, PRM experiments with two replicates were processed each day across 5 days, and the intra- and inter-day coefficient of variation (CV) was calculated for all the 72 peptides. Most of the peptides displayed an intra- and inter-day CV of less than 20% (Fig. 3). Only eight of the initial 72 peptides had a CV above 20%, with seven from chitinase-3-like protein 1 (CH3L1), failing only the inter-day CV, and one from Seizure 6-like protein 2 (SEZ6L2), failing both intra- and inter-day CV. Notably, none of the peptides from CH3L1 showed an acceptable inter-day CV. Peptides not fulfilling the intra- and inter-day CV limits were discarded, resulting in 64 peptides from 24 proteins retained. Detailed results from this experiment can be found in Additional file 5: Table S5.
Peptide digestion testing
In order to create assays for absolute protein concentrations in CSF samples, it is important to investigate how the trypsin incubation time affects the quantitative results. The main question is whether the detected amount of a peptide continues to increase after completing a standard trypsin digestion protocol with 16 h digestion time (see “Materials and methods” section), as then the absolute concentration of the corresponding protein cannot be determined via such a standard digestion protocol using the given peptide.
The experiment investigated five different digestion times (1, 5, 16, 24 and 30 h), each with three replicates, and was repeated three times. A peptide was considered stably digested after 16 h if the percentage change from 16 to 24 h and from 16 to 30 h was both less than 20%. In addition, the resulting peaks had to be satisfactory with regards to intensity, interference and shape, evaluated through the Skyline  data analysis. A total of 44 peptides, with at least one peptide from each of the 24 proteins, passed the digestion test. In other words, 20 peptides, but no proteins, were discarded based on this test.
How the peptides changed (ratio light/heavy) after 16 h is illustrated in Fig. 4a, b, where red dots represent the peptides who failed the test. Examples of observed peptide profiles for two selected proteins are shown in Fig. 5a, b.
As can be seen from Fig. 5a, all of the three peptides from neuronal cell adhesion molecule show limited increase after 16 h of digestion, i.e. they all passed the test. However, some proteins demonstrated an increase in peptide amount after 16 h and/or a varying digestion profile for the different peptides. As an example, we see that peptide SLPSEASEQYLTK in Fig. 5a appears to be readily digested already after 1 h. Other proteins had some peptides passing and some failing the digestion test. In Fig. 5b, we see that one peptide for the protein Seizure-6-like protein 2 increase in amount up to 16 h of digestion, and then no increase beyond 20% is found (peptide passed), while a different peptide from the same protein continue to increase after 16 h, thereby failing the test, notably with a high variation in the minimum and maximum values.
All the data from this experiment is available in Additional file 6: Table S6 and the complete digestion profile for all peptides can be found in Additional file 7: Fig. S1 and peptide abundance change at all measured time points are in Additional file 8: Fig. S2.
RRMS vs non-inflammatory controls
To test whether the changes indicated in CSF-PR between MS and Non-MS could be reproduced, a small PRM study was conducted using six pools of CSF from MS (three pools of seven RRMS patients) and control (three pools of seven OND patients) patients. These pools have previously been analysed in-depth by shotgun TMT-based proteomics , and were selected to test how well the optimized PRM assays reflect the differences between the two patient categories.
The majority of the 44 tested peptides showed similar regulation trends as previously reported in the literature (Fig. 6), but a couple of the peptides were not found significantly changed in this study (Osteomodulin—two peptides, Complement C1q subcomponent subunit C—one peptide, Collagen alpha-3(VI) chain—one peptide, CD44 antigen—one peptide), and two peptides showed the opposite direction of regulation compared to CSF-PR (Complement C1q subcomponent subunit C—one peptide, Complement C1r subcomponent—one peptide). These were therefore discarded from further assay development. In conclusion, seven of 44 tested peptides failed this test.
We concluded that the remaining 37 peptides were suitable to reflect the previously reported regulations, and therefore represent the most promising biomarkers for MS. The complete results from this test is found in Additional file 9: Table S7.
Final assay peptides
After all the steps detailed above, we finally arrived at a list of 37 peptides from 21 proteins for which promising absolute quantitative PRM assays could be developed (Table 3). These represent the best surrogates to precisely and reproducibly quantitate proteins affected by MS. A complete table of all the tested peptides, important results and data from each experiment is collected in Additional file 10: Table S8, and a protein level overview of the number of peptides passing the various tests can be found in Additional file 11: Table S9.
To ensure that the peptides could be accurately quantified by PRM mass spectrometry around the level of its observed concentration in CSF, calibration curves determining the linear areas of quantification have to be determined. Calibration curves have so far been generated for 17 of the peptides passing the initial testing and the work is ongoing. Rat plasma was used as matrix with varying amounts of synthetic light peptide and stable spike-in of heavy surrogate peptide. The linear area was determined by weighed least squares regression. For further details see “Materials and methods” section. An example of a calibration curve is shown in Fig. 7 and all of the calibration curves developed so far are provided in Additional file 12: Fig. S3 and additional details related to slope, intercept and linear areas are in Additional file 10: Table S8.
The verification of biomarker candidates in CSF from discovery studies has been challenging due to the many issues pointed out in the introduction. One of the major bottlenecks has been to simultaneously measure a larger number of proteins in a high number of samples in a reproducible fashion and over time. In our view there is a need to develop well-described high-quality assays able to generate reproducible data over time, and ideally also between laboratories, in order to achieve efficient biomarker verification in CSF.
Recently, there has been at least two publications going in the direction of generating high-quality PRM assays for CSF-proteins; one describing assays for 30 brain proteins , and another assays for monitoring a set of defined biological process . In our study, we have contributed towards this idea by developing 37 well-described high-quality PRM peptide assays representing 21 proteins found to be affected in multiple sclerosis across multiple studies. Our goal is that these assays can be used to generate comparable data over time and provide the possibility to analyse and compare the protein levels in a large number of patient samples in a long-term perspective.
An important aspect of this work is that the biomarker candidates have been selected based on data from several studies using the online database CSF-PR. This approach is likely to provide less false positive candidates as more data, most often from several research groups, is used in the selection. Furthermore, the surrogate peptides have been selected based on quantitative information available from CSF-PR (when available), indicating that the particular peptide is actually regulated in the target protein. In sum, we argue that our approach is a step forward in increasing the effectiveness of verifying biomarker candidates in CSF.
Selection of proteins and peptides—what is important to consider?
Using CSF-PR as a starting point for selecting proteins affected by MS differs substantially from using a single experiment as the basis for selection, and the 133 proteins initially identified thus represents the proteins collectively reported to be affected by MS from the mass spectrometry proteomics literature. In our view, this approach increase the chance of including the most relevant proteins and those more likely affected by MS, compared to basing the selection on a single study.
Next, we wanted to make sure that all of the proteins included in the assay development was possible to quantify in a PRM experiment without the need for protein depletion or peptide fractionation, as these steps both have their drawbacks. Targeted immunoaffinity depletion of high-abundant proteins is a useful way to increase the chance of measuring low-abundant proteins. Depletion is however a debated approach in biomarker studies, given that proteins not targeted by the depletion column may be co-depleted due to unspecific binding and protein interactions, potentially introducing a bias already at an early stage in the sample preparation [32,33,34]. As for peptide fractionation, this step will increase the analysis time, cost and complexity, and is therefore not desirable.
To arrive at a more manageable number, and as a demonstration of our suggested workflow, we selected 25 proteins. These proteins will of course not reflect all off the disease-affected processes represented by the complete list of 120 proteins, nor are they meant to represent a final list of biomarker candidates for multiple sclerosis. However, they do cover a range of functions and processes relevant in the MS disease as summarized in Table 1.
Creating PRM assays for all peptides from all potentially interesting CSF-proteins would be preferable, but as there will always be a cost vs. benefit consideration this is not realistic
Using the peptide level quantitative information available in CSF-PR as part of the surrogate peptide selection was also considered important. For the disease-affected proteins we observed that not all of the peptides were regulated in the discovery data, and that some peptides were also regulated in the opposite direction (Fig. 8). We suggest inspecting and using the peptide level data available in CSF-PR to select peptides that are observed as significantly changed, thus increasing the chance of the peptide actually representing the regulation reported at the protein level.
Different peptides from the same protein may show different (relative) abundance due to: (i) peptides mapping to multiple proteins or proteoforms with different regulation status, (ii) some peptides can be less suitable for mass spectrometry or in too low amounts for stable and accurate quantitation, and (iii) certain peptides have varying degrees of post-translational modifications, resulting in unstable concentration for the unmodified form. In addition, it is important to consider the general guidelines for peptide selection in targeted proteomics, i.e. to avoid non-unique peptides and peptides prone to mis-cleavages.
Peptide stability testing—most peptides are stable across runs
A large majority of the tested peptides passed our stability test, indicating that they are suitable for PRM monitoring robustly over time. We can conclude that they are in the appropriate concentration range in CSF for the method to consistently provide a sufficient signal for stable measurements.
Chitinase-3-like protein 1 peptides give unstable measurements over time
Chitinase-3-like protein 1 (CH3L1) has been linked to several neurological diseases [23, 26, 35,36,37,38,39], including MS [22, 24, 25, 27, 28]. However, it seems that this protein is not ideal for an absolute targeted assay, due to the unstable peptide measurements across runs. One explanation is that CH3L1 is low-abundant in many patient categories used in testing and as controls, but more abundant in MS patients. This may explain why several discovery studies have found it regulated [5, 24, 25, 28] compared to controls and why we find high variations in this stability test using CSF from Control (OND) patients. Some peptides for CH3L1 were not far from being acceptable in terms of variation, having inter-day CV values between 20 and 30%. Due to the potential importance of this protein one could consider including PRM assays for these peptides, but then taking into account that the variation in the data is larger.
Peptide digestion testing—most peptides show no increase after 16 h
Considering our goal of creating PRM assays able to measure absolute protein amounts, we found that it was crucial to examine the digestion status after the standard 16 h of trypsin incubation. This is not a standard test for PRM assays, but in order for an assay to come as close as possible to reflect the absolute protein amount, we found it essential. For most proteins, digestion times of 16 h were sufficient, in that no significant increase (or decrease) in peptide amount was observed after prolonged incubation. But there were a couple of peptides increasing also after 16 h of digestion, and nine proteins having some peptides passing and some failing the digestion test (Fig. 4b).
When examining the full profile of peptide amount (L/H) measured after various trypsin incubation times (Additional file 7: Fig. S1), we also note that certain peptides show a decrease already before 16 h of digestion. Although, no decrease beyond 20% was observed before 16 h, this may still indicate that some undesired peptide degradation is occurring already before 16 h. Such “fast digesting” peptides should ideally have shorter trypsin incubation times. For the peptides where the digestion was not completed after 16 h, one could envision longer digestion times in order to reach complete digestion, or consider testing other digestion conditions. However, applying individual trypsin digestion times and conditions for a large number of peptides in assays run over time is tedious and unrealistic. The peptides where the digestion was not complete after 16 h are not suitable for absolute quantification, but the assays could still be used for relative quantitation if trypsin incubation times are equally long between experiments. An alternative would be to search for other peptides better representing these proteins when using 16 h digestion time.
The overall results from the digestion testing show that each peptide can have specific physiochemical properties affecting the digestion process and demonstrate the necessity of obtaining peptide digestion profiles for each individual peptide. Another approach could be to use isotopically labelled proteins as internal standards, instead of peptides, in which case digestion variability would be adjusted for by the internal standard. This is however a quite costly approach.
PRM RRMS vs control—confirmation of previously found changes
This small PRM study was designed to investigate whether the selected peptides could reproduce the previously reported quantitative differences between MS and controls. As we had merged some of the disease sub-categories when performing the initial CSF-PR search, it was not possible to find identical MS and control groups. We concluded that using relapsing–remitting MS (RRMS) and OND controls, was a good choice for this experiment.
Most of the peptides also passed this test, but seven peptides did not show the same significant difference between RRMS and OND as reported between MS and Non-MS in the CSF-PR publications , either because the difference was not significant or they showed the opposite change. One reason may be that the number of patients included in both the PRM study described here and the studies in CSF-PR is not large enough to eliminate the biological variation as a factor, creating false positive biomarker candidates. The patient groups used were also not identical, which could result in variation in differentially abundant proteins.
Another reason for the discrepancy may be that many of the studies in CSF-PR used depletion of high-abundant proteins, which on purpose was not performed in our PRM pilot experiment. Depletion could potentially affect the protein quantitation and thereby the resulting differentially abundant proteins as variation is introduced, as discussed above. This is particularly relevant if the number of patients is low. Even though the seven rejected peptides did not pass this test, they could still prove valuable upon more thorough testing with larger patient numbers. In the current study, the 37 peptides displaying similar differential abundance as previously reported were prioritized.
Development of calibration curves—linearity down to the highest dilution point
Calibration curves were generated in rat plasma as it is a somewhat similar matrix to CSF and from a different species (non-human CSF was not possible to obtain), so that there would be no endogenous presence of analyte signal in the matrix, which would add to the spiked signal . The calibration curves displayed a high degree of linearity down to the highest dilution point, with adjusted R2 values all over 0.99. Ideally, the calibration curves would include endogenous analyte concentrations so that the signal would be indistinguishable from the background, yielding a hockey-stick shape of the curve. The % CV of the replicates of the lowest endogenous peptide concentrations was less than 20% for all but one peptide, indicating that the limit of quantification (LOQ) was not reached for these. As the analyte concentrations rarely varies more than the span covered by the linear curve, the assays were deemed sufficient for our purposes, and the concentration span between the lowest and highest measured endogenous concentration will be used for absolute quantitation.
Relevance for multiple sclerosis and other neurological disorders
The assays have been developed to monitor processes affected by MS, but through CSF-PR, we find that several of the protein candidates are also found changed in Alzheimer’s and Parkinson’s disease. This is the case for nine of the 21 proteins having peptides that passed all test (CD44 , Follistatin-related protein 4 , Secretogranin-2 [31, 42] ProSAAS , Neurexin-1 , Cadherin-13 , Kallikrein-6 , Amyloid-like protein 1  and Ephrin type-A receptor 4 ). It indicates, not surprisingly, that many of the processes affected by MS are also affected by other neurological disorders, and are thus not specific to MS. Which in turn can mean that the diagnostic value of these particular proteins is limited, however they could still be very valuable as biomarkers for disease status, treatment effect and prognosis.
Validation in larger cohorts using the developed assays is necessary to determine the value of the proteins as biomarkers in a clinical setting. After validation of a subset of the proteins using PRM, one likely way to implement the measurement of the proteins into the clinic would be to develop ELISA-assays for the most valuable proteins. In the future the PRM-assays could perhaps also be used directly in the clinic.
It is also expected that these assays will be useful in shedding light on the disease status for other diseases where similar processes are affected. The remaining 12 proteins with peptides that passed all tests are however only changed in the MS categories in CSF-PR. These proteins may therefore be the most useful for diagnostic purposes and monitoring of processes occurring specifically in MS patients. For more details, the proteins can be searched and available data investigated in CSF-PR.
In this study, we have developed 37 robust PRM peptide assays for 21 CSF proteins likely affected by MS. The selected proteins cover many of the pathways and processes recently reported to be affected in MS, but also in other neurological diseases such as Alzheimer’s and Parkinson’s disease. The peptides chosen as protein surrogates are quantifiable without the need for depletion, fractionation or enrichment prior to mass spectrometry. Due to the documented inter- and intra-day stability of the assays and the digestion stability, comparable quantitative values over time is expected. This allows for large-scale analyses of patient samples to reveal the relationship between the monitored MS-affected processes, disease progression and treatment response, and results from future large-scale patient analyses using these assays are expected to aid in treatment decisions.
These well-documented absolute quantitative assays could also be adopted by other laboratories and have the potential to generate comparable quantitative measurements between laboratories. To explore this potential future, inter-laboratory comparisons must be conducted.
We recommend that the presented workflow should be used as a general guideline for the development of targeted PRM biomarker assay in CSF, and consider this work to be a contribution towards standardizing CSF protein quantification allowing us to move from non-comparable data between single experiments to accumulation of reproducible quantitative data over time. In our view this is essential in order to enable the analyses of large enough patient cohorts to reveal disease-related changes in the CSF proteome related to disease status and progression.
Materials and methods
Human cerebrospinal fluid (CSF) was obtained by diagnostic lumbar puncture, according to the standardized protocol for collection and biobanking . Patients gave written consent and the study was approved by the Regional Committee for Medical Research Ethics of Western Norway. Various pools of CSF were used in the experiments described in this paper, mainly due to the limited availability of CSF samples to use for assay optimization and testing. Details about the pools can be found in Table 4 and in . The same pools as in  were used for the MS vs. Control PRM study (MS and OND pools, here: Pools 3–8). The pool used for the DDA proteome depth and peptide stability experiment consisted of various OND patients (Pool 2), the pool used in the digestion test consisted of 3 OIND patients (Pool 1, all with myelitis).
Rat plasma (P2516, Sigma Aldrich) was used to construct calibration curves for high-purity peptides. The purchased rat plasma contained lyophilized material derived from 1 ml pooled and filtrated rat blood with the addition of anticoagulant, 3.8% trisodium citrate. The concentration of the rat plasma was estimated by BCA, and diluted in 1xPBS to a final concentration of 0.5 µg/µl prior to trypsination.
Literature curation using CSF-PR
We used CSF-PR (https://proteomics.uib.no/csf-pr) to extract biomarker candidates between MS and non-inflammatory control patients (Non-MS) as of August 2017. To specifically extract quantitative data relevant to this comparison, we first merged certain MS and control subcategories listed in CSF-PR as follows: RRMS (relapsing-remitting MS), CDMS (clinically definite MS) and CIS-MS (clinically isolated syndrome with conversion to MS) were merged to the general category “MS” and the subcategories OND (other neurologically controls), symptomatic controls, Non-MS, healthy and healthy controls were all merged to the general category “Non-MS”. In this way we identified protein data from all papers in the resource comparing MS to non-inflammatory controls. The protein table with the quantitative data from these studies contained thousands of proteins, so we applied some selection criteria by using the table filtering options in CSF-PR before exporting the protein list: (i) proteins quantified in at least two studies and (ii) having a CSF-PR score () of >=0.5 (50%) or < = − 0.5 (− 50%) according to the equation used in CSF-PR for summarizing overall reported protein regulation (see below, ×100 for %), indicating that each protein was increased or decreased in at least 50% when averaging the results from all studies.
This resulted in 194 proteins, which were exported from CSF-PR, and further analysis was performed using Excel. To identify the most consistently changed proteins, we applied an additional criterion that (iii) proteins were found changed in the same direction (up or down) in at least two studies. This reduced the list to 133 proteins (Additional file 1: Table S1), representing the most promising and consistently reported biomarker candidates for MS. A separate list of proteins that were significantly changed between MS and Non-MS, but quantified in only one study in CSF-PR was also created (Additional file 2: Table S2).
CSF sample preparation—general
Protein concentration in the CSF pools was measured by the QubitTM fluormeter (InvitrogenTM, Thermo Scientific) and the Qubit™ protein assay kit (InvitrogenTM, Thermo Scientific), following the manufacturers protocol. CSF samples were lyophilised at 30 °C in an Acid-Resistant CentriVapTM Concentrator System (LabconcoTM), and dissolved in 20 µl freshly made Urea solution (8 M Urea/20 mM methylamine (Sigma Aldrich)). All CSF samples were in-solution digested as previously described  using trypsin porcine (Promega, art. V5111) added to samples in a 1:50 relationship, and desalted using OASIS ® HLB µElution plates 30 µm (Waters Corp, Millford, MA, USA) according to the manufacturer’s instructions. Samples were vacuum dried following desalting, and dissolved in 2% ACN, 0.1% TFA to a concentration of approximately 0.5 µg/µL for the MS analysis. About 0.5 µg were injected if not otherwise stated.
Preparation and spike-in of synthetic peptides
All isotopic labelled peptides (IS peptides) used as internal standards were purchased from Thermo Scientific at crude (unknown purity) and AQUA Ultimate (> 95% purity) quality for peptide testing and AQUA Ultimate quality only for calibration curves, whilst synthetic light peptides (SpikeTides) were acquired from JPT. Heavy labelled peptides have been C-terminally modified with 13C and 15 N isotope arginine or lysine. The synthetic heavy peptides were added to the samples before the desalting step, and the synthetic light peptides from JPT used to make calibration curves were added prior to digestion as they contain a tag that needs to be enzymatically released. Heavy peptides were spiked to the samples in an approximate 1:1 relationship between the heavy IS peptide and the endogenous analyte, which was estimated from initial peptide tests (data not shown). Notably, the lowest endogenous concentration was adjusted to 3 fmol/µg. Spike-in for calibration curve development is described under “calibration curve” section.
PRM mass spectrometry—general for all PRM experiments
The separation of peptides was performed by an Ultimate™ 3000 RSLCnano System (Thermo Fisher Scientific™) with an Acclaim PepMap™ 100 trap column (diameter width at 75 µm × 2 cm nanoviper C18 column, with particle size 3 µm and length at 100 A) and 5 µL 0.1% TFA solution. Peptides were separated on an analytical column PepMapTM RSLC C18 (diameter width 75 µm × 50 cm, particle size at 2 µm and 100 A in length) with the combination of 95% solvent A (0.1% FA) and 5% solvent B (100% ACN, 0.1% FA) with a flow rate of 200 µl/min. The column gradient for peptide elution went from 0 to 5 min with 5% solvent B, then an increase at 5–5.5 min to 8% of solvent B, 5.5–140 min 35% B, 140–155 min 90% B and 155–170 min 90% B. At 170–175 min solvent B decreased to 5% B and held at 5% solvent B from 175 to 190 min. Column temperature was specified to be 35 °C, whilst the auto sampler had a temperature of 4 °C. Ionization of samples occurred with an Easy-Spray™ (Thermo Scientific) ion source, with a spray voltage at 1.8 kV. The capillary temperature was set at 250 °C, heater temperature at 350 °C, whilst the S-lens RF value were at 60. Sheath and auxiliary gases were not used. As a result of the ion source, samples were obtained in a positive ionization mode.
Mass spectrometry analysis
The method duration was 195 min (runtime 10–175 min). The mass spectrometer was operated in PRM scheduled mode and switched between full scan MS1 between every 12th PRM MS2 scan. The instrument was controlled through Q Exactive HF Tune 2.4 and Xcalibur 3.0. MS1 spectra were acquired in profile mode in the scan range of 375–1500 m/z with resolution of 15,000, automatic gain control (AGC) target of 3e6, and a maximum injection time (IT) of 15 ms. The target peptides on the inclusion list were sequentially isolated for higher-energy collision dissociation (HCD) fragmentation and MS2 acquisition to a normalized HCD collision energy of 28%, target AGC value of 1e5, resolution R = 15 000, and IT of 100 ms. The precursor isolation window was set to 1.6 m/z with no isolation offset or dynamic exclusion. Lock-mass (445.12003 m/z) internal calibration was used.
Skyline  was used for the creation of inclusion lists prior to PRM-mass spectrometry analysis and for data refinement of the PRM-mass spectrometry data. Skyline settings were overall kept at default, or updated depending on the parameters in the mass spectrometry analysis used to acquire data. Notably, structural modifications were specified with carbamidomethyl (C) and isotope modification “label: 13C(6) 15 N(2) (C-term K)” and label: “13C(6) 15 N(4) (C-term R)”. Both 2+ and 3+ charged precursors and b- and y-ions were investigated in the stability test experiment, while in the following (digestion, MS vs. OND and calibration curves), only 2+ precursors and y-ions were used, as these most often gave the best signal. Detailed Skyline settings for each experiment, e.g. the peptide and transition settings and filters, can be inspected in the Skyline documents uploaded to Panorama Public (https://panoramaweb.org/PRM_Assay_CSF.url).
The peak signal for each peptide was determined by the Skyline peak picking algorithm, and manually verified or re-integrated based on the fragment pattern of the peptide, elution profile and simultaneous retention time of the endogenous and the IS peptide. Spectral libraries from CSF samples generated on the same Q Exactive HF instrument were used as a reference to make sure the correct peak for the various peptides were chosen. The three fragments with the highest intensity, low interference, and mass error less than 10 ppm was selected for quantitation. Additional file 13: Fig. S4A and B shows examples of typical transitions used in the assay. All other transitions can be inspected in Panorama Public (https://panoramaweb.org/PRM_Assay_CSF.url) where the Skyline documents from all experiments can be downloaded.
Notably, for most peptides, one to three of the transitions were significantly more intense compared to the rest, only the top three where therefore chosen for quantification. A typical example of this is illustrated in Additional file 13: Fig. S4C and D. For some peptides in certain tests or replicates, only two transitions were used for quantitation, due to missing data or bad peaks in specific replicates. These were mainly from very low abundant proteins and/or from peptides with only low intensity transitions. The area under the curve, excluding background, were summed to give one peak area value for each peptide.
Furthermore, the endogenous peak area was divided by the peak area of the heavy internal standard peptide to generate a ratio to standard which was used for quantitation. From Skyline, a.csv file was exported containing the quantitative data needed for follow-up processing in Microsoft Excel or R. To determine the difference between the two patient groups in the final PRM experiment, an unpaired two tailed, homoscedastic student’s t test was performed using Microsoft Excel. A p-value of ≤ 0.05 was used to determine a significant difference.
CSF protein depth investigation
We tested the identification of CSF proteins from a 20 µg un-depleted CSF sample (pool 2) subjected to peptide fractionation into 11 fractions following trypsin digestion (as described above). Peptide fractionation was performed by mixed mode reversed phase-anion exchange chromatography (MM)  on a Promix MP column (MP10.250.0530, 1.0 × 250 mm, 5 μm, 300 Å, Sielc Technologies), as previously described .
Data dependent acquisition mass spectrometry analysis
Approximately 0.5 μg of peptides from each fraction was injected into the same LC system, trap column and mass spectrometer as described above. However, a 25 cm analytical column (PepMap RSLC, 25 cm × 75 μm i.d. EASY-spray column, packed with 2 μm C18 beads (Thermo Scientific)) was used (flow rate of 0.250 μL/min). Solvent A and B was the same as above as was the other MS general instrumental parameters related to ionization, voltage, temperature etc.
The mass spectrometer was operated in data-dependent acquisition mode to automatically switch between full scan MS1 and MS2 acquisition. The method duration was 120 min (runtime 8-105 min). The instrument was controlled through Q Exactive HF Tune 2.4 and Xcalibur 3.0. MS spectra were acquired in the scan range of 375–1500 m/z with resolution of 60 000, automatic gain control (AGC) target of 3e6, and a maximum injection time (IT) of 25 ms. The 12 most intense eluting peptides above intensity threshold 5e4, and charge states two or higher, were sequentially isolated for higher-energy collision dissociation (HCD) fragmentation and MS2 acquisition to a normalized HCD collision energy of 28%, target AGC value of 1e5, resolution R = 60,000, and IT of 110 ms. The precursor isolation window was set to 1.6 m/z with an isolation offset of 0.3 m/z and a dynamic exclusion of 20 s. Lock-mass (445.12003 m/z) internal calibration was used and isotope exclusion was active.
All raw files were converted to mgf using ProteoWizard  and searched using X! Tandem , MyriMatch  and MS Amanda  via SearchGUI (v2.1.3)  against the homo sapiens complement of the UniProt/SwissProt reviewed database downloaded October 2015 (20 196 entries)  with the reversed version of every sequence added as decoys. The search settings were: carbamidomethylation of C; oxidation of M as variable modification; trypsin as enzyme with a maximum of two missed cleavages; precursor charge 2–5; peptide length 6–30; precursor mass tolerance 10 ppm and fragment mass tolerance 0.005 Dalton. All other settings were left as the defaults. The search engine results were combined and assembled in PeptideShaker  (v1.1.2). Hits were thresholded to retain only the best scoring until a false discovery rate (FDR) of 1% was reached, estimated using the distribution of target and decoy hits .
Peptide stability test
To test the intra- and inter-day variability of measurements for the peptides, we analysed aliquots from the same CSF samples (pool of OND patients) at two different time points at the same day across 5 days. Two 10 µg aliquotes of CSF-pool 2 were in-solution digested, spiked, desalted, dried and stored in − 20 °C. This was repeated on five different days, and all samples were analysed by PRM MS as described above. Data was inspected and refined in Skyline as described above, and in Excel, the intra- and inter-day variation for each peptide (2+ and 3+ separately) was calculated on the exported total area ratio (light/heavy). This value for each peptide was compared between samples prepared on the same day (intra-day) and between each of the three sample sets (inter-day). Peptides with intra- and inter-day CV less than 20% was considered reproducible.
Peptide digestion test
Fifteen 10 µg aliquots of CSF-pool 1 was in-solution digested, spiked, and desalted as described above. Trypsination was however, stopped at five different time points (1, 5, 16, 24 and 30 h), and three replicates was stopped at each time point. This experiment was repeated three times (across three different weeks). Data was refined in Skyline, as described above and the percentage change in ratio to standard was calculated for each peptide between 16 to 24 h and 16 to 30 h. A peptide was considered stably trypsinated after 16 h if the percentage change from 16 to 14 h, and 16 to 30 h was less than 20%. Individual protein plots showing ratios at all time points for all peptides were generated using R (http://www.R-project.org). The plots were generated using the graphics package ggplot2 (https://ggplot2.tidyverse.org) (Additional file 7: Fig. S1).
PRM RRMS vs control
Samples from six CSF-pools were used in this experiment. The samples were crude 100 µg aliquotes from the experiment described in , which was three pools of MS patients and three pools of OND patients (pools 3-8). The samples were purified and concentrated using 3 kDa ultracentrifugation filters as described in , before in-solution digestion and Oasis desalting as described above, except a 10 mg plate was used, as in , due to the high protein amount (100 µg). The eluate after the desalting was divided to 5 µg aliquots and two aliquots (replicates) from each pool was used for this experiment. The aliquots were spiked with heavy peptides, dried and dissolved in 2% ACN, 0.1% TFA to approximately 0.5 µg/µL. Approximately 1 µg sample was injected for MS analysis and analysed by PRM as described, except MS2 resolution was 30,000. The ratio to standard was used to calculate fold change and significance between groups.
Development of calibration curves
Calibration curve generation in rat plasma
The calibration curves were made in rat plasma by preparing a dilution series of synthetic normal mass (light) peptides based on the estimated endogenous concentration of each peptide in CSF Pool 2. An 11-point dilution curve was centered around this estimate, so that the varying analyte spanned 32 times the endogenous concentration and 32 times less than the estimated endogenous concentration. The dilution series was prepared for the synthetic light peptides for analysis in rat plasma (dilutions prepared in 8 M Urea/20 mM methylamine directly prior to trypsination), Additionally, a mix of heavy AQUA peptides in levels 1:1 with endogenous peptide was generated in 5% ACN. Eleven 10 µg aliquots of rat plasma were added synthetic light peptides in different dilutions. In addition, one sample without added light peptides and was used as a blank. The twelve samples were in-solution trypsin digested, and equal amounts of the AQUA heavy peptides mix were added to each sample prior to desalting. This procedure was performed in trypsinated triplicates.
Mass spectrometry analysis
For the calibration curve experiment, some optimized parameters were used in the PRM analysis. Peptides were separated on an analytical column PepMapTM RSLC C18 (diameter width 75 μm × 25 cm, particle size at 2 μm and 100 A in length) with the combination of 95% solvent A (0.1% FA) and 5% solvent B (100% ACN, 0.1% FA) with a flow rate of 250 μl/min. The column gradient for peptide elution went from 0 to 5 min with 5% solvent B, then an increase at 5–5.5 min to 7% of solvent B, 5.5–65 min 22% B, 65–87 min 35% B, 87–92 min 90% B and 92–102 min with 90% B. At 102–105 min solvent B decreased to 5% solvent B and held a 5% solvent B from 105 to 120 min. The method duration for calibration curve runs was 120 min (runtime 10–110 min). The mass spectrometer was operated and MS spectra acquired as described above for PRM analysis, except MS2 spectra were acquired with optimized collision energies, resolution R = 60,000 at 200 m/z, IT of 118 ms, AGC target value at 2e5 and precursor isolation window was set to 0.7 m/z. All other parameters related to the LC and MS instrumentation and settings were as described above for general PRM experiments.
Calibration curve development in R
Following data refinement in Skyline, the ratio to standard values were exported for analysis in the programming language R (http://www.R-project.org). For the peptides measured in rat plasma, the measured ratio to standard was multiplied with the spike-in level to give the measured concentration at each dilution point and was plotted against the theoretical concentration. Notably, as more variation is common in the high concentration measurements, the linear regression was weighted with 1/sd^2 to limit the impact of the points with the highest variability on the regression equation . The slope, intercept and the lowest and highest theoretical concentration points of the linear curve was exported. The plots were generated using the graphics package ggplot2 (https://ggplot2.tidyverse.org).
The mass spectrometry proteomics data from the DDA experiment have been deposited to the ProteomeXchange Consortium via the PRIDE  partner repository with the dataset identifier PXD013449 and https://doi.org/10.6019/pxd013449. All other mass spectrometry data are targeted PRM data, and Skyline documents and raw files are available through Panorama Public  at https://panoramaweb.org/PRM_Assay_CSF.url and at ProteomeXchange with dataset identifier PXD017281.
Availability of data and materials
The .raw files and datasets supporting the conclusions of this article are available in the PRIDE repository (ID: PXD013449 and PXD017281) and in Panorama Public (https://panoramaweb.org/PRM_Assay_CSF.url).
Automatic gain control
Clinically definite multiple sclerosis
Chitinase-2-like protein 2
Clinically isolated syndrome with conversion to multiple sclerosis
CSF Proteome Resource
Coefficient of variation
Data dependent acquisition
False discovery rate
Higher-energy collision induced dissociation
Limit of quantitation
Mixed-mode reversed phase anion exchange
Other inflammatory neurological diseases
Other neurological diseases
Parallel reaction monitoring
Relapsing–remitting multiple sclerosis
Selected reaction monitoring
Paul A, Comabella M, Gandhi R. Biomarkers in multiple sclerosis. Cold Spring Harb Perspect Med. 2019;9(3):a029058.
Kroksveen AC, Aasebø E, Vethe H, Van Pesch V, Franciotta D, Teunissen CE, et al. Discovery and initial verification of differentially abundant proteins between multiple sclerosis patients and controls using iTRAQ and SID-SRM. J Proteomics. 2013;78:312–25.
Kroksveen AC, Jaffe JD, Aasebø E, Barsnes H, Bjørlykke Y, Franciotta D, et al. Quantitative proteomics suggests decrease in the secretogranin-1 cerebrospinal fluid levels during the disease course of multiple sclerosis. Proteomics. 2015;15(19):3361–9.
Opsahl JA, Vaudel M, Guldbrandsen A, Aasebø E, Van Pesch V, Franciotta D, et al. Label free analysis of human cerebrospinal fluid addressing various normalization strategies and revealing protein groups affected by multiple sclerosis. Proteomics. 2016;16(7):1154–65.
Kroksveen AC, Guldbrandsen A, Vaudel M, Lereim RR, Barsnes H, Myhr KM, et al. In-depth cerebrospinal fluid quantitative proteome and deglycoproteome analysis: presenting a comprehensive picture of pathways and processes affected by multiple sclerosis. J Proteome Res. 2017;16(1):179–94.
Lange V, Picotti P, Domon B, Aebersold R. Selected reaction monitoring for quantitative proteomics: a tutorial. Mol Syst Biol. 2008;4:222.
Bourmaud A, Gallien S, Domon B. Parallel reaction monitoring using quadrupole-Orbitrap mass spectrometer: principle and applications. Proteomics. 2016;16(15–16):2146–59.
Peterson AC, Russell JD, Bailey DJ, Westphall MS, Coon JJ. Parallel reaction monitoring for high resolution and high mass accuracy quantitative, targeted proteomics. Mol Cell Proteomics. 2012;11(11):1475–88.
Bridel C, Eijlers AJC, van Wieringen WN, Koel-Simmelink M, Leurs CE, Schoonheim MM, et al. No plasmatic proteomic signature at clinical disease onset associated with 11 year clinical, cognitive and MRI outcomes in relapsing-remitting multiple sclerosis patients. Front Mol Neurosci. 2018;11:371.
Schilde LM, Kosters S, Steinbach S, Schork K, Eisenacher M, Galozzi S, et al. Protein variability in cerebrospinal fluid and its possible implications for neurological protein biomarker research. PLoS ONE. 2018;13(11):e0206478.
Disanto G, Berlanga AJ, Handel AE, Para AE, Burrell AM, Fries A, et al. Heterogeneity in multiple sclerosis: scratching the surface of a complex disease. Autoimmune Dis. 2010;2011:932351.
Hensiek AE, Sawcer SJ, Compston DA. Searching for needles in haystacks-the genetics of multiple sclerosis and other common neurological diseases. Brain Res Bull. 2003;61(3):229–34.
Zhang J. Proteomics of human cerebrospinal fluid - the good, the bad, and the ugly. Proteomics Clin Appl. 2007;1(8):805–19.
Guldbrandsen A, Farag Y, Kroksveen AC, Oveland E, Lereim RR, Opsahl JA, et al. CSF-PR 2.0: an interactive literature guide to quantitative cerebrospinal fluid mass spectrometry data from neurodegenerative disorders. Mol Cell Proteomics. 2017;16(2):300–9.
Huhmer AF, Biringer RG, Amato H, Fonteh AN, Harrington MG. Protein analysis in human cerebrospinal fluid: physiological aspects, current progress and future challenges. Dis Markers. 2006;22(1–2):3–26.
Bauer M, Ahrne E, Baron AP, Glatter T, Fava LL, Santamaria A, et al. Assessment of current mass spectrometric workflows for the quantification of low abundant proteins and phosphorylation sites. Data Brief. 2015;5:297–304.
Apweiler R, Bairoch A, Wu CH, Barker WC, Boeckmann B, Ferro S, et al. UniProt: the universal protein knowledgebase. Nucleic Acids Res. 2004;32(Database issue):D115–9.
Phillips HL, Williamson JC, van Elburg KA, Snijders AP, Wright PC, Dickman MJ. Shotgun proteome analysis utilising mixed mode (reversed phase-anion exchange chromatography) in conjunction with reversed phase liquid chromatography mass spectrometry analysis. Proteomics. 2010;10(16):2950–60.
Chiva C, Sabido E. Peptide selection for targeted protein quantitation. J Proteome Res. 2017;16(3):1376–80.
Pan S, Aebersold R, Chen R, Rush J, Goodlett DR, McIntosh MW, et al. Mass spectrometry based targeted protein quantification: methods and applications. J Proteome Res. 2009;8(2):787–97.
Rauniyar N. Parallel reaction monitoring: a targeted experiment performed using high resolution and high mass accuracy mass spectrometry. Int J Mol Sci. 2015;16(12):28566–81.
Borras E, Canto E, Choi M, Maria Villar L, Alvarez-Cermeno JC, Chiva C, et al. Protein-based classifier to predict conversion from clinically isolated syndrome to multiple sclerosis. Mol Cell Proteomics. 2016;15(1):318–28.
Coffman FD. Chitinase 3-Like-1 (CHI3L1): a putative disease marker at the interface of proteomics and glycomics. Crit Rev Clin Lab Sci. 2008;45(6):531–62.
Comabella M, Fernández M, Martin R, Rivera-Vallvé S, Borrás E, Chiva C, et al. Cerebrospinal fluid chitinase 3-like 1 levels are associated with conversion to multiple sclerosis. Brain. 2010;133(Pt 4):1082–93.
Hinsinger G, Galeotti N, Nabholz N, Urbach S, Rigau V, Demattei C, et al. Chitinase 3-like proteins as diagnostic and prognostic biomarkers of multiple sclerosis. Mult Scler. 2015;21(10):1251–61.
Johansen JS, Jensen BV, Roslind A, Nielsen D, Price PA. Serum YKL-40, a new prognostic biomarker in cancer patients? Cancer Epidemiol Biomarkers Prev. 2006;15(2):194–202.
Modvig S, Degn M, Roed H, Sorensen TL, Larsson HB, Langkilde AR, et al. Cerebrospinal fluid levels of chitinase 3-like 1 and neurofilament light chain predict multiple sclerosis development and disability after optic neuritis. Mult Scler. 2015;21(14):1761–70.
Stoop MP, Singh V, Stingl C, Martin R, Khademi M, Olsson T, et al. Effects of natalizumab treatment on the cerebrospinal fluid proteome of multiple sclerosis patients. J Proteome Res. 2013;12(3):1101–7.
MacLean B, Tomazela DM, Shulman N, Chambers M, Finney GL, Frewen B, et al. Skyline: an open source document editor for creating and analyzing targeted proteomics experiments. Bioinformatics. 2010;26(7):966–8.
Begcevic I, Brinc D, Dukic L, Simundic AM, Zavoreo I, Basic Kes V, et al. Targeted mass spectrometry-based assays for relative quantification of 30 brain-related proteins and their clinical applications. J Proteome Res. 2018;17(7):2282–92.
Brinkmalm G, Sjodin S, Simonsen AH, Hasselbalch SG, Zetterberg H, Brinkmalm A, et al. A parallel reaction monitoring mass spectrometric method for analysis of potential CSF biomarkers for Alzheimer’s Disease. Proteomics Clin Appl. 2018;12(1):1700131.
Koutroukides TA, Guest PC, Leweke FM, Bailey DM, Rahmoune H, Bahn S, et al. Characterization of the human serum depletome by label-free shotgun proteomics. J Sep Sci. 2011;34(13):1621–6.
Guldbrandsen A, Vethe H, Farag Y, Oveland E, Garberg H, Berle M, et al. In-depth characterization of the cerebrospinal fluid (CSF) proteome displayed through the CSF proteome resource (CSF-PR). Mol Cell Proteomics. 2014;13(11):3152–63.
Ramström M, Hagman C, Mitchell JK, Derrick PJ, Håkansson P, Bergquist J. Depletion of high-abundant proteins in body fluids prior to liquid chromatography fourier transform ion cyclotron resonance mass spectrometry. J Proteome Res. 2005;4(2):410–6.
Houston DR, Recklies AD, Krupa JC, van Aalten DM. Structure and ligand-induced conformational change of the 39-kDa glycoprotein from human articular chondrocytes. J Biol Chem. 2003;278(32):30206–12.
Bereman MS, Beri J, Enders JR, Nash T. Machine learning reveals protein signatures in CSF and plasma fluids of clinical value for ALS. Sci Rep. 2018;8(1):16334.
Wildsmith KR, Schauer SP, Smith AM, Arnott D, Zhu Y, Haznedar J, et al. Identification of longitudinally dynamic biomarkers in Alzheimer’s disease cerebrospinal fluid by targeted proteomics. Mol Neurodegener. 2014;9:22.
Kester MI, Teunissen CE, Sutphen C, Herries EM, Ladenson JH, Xiong C, et al. Cerebrospinal fluid VILIP-1 and YKL-40, candidate biomarkers to diagnose, predict and monitor Alzheimer’s disease in a memory clinic cohort. Alzheimers Res Ther. 2015;7(1):59.
Llorens F, Thune K, Tahir W, Kanata E, Diaz-Lucena D, Xanthopoulos K, et al. YKL-40 in the brain and cerebrospinal fluid of neurodegenerative dementias. Mol Neurodegener. 2017;12(1):83.
Mani DR, Abbatiello SE, Carr SA. Statistical characterization of multiple-reaction monitoring mass spectrometry (MRM-MS) assays for quantitative proteomics. BMC Bioinform. 2012;13(Suppl 16):S9.
Barucker C, Sommer A, Beckmann G, Eravci M, Harmeier A, Schipke CG, et al. Alzheimer amyloid peptide abeta42 regulates gene expression of transcription and growth factors. J Alzheimers Dis. 2015;44(2):613–24.
Spellman DS, Wildsmith KR, Honigberg LA, Tuefferd M, Baker D, Raghavan N, et al. Development and evaluation of a multiplexed mass spectrometry based assay for measuring candidate peptide biomarkers in Alzheimer’s Disease Neuroimaging Initiative (ADNI) CSF. Proteomics Clin Appl. 2015;9(7–8):715–31.
Ebers GC. Randomised double-blind placebo-controlled study of interferon beta-1a in relapsing/remitting multiple sclerosis. PRISMS (Prevention of Relapses and Disability by Interferon beta-1a Subcutaneously in Multiple Sclerosis) Study Group. Lancet. 1998;352(9139):1498–504.
Shi M, Movius J, Dator R, Aro P, Zhao Y, Pan C, et al. Cerebrospinal fluid peptides as potential Parkinson disease biomarkers: a staged pipeline for discovery and validation. Mol Cell Proteomics. 2015;14(3):544–55.
Teunissen CE, Petzold A, Bennett JL, Berven FS, Brundin L, Comabella M, et al. A consensus protocol for the standardization of cerebrospinal fluid collection and biobanking. Neurology. 2009;73(22):1914–22.
Kessner D, Chambers M, Burke R, Agus D, Mallick P. ProteoWizard: open source software for rapid proteomics tools development. Bioinformatics. 2008;24(21):2534–6.
Craig R, Beavis RC. TANDEM: matching proteins with tandem mass spectra. Bioinformatics. 2004;20(9):1466–7.
Tabb DL, Fernando CG, Chambers MC. MyriMatch: highly accurate tandem mass spectral peptide identification by multivariate hypergeometric analysis. J Proteome Res. 2007;6(2):654–61.
Dorfer V, Pichler P, Stranzl T, Stadlmann J, Taus T, Winkler S, et al. MS Amanda, a universal identification algorithm optimized for high accuracy tandem mass spectra. J Proteome Res. 2014;13(8):3679–84.
Barsnes H, Vaudel M. SearchGUI: a highly adaptable common interface for proteomics search and de Novo engines. J Proteome Res. 2018;17(7):2552–5.
Vaudel M, Burkhart JM, Zahedi RP, Oveland E, Berven FS, Sickmann A, et al. PeptideShaker enables reanalysis of MS-derived proteomics data sets. Nat Biotechnol. 2015;33(1):22–4.
Elias JE, Gygi SP. Target-decoy search strategy for increased confidence in large-scale protein identifications by mass spectrometry. Nat Methods. 2007;4(3):207–14.
Guldbrandsen A, Barsnes H, Kroksveen AC, Berven FS, Vaudel M. A simple workflow for large scale shotgun glycoproteomics. Methods Mol Biol. 2016;1394:275–86.
Vizcaino JA, Csordas A, Del-Toro N, Dianes JA, Griss J, Lavidas I, et al. 2016 update of the PRIDE database and its related tools. Nucleic Acids Res. 2016;44(D1):D447–56.
Sharma V, Eckels J, Schilling B, Ludwig C, Jaffe JD, MacCoss MJ, et al. Panorama public: a public repository for quantitative data sets processed in skyline. Mol Cell Proteomics. 2018;17(6):1239–44.
Schutzer SE, Angel TE, Liu T, Schepmoes AA, Xie F, Bergquist J, et al. Gray matter is targeted in first-attack multiple sclerosis. PLoS ONE. 2013;8(9):e66117.
Kroksveen AC, Guldbrandsen A, Vedeler C, Myhr KM, Opsahl JA, Berven FS. Cerebrospinal fluid proteome comparison between multiple sclerosis patients and controls. Acta Neurol Scand Suppl. 2012;195:90–6.
A.G. and H.B. are supported by the Research Council of Norway (251235). H.B. and R.R.L. are supported by the Trond Mohn Foundation (BFS2016REK02). FB is supported by the Western Norway Regional Health Authority (912166).
Consent for publication
The authors declare that they have no competing interests.
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
133 proteins significantly changed (=/± 50%) between MS and Non-MS from CSF-PR. Quantified in minimum two studies, and changed in same direction in minimum two studies.
287 proteins significantly changed between MS and Non-MS, but that were quantified in only one study in CSF-PR.
120 proteins from Table S1 found in the DDA protein depth experiment.
All 1194 identified proteins from DDA protein depth experiment.
Data from the peptide stability experiment.
S6. Data from the trypsin digestion experiment.
S1. Digestion profile plot across all time points for all peptides.
S2. Scatter plots comparing peptide amount (ratio L/H) at all times points tested in the digestion experiment.
S7. Data from the PRM RRMS vs. OND experiment.
Full table of all tested peptides with essential results from each test and whether they passed or failed the tests.
S9. Overview of the tests for stability, digestion and MS vs. OND experiments, for the 25 proteins. The colours and numbers refer to the peptides passing the specific test. Green: protein passed with two or more peptides; yellow: passed with one peptide; and, red: either no peptides passed (0) or none were tested (-).
S3. Calibration curves for 17 assay peptides that passed all quality control tests.
S4. Representative transition peaks from the Skyline analysis. A and B show typical examples of used transitions. The transition intensity, integration limits, retention time and mass error (ppm) is illustrated. C and D show examples of how 1-3 transitions were often much higher than the rest. Peak smoothing (Savitzky-Golay) was used in Skyline, which notably does not affect the quantification.
About this article
Cite this article
Guldbrandsen, A., Lereim, R.R., Jacobsen, M. et al. Development of robust targeted proteomics assays for cerebrospinal fluid biomarkers in multiple sclerosis. Clin Proteom 17, 33 (2020). https://doi.org/10.1186/s12014-020-09296-5
- Parallel reaction monitoring
- Cerebrospinal fluid
- Multiple sclerosis
- Assay development
- Neurological diseases