Final Report Summary - PROACTIVE (High throughput proteomics systems for accelerated profiling of putative plasma biomarkers)
The prospect of being able to analyse hundreds of low abundance proteins in blood samples at high throughput entails a promise of biomarker discovery leading to improved disease detection and thus better human health. The goal of the PROACTIVE consortium was therefore to develop a validated and fully operational high throughput targeted proteomics pipeline for analysis of human blood samples capable of detecting very large sets of putative cancer biomarkers.
The technology development of the PROACTIVE-project has centred on the proximity probing assays (PLA/PEA), which have been invented and developed to aid in the discovery and validation of protein biomarkers present in blood. Proximity probing is based on antibodies linked with synthetic oligonucleotide sequences, and upon proximal and pair-wise target protein binding, these molecular probes are united by the action of either a deoxyribonucleic acid (DNA) polymerase or a DNA ligase forming a set of sequences detectable by quantitative real-time polymerase chain reaction (qPCR). A number of unique features of these proximity probing technologies make them especially suitable for these types of studies. The assays have high sensitivity, multiplex capability, low sample consumption, and potential for rapid assay development time and high throughput.
First of all, the antibody / oligonucleotide conjugation protocol was improved in several ways that made it possible to generate several large panels of potentially relevant assays. Secondly, the degree of multiplexing was incrementally increased during the course of the project, going from 9-plex, 24-plex and up to 96-plex. Although the number of proteins analysed in the same sample was increased 10-fold, there was no apparent reduction in sensitivity (down to femtomolar detection), and the high specificity remained.
The consortium has also developed data management and analysis tools specifically tailored for use with the data generated by the detection technology combined with clinical sample information. The performance of the developed detection and analysis systems was validated in a pilot biomarker research project using biobanked plasma samples from colorectal cancer patients and non-colorectal cancer individuals. Altogether 35 biomarkers were identified as being increased selectively in colorectal cancer patients, of which some markers were able to detect colorectal cancer already at an early stage. Finally data generated with proximity ligation assay (PLA) and proximity extension assay (PEA) were compared to data generated with enzyme linked immunosorbent assay (ELISA) for the same samples demonstrating very good inter-technology correlation and reproducibility.
Currently, the proximity probing technology plat form is capable of quantifying 96 selected proteins in 96 blood samples in one day, consuming only 1 µL of each plasma sample. This technology can easily be utilised to detect measure several hundreds of putative biomarkers in larger sample collections with reliable results.
Project context and objectives:
The prospect of being able to analyse hundreds of low abundance proteins in blood samples at high throughput entails a promise of biomarker discovery leading to improved disease detection and thus better human health. No such capability exists today. The goal of the PROACTIVE consortium was therefore to develop a validated and fully operational high throughput targeted proteomics pipeline for analysis of human blood samples capable of detecting very large sets of putative cancer biomarkers. The consortium has also developed data management and analysis tools specifically tailored for use with the data generated by the detection technology combined with clinical sample information.
The objectives of the project were the following:
1. Develop a high throughput multiplexed protein detection technology for blood-based biomarkers. Capacity of at least 180 targeted proteins quantified in 100 human blood samples in less than one week (potentially only one day) and consuming only 4 µL of each sample.
2. Develop software tools for data management and analysis. With integrated laboratory information data management system (LIMS), data visualisation, and multivariate biostatistics for biomarker data and clinical patient information.
3. Validate performance of these detection and analysis systems in a pilot biomarker research project in biobanked plasma samples from colorectal cancer patients and non-colorectal cancer individuals. In practice, 312 biobanked colorectal cancer plasma samples and controls have been assayed.
4. Compare data from putative and clinically used biomarkers with orthogonal standard methods for protein detection to assess inter-technology correlation.
Detection technology and biomarkers indicative of disease states and outcomes are actively pursued around the world in both academia and industry. The term 'proactive diagnostics' has been coined to reflect the fields desire to shift from the current state of 'reactive diagnostics' to early stage disease detection where many diseases, especially cancer, are at their most curable stage. To enable this vision, screening test of high specificity (low false positive rate) and high sensitivity (low false negative rate) are needed. Multi-marker approaches are here essential for success in such a heterogeneous disease as cancer. These tests will eventually lead to screening of high-risk populations or even populations of asymptomatic individuals for early detection. Other critical applications in need of powerful biomarkers are patient management through disease stratification, eventually leading to personalised medicine.
The technology development of the PROACTIVE-project has centred on the proximity ligation / extension assays (PLA/PEA), which have been invented and developed to aid in the discovery and validation of protein biomarkers present in blood. A number of unique features of these technologies make them especially suitable for these types of studies. The assays have high sensitivity, multiplex capability, low sample consumption, and potential for rapid assay development time and high throughput. At its present state of development, PLA/PEA can detect and quantify tens of putative biomarkers in a few micro-litres of a blood sample with femtomolar sensitivity at medium to low throughput. The PROACTIVE consortium has improved upon this technology through innovative research in order to dramatically increase its capacity to be able to detect hundreds of putative biomarkers in large sample collections.
Multivariate data analysis is a critical component of all biomarker research. It aims at studying in a potentially very large space of tens of variables (each one being a 'dimension'), which are the key relationships in the data set. In particular, much of the available experience gained from large-scale mRNA profiling available should be directly applicable to this project. However, the large number of putative biomarkers available in comparison to the often quite limited number of patient samples in mRNA and protein profiling makes the analyses very challenging. In particular, classical statistical tools and theorems developed for large sample sizes are not valid. This has resulted in a variety of novel and therefore still immature approaches based on ideas merged mainly from statistics, computer science and engineering. In this project, the current research front in data analysis has been translated and adapted to develop a high throughput protein profiling system.
By coordinated research and innovation, the four small and medium-sized enterprises (SMEs) have collaborated to drive the development of these technologies and analysis tools into high throughput systems. Together with the academic partners of the consortium, the respective expertise of the SMEs has synergistically enhanced the quality of this project. The development of this high content protein detection tool is targeting applications within the early phases of biomarker research. Naturally, it requires expert input from several areas such as statistics, clinical research, and leading diagnostics industry in order to build a valid analysis system. Indeed, all these components have been assembled within the PROACTIVE-consortium bringing together wet lab technology developers (Olink and Innova Bioscience), in silico scientist in biostatistics (UAH and Integromics), clinical cancer research specialist (KU), and the diagnostics industry (FDAB).
The project has been conducted in phases of incremental levels of development of the technologies as of sample capacity, biomarker content and analysis tools capability. Three pilot studies of successive system versions have been carried out during the development of this biomarker research pipeline. At these events all components of the system developed at that stage have been tested and the results verified by orthogonal / standard methods.
Project results:
Development and improvements of the conjugation protocol
Innova has continued to refine methods for attaching oligonucleotides to antibodies in the light of feedback from colleagues about the performance of the different types of conjugates in PEA/PLA assays. While there is now a choice of possible conjugation methods on the project, antibody reagents prepared from one of the more complex protocols (the 'four-step method'), which includes a chromatographic separation step, generally give the best performance in PLA/PEA assays, possibly because higher levels of oligonucleotide can be incorporated into the conjugate. For production purposes the four-step method is not particularly attractive because of its column separation step, which causes dilution of the sample and negatively impacts scalability. This step also introduces reagent losses that are hard to quantify when starting with small amounts of antibody (e.g. typically 100 micrograms from commercial sources). This means that the concentrations and ratios of reactants in the subsequent conjugation reactions are not known with the required certainty, and dilution of the sample inevitably reduces conjugation efficiency.
At the end of the last reporting period, we alluded to a promising three-step variant method, which does not contain the awkward column separation of the four-step method. The new three-step method has now been fully optimised and shown to be applicable to antibodies from mouse, goat, sheep, donkey, and rabbit. It is likely that the procedure will be applicable to many other species as well. The new three-step method allows tight control over the level of oligonucleotide incorporation into antibodies, thus conjugate performance to be fine-tuned for each antibody or for particular applications if required. By altering the reaction conditions the amount of attached oligonucleotide may be varied at will, as evidenced by the variable band shift on a sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) gel.
Based on PLA/PEA assay results a relatively high level of oligo incorporation is desirable. Conjugates are typically prepared with 50 ug of antibody (half of a vial), allowing the remaining 50 µg to be used as a back up or for further production if required. Typically twenty conjugates are made in parallel and then analysed by SDS-PAGE before the generation of the next batch is initiated. The new three-step method generates conjugates that exhibit good sensitivity in PEA/PLA assays. Compared with some of the alternative one-step methods the three-step method is more resistant to interfering substances found in some preparations of antibodies and can be used with as little as 25 µg of antibody (albeit somewhat higher than the 5-10 µg possible with one-step methods).
For ease of operation and to facilitate parallel working oligonucleotides preferably are first activated and freeze-dried. The high stability of material stored in this way permits the preparation of a large bank of 'conjugation-ready' oligonucleotides. For example the entire set of oligonucleotides prepared in early 2011 for milestone #3 were re-used several months later to prepare nearly 200 oligonucleotide conjugates in separate reactions with a donkey-anti-chicken antibody for assessing relative oligonucleotide-pair efficiency in PEA.
Despite the fact that free oligonucleotide is present in conjugates the PLA and PEA technology appears sufficiently robust that purification of the conjugates are not required as a matter of routine. The three-step method is ideal for antibodies that are available in purified form in a suitable formulation. An upstream purification module has been inserted in about 10 % of cases, either because antibodies are impure (e.g. crude serum, ascites fluid) or because undesirable additives are present (e.g. BSA, interfering compounds). Whilst methods of purification are well known we have reworked the various procedures so that only conjugation friendly reagents are employed.
In summary, a three-step method without any column separation steps has been employed for the production of hundreds of conjugates to support the development of PLA and especially the newer PEA assay technology. The three-step method has now become the method of choice for conjugate production on the project.
Development of the PEA
Previous work on PLA demonstrated that accurate protein quantifications were difficult to perform in complex biological samples, such as blood plasma as the DNA ligase activity seem to be inhibited to varying degree in different samples (recovery loss). In parallel with the work on milestone #3 we developed a new technology called the PEA. PEA is based on the same concept as PLA but uses a DNA polymerase to unite the two probe oligos instead of a DNA ligase.
As with PLA, antibodies (either two matched mAb, or one batch of pAb split in two fractions) are covalently linked with two different oligonucleotides at their 5'-ends. To one of the probes, a 56-mer DNA oligo (extension primer) comprising 40 nt that are complementary to probe A, a 7 nt spacer, and 9 nt complementary to the corresponding probe B is hybridised. Next, the hybridised probe pair is incubated with a sample containing the antigen of interest (e.g. blood plasma). This results in binding between the probe pair and the antigen, and as a result, the probe oligonucleotides come in close proximity and hybridize to each other. The addition of a DNA polymerase leads to an extension of the extension primer over probe oligo B. Finally, the generated DNA template can be detected and quantified by qPCR.
Recovery (the difference in measured signal between a complex and a non-complex matrix) reflects the ability of an assay to accurately quantify an analyte in biological materials. As mentioned above, the activity of DNA ligase utilised in these assays was impaired in blood plasma, sometimes resulting in poor recovery. The average recovery determined for 13 PLA assays was only 33 %, but with PEA the average recovery for the same analytes was 77 % (data not shown). PEA was found to perform equally well as PLA with regards to all other immunoassay parameters tested. Therefore, we decided to run all panel evaluations and the final biomarker study for milestone #3 with PEA.
Development of a 96-plex panel
Initially we planned to assess 180 markers (and 20 repeats) in 4 x 78 samples by multiplex PLA. After the successful completion of milestone #2, and with the successful development of a new technology, we decided to use PEA for all further optimisations, evaluations and the pilot biomarker study. Furthermore, we aimed at increasing the number of analytes to be assessed in multiplex to 96. By doing so, we would be generating a sensitive antibody-based immunoassay with the highest degree of multiplexing to date. All-in-all we generated 5 biomarker panels: four 24-plex and one 96-plex PEA panels.
In the 24-plex panels generated in milestone #2 we had included three non-human proteins, PE, GFP, APC, as well as a DNA oligo representing a complete PLA product as spike-in controls. As APC never worked well as a spike-in control in the 24-plex panels it was replaced with an 'extension control' in the 96-plex panel. The extension control was included as a way to measure the efficiency of the extension reaction in different samples.
In the work in milestone #1 and #2 we had some difficulties getting a robust pre-amplification of all amplicons for all analytes. In order to improve the pre-amplification, and thereby achieving better data quality, we redesigned all DNA sequences. About 2000 sequences were taken from the literature and used to build the new sequence sets including: probe oligos A and B, extension primers, unique and universal Frw and Rev primers, and a binding site for molecular beacon. All sequences were tested in silico for their tendency to form hairpin formation and ranked accordingly. The top 96 sequences were used for conjugations.
In pilot biomarker studies #1 and #2, we had problems with uneven pre-amplification in the sense that some sequences were amplified better than others, and that the precision was poor for some assays. In order to achieve a better pre-amplification step, new pre-amplification primer sites were added on both probe oligonucleotides. Primers were removed after the preamplification step. As in the previous design, unique primer sites are also present in each probe oligo for qPCR detection.
In order to evaluate the sequences used for conjugation, we conjugated all of the selected sequences to the same antibody (donkey anti-chicken) and measured their response to 10 pM chicken IgY. When plotting the signal-to-noise levels as a function of the background signal in buffer, it was striking that a lower buffer background level leads to higher signal-to-noise levels, and vv. When comparing the background levels for the anti-chicken conjugates to those of the 96-plex panel, we found that the signal and signal-to-noise levels are to a large part determined by which particular sequences are being used for the conjugations. These results are very informative for selecting sequences for future 96-plex assays.
Evaluations of preamplification and qPCR protocols
The pilot #1 and #2 biomarker studies used 48 x 48 microfluidic chips that were analysed with BioMark™ real time PCR system. Following that work, an upgraded chip version with a higher density (96 x 96) was made available. The 96 x 96 format suited our 96-plex format perfectly, since all analytes can be analysed on the same chip, the number of pipetting steps was even further reduced, and the number of chips run was reduced by 4 when comparing with the 48 x 48 system. Therefore we decided to run the pilot biomarker study #3 on this new system. We purchased the corresponding IFC Controller and chips. In addition, the risk of making mistake when pipetting is also lowered, and the data analysis is faster with the 96 x 96 system. We decided to use the 96 x 96 system also for the 24-plex panels, even though the 24-plex panels would fit on the 48 x 48 chips, mainly because the 96 x 96 system would give us quadruplicates for all data points, and thereby improve data quality.
To assess whether there were other assays that would run sub-optimally because of poor qPCR efficiency, the amplification efficiency was assessed for all primers. This was achieved by performing serial dilutions of a pre-amplified Ag-spiked plasma sample and measured the qPCR signal at each dilution. For some assays the number of qPCR templates was too low to get good estimations, and their data was excluded from the analysis. The median efficiency was always at least 70 % and on average 91 %.
Evaluation of immunoassay parameters
The sensitivity of the 96-plex PEA assay was estimated for 44 assays, for which antigen was available, using the following formula: [Sensitivity is the concentration of spike-in closest to background /2^dCt]. 12 assays had a sensitivity <1 pM, and the median sensitivity across the panel was around 3 pM. Previous evaluations of the 24-plex PLA and PEA assays demonstrated better sensitivities. All but 12 or 13 of 60 assays displayed a sensitivity of less than 1 pM. The median sensitivity was 0.2 and 0.3 pM, and down to 11 and 3 fM for PLA and PEA, respectively. However, these estimations were done using slightly different protocols for panel evaluations. To get a fair estimation, we calculated the median fold-change between healthy plasma samples and buffer background for the four markers present in both 24- and 96-plex panels. The fold-change was similar with the different panels suggesting that the different conjugate designs were equally good with respect to generating sensitive assays.
In addition, with more sequences being included in the panel the number of possible interactions and unspecific events are exponentially increased. Therefore, it is of certain importance to test the assay specificity. This was done by analysing samples in which only a subset of the antigens are present and at varying concentrations. For the antigen combinations tested there was no indication of unspecific signal for either of the conjugates.
The level of multiplexing of PLA/PEA had not previously exceeded 24. With more conjugates included in the panel there is a risk that interference between sequences/Abs might lead to a decrease in sensitivity. To test if this was the case, we compared the signal-to-noise levels for some conjugates using panels with increasing levels of multiplexing. For some conjugates, we compared their performance when included in a 16-, 32- or 96-plex panel. However, the signal-to-noise does not seem to be significantly affected by the different degree of multiplexing.
The 96 anti-chicken conjugate pairs from above were used to study the relationship between signal level and precision. As observed with the 24-plex PLA/PEA and with the 96-plex PEA panel, we found that the assay precision worsened (increased CV %) with decreased signal. The level of variation was further assessed for different parts of the PEA procedure by analysing replicate samples. For both pre-amplification and qPCR, there was a clear correlation between the signal level of the assay and variation; the higher the signal, the better precision.
The assay precision in plasma was determined for the spike-in proteins, for which the expected plasma levels are equal for all samples. Their signals across all samples on one PEA plate were used to compare precision (CV %) between the 24- and 96-plex assays, before or after normalisations. This demonstrated that, given a sufficient signal level of the assay, the precision was significantly improved in the 96-plex panel: CV of 25 % before and 10 % after normalisation. This is very good precision as the qPCR readout alone contributes to a variation of 5 %. The improved precision is likely due to a more robust pre-amplification and qPCR attained by using universal pre-amplification primers that are efficiently removed before qPCR. After summarising the panel evaluations, it is clear that one must ensure that the signal level for all conjugates is sufficient to obtain good overall assay performance.
24-plex versus 96-plex
Finally, data from immunoassay evaluations were summarised and compared for the 24-plex versus 96-plex panels. As discussed above, we found that the average sensitivity was better with the 24-plex and the fraction of analytes detectable in plasma was higher. This is partly explained by the use of lower probe concentration in the 24-plex assays. However, when comparing these two parameters on assays present in both PEA panel types, we found no significant difference. This suggested to us that the lower success rate obtained with the 96-plex panel was most likely due to the use of less well-studied analytes of which many had unknown, or low expected levels in plasma. When addressing precision for the different panel types by determining the variation of the PE signal across an entire PEA plate (93 plasma samples) before (37 to 70 % versus 25%) and after (17 to 28 % versus 10 %) GFP normalisations. This demonstrated that the new sequence design and new pre-amp protocol significantly improved precision, and that our normalisation strategy seems potent.
LIMS
A LIMS was developed by Integromics on top of an existing code base. The LIMS is built as a client-server application where the data is stored in a central repository (server) that is accessed using a web browser (client). This approach makes it very easy to share experiments between researchers and to roll out updates of the client. The data model was designed specifically for the needs of the PROACTIVE project, incorporating not only experiment parameters but also clinical information about the samples.
A prototype was implemented based on a requirements document produced by Olink and Copenhagen University, and the design was refined for each deliverable. A development system and a production system were hosted in parallel allowing testing of features without interfering with production at the address http://proactive.integromics.com/proactive/ Visualisations useful for quality control were implemented within the LIMS. Integration with more advanced data compression, visualisation and statistical analyses provided through an export feature. Raw data can be exported from the LIMS and normalised in a desktop software application developed by Uppsala University. The application outputs normalised data in the Microsoft Excel file format which the majority of statistical software applications can work with, including free open source suites such as R. This solution was deemed to produce the maximal amount of flexibility for downstream data analyses.
Multivariate analysis
For each release of PLA data, multivariate analysis work by Uppsala University within WP2 utilised principal component analysis and hierarchical clustering and other visualisation techniques for multidimensional data to identify subgroups in the data set. Then a variety of different machine learning techniques were evaluated for design of detectors of colorectal cancer, such as discriminant analysis in combination with forward and backward feature selection methods, k-nearest neighbour classification, support vector machine classification, nearest shrunken centroid classification and random forests. The main focus was the establishment of a procedure for demonstrating presence of a signal for classification of colorectal cancer samples based on the proximity ligand assay. Of the methods evaluated, random forest classification emerged as a good choice for finding a biomarker signature that can be used to predict colorectal cancer (CRC) status in plasma samples. Random forest is an ensemble method that builds many models that each casts a vote for the predicted group label. Hence interpretation of the model structure is difficult, but the value of each potential marker in predicting the label can be estimated using variable importance measures (data not shown). Performance and variable importance were estimated using re-sampling, i.e. by running the procedure on random subsamples of the data set to assess the variability of performance and variable importance measures in small data sets. Assays important for classification can then be further studied in a larger set of samples.
Classifications
In full agreement with the project plan, the research activity within WP2 has focussed on procedures for obtaining good performance estimates for small data sets. For instance, the cross-validation procedure in common use only produces a point estimate of the expected performance of a classifier built using datasets of a given size sampled from the population. However, the variability in performance is also of interest when choosing between different designing procedures. Hence a procedure using re-sampling for estimating the upper bound of the error rate was devised. Towards that end, a novel kernel density estimator was developed to be used for generating new samples to be used in estimating the distribution of error rates. Using simulations, it was shown, that except in rare cases, choosing between design procedures using expected error rate yields the same choice as the upper bound. Therefore, focus shifted towards computing confidence interval for the expected error rate. A procedure for producing a conservative estimate of the confidence interval for the expected performance was found in the literature and was subsequently used in analysing the performance. For instance, for PLA release #2, expected classification error rate of CRC versus healthy using random forest was estimated at 29 % with the conservative 95 % confidence interval (22 %, 37 %), for PLA release #3 the corresponding (preliminary) figures are 24 % (13 %, 36 %).
Plasma samples
The sample set was selected from a larger endoscopy study where subjects undergoing a sigmoidoscopy or colonoscopy either following symptoms consistent with CRC or patients attending surveillance programs due to hereditary CRC (HNPCC and FAP) were included in a cross sectional study. A total of 5165 subjects were included and according to the Helsinki II Declaration, oral and written consent was given from each subject. The study was approved by the Regional Ethical Committee of Greater Copenhagen, Denmark (KF 01-080/03). A case-control study was designed for the present EU project by randomly selecting 78 biobanked stage I-IV CRC samples. By matching age and gender of the 78 CRC patients, 78 individuals with no pathological findings by endoscopy and/or no self reported diseases or medication, 78 individuals with adenoma, and 78 individuals with non-cancer-disease were selected from the cohort. Due to size limitations in the current technical setup the analyses were restricted to 70 samples from each of the above mentioned groups resulting in a total of 280 samples.
Biomarker screening
For this screening, the basic proximity probing assay was shifted from PLA to PEA (described above). The 24-plex PLA assays (Lundberg et al., 2011) was applied as a model to transfer the PLA approach to the PEA. Screening #3 was divided into four panels in 24-plex and a single panel of 96-plex. The biomarkers in the 24-plex are identical to PLA release two and the 96-plex is composed of novel biomarkers and a few repeats from the 24-plex panels. For this screening all four groups (CRC, healthy, adenoma and non-cancer disease) were included.
In close collaboration with Olink, new putative biomarkers were selected for Biomarker screening #3. For the new PEA 96-plex, the biomarkers were carefully selected by thorough literature searches', including criteria's such as proteins present in pathways involved in CRC, proteins involved in inflammation and cancer, proteins found in screenings looking into the tumour tissue, and general cancer markers were included. Another important criterion was again the availability of appropriate antibodies. Based on the recommendations from the reviewers it was decided to focus on novel low abundance proteins, as the proximity assays possess a high sensitivity. To achieve this, we included markers with reported low plasma levels and several markers based on their increased expression in CRC tissue, with no available data on plasma levels.
Correlations between ELISA, PLA and PEA
At KU, six proteins have been quantified by ELISA in samples from healthy individuals and CRC patients. Correlations between data from PLA release #2 and PEA release #1 (WP3) and data from the different ELISAs has been made for these. Generally, the Spearman correlation (R) between PLA release#2 and PEA#3 was good. For all the analysed assays, except for one, we found that switching from PLA to PEA improved the Spearman correlation (R) to the ELISA data.
Identification of CRC-specific biomarkers
We focused on CRC specific markers by comparing CRC to each of the three other groups (healthy, non-cancer disease and adenoma). Applying these criteria we identified 35 assays which display specific changes only in CRC (data not shown). Altogether, we have identified 35 assays out of 150 individual assays (23 %) that specifically identify CRC from the other groups in the case-control study. Among the 96-plex assays, which comprised 86 assays, we identified six novel CRC specific assays. These markers are very promising for further investigation since they were identified among potential marker of low plasma abundance and tumour tissue biomarkers.
Potential impact:
The results generated within the PROACTIVE project have been presented in a several large scientific conferences, such as HUPO (Sidney/Geneva 2010/2011) and AACR (Philadelphia 2010), and published in two peer-reviewed journals (M. Lundberg et al., Nucleic Acids Res, 2011 and M. Lundberg et al., Mol Cell Proteomics, 2011).
The PROACTIVE project has enabled Olink to develop our proximity probing technologies both in terms of assay performance and at the multiplexing level. A plasma biomarker service based on the biomarker panels generated within the project have been released and resulted in a large interest amongst existing and new customers, generated significant revenue, as well as important collaborations for Olink. We are now continuing our work on the 96-plex protein detection technology and plan to develop and commercialise four panels within the coming year. This will make high-throughput biomarker screenings available for academic laboratories as well as biotech companies, and hopefully lead to the discovery of many new interesting biomarkers and/or biomarker panels.
The chosen strategy of developing panels of well selected and qualified analytes has several implications. It will be more indicative of the status of a disease than just analysing with any single marker, and help to reduce sample costs efficiently due to the high throughput capacity of the PLA platforms. Such a powerful tool will contribute to improve health care but also to significantly increase competitiveness and commercial success of the participating SMEs as well as to the PROACTIVE seven scientific recognitions for the academic partners. The intellectual property will be commercialised by the consortium participants as proprietary products and by out-licensing agreements.
CRC is the second most frequent cancer type second only to breast cancer. It is proven that colorectal cancer screening reduces disease-related mortality and is cost-effective. However, current compliance is poor. The potential use of the PROACTIVE high throughput technology enables biomarker screening to define a set of protein markers that potentially will identify individuals who should undergo colonoscopy by placing them in a certain risk group. CRC was chosen as the tumour type because of its enormous social and economic impact. In addition, the knowledge obtained from our studies in patients with CRC may be translated into the adjuvant treatment and perhaps into other cancer diseases and thereby be of benefit for a significant additional number of cancer patients. At the conclusion of this collaborative project, the research and innovation conducted within the consortium will enable the partners to position themselves at the very forefront of high-throughput biomarker research strengthening their competitiveness in the international arena.
Project website: http://www.olink.com/proactive
Simon Fredriksson
simon.fredriksson@olink.com via e-mail
Telephone: +46-184-443984
The technology development of the PROACTIVE-project has centred on the proximity probing assays (PLA/PEA), which have been invented and developed to aid in the discovery and validation of protein biomarkers present in blood. Proximity probing is based on antibodies linked with synthetic oligonucleotide sequences, and upon proximal and pair-wise target protein binding, these molecular probes are united by the action of either a deoxyribonucleic acid (DNA) polymerase or a DNA ligase forming a set of sequences detectable by quantitative real-time polymerase chain reaction (qPCR). A number of unique features of these proximity probing technologies make them especially suitable for these types of studies. The assays have high sensitivity, multiplex capability, low sample consumption, and potential for rapid assay development time and high throughput.
First of all, the antibody / oligonucleotide conjugation protocol was improved in several ways that made it possible to generate several large panels of potentially relevant assays. Secondly, the degree of multiplexing was incrementally increased during the course of the project, going from 9-plex, 24-plex and up to 96-plex. Although the number of proteins analysed in the same sample was increased 10-fold, there was no apparent reduction in sensitivity (down to femtomolar detection), and the high specificity remained.
The consortium has also developed data management and analysis tools specifically tailored for use with the data generated by the detection technology combined with clinical sample information. The performance of the developed detection and analysis systems was validated in a pilot biomarker research project using biobanked plasma samples from colorectal cancer patients and non-colorectal cancer individuals. Altogether 35 biomarkers were identified as being increased selectively in colorectal cancer patients, of which some markers were able to detect colorectal cancer already at an early stage. Finally data generated with proximity ligation assay (PLA) and proximity extension assay (PEA) were compared to data generated with enzyme linked immunosorbent assay (ELISA) for the same samples demonstrating very good inter-technology correlation and reproducibility.
Currently, the proximity probing technology plat form is capable of quantifying 96 selected proteins in 96 blood samples in one day, consuming only 1 µL of each plasma sample. This technology can easily be utilised to detect measure several hundreds of putative biomarkers in larger sample collections with reliable results.
Project context and objectives:
The prospect of being able to analyse hundreds of low abundance proteins in blood samples at high throughput entails a promise of biomarker discovery leading to improved disease detection and thus better human health. No such capability exists today. The goal of the PROACTIVE consortium was therefore to develop a validated and fully operational high throughput targeted proteomics pipeline for analysis of human blood samples capable of detecting very large sets of putative cancer biomarkers. The consortium has also developed data management and analysis tools specifically tailored for use with the data generated by the detection technology combined with clinical sample information.
The objectives of the project were the following:
1. Develop a high throughput multiplexed protein detection technology for blood-based biomarkers. Capacity of at least 180 targeted proteins quantified in 100 human blood samples in less than one week (potentially only one day) and consuming only 4 µL of each sample.
2. Develop software tools for data management and analysis. With integrated laboratory information data management system (LIMS), data visualisation, and multivariate biostatistics for biomarker data and clinical patient information.
3. Validate performance of these detection and analysis systems in a pilot biomarker research project in biobanked plasma samples from colorectal cancer patients and non-colorectal cancer individuals. In practice, 312 biobanked colorectal cancer plasma samples and controls have been assayed.
4. Compare data from putative and clinically used biomarkers with orthogonal standard methods for protein detection to assess inter-technology correlation.
Detection technology and biomarkers indicative of disease states and outcomes are actively pursued around the world in both academia and industry. The term 'proactive diagnostics' has been coined to reflect the fields desire to shift from the current state of 'reactive diagnostics' to early stage disease detection where many diseases, especially cancer, are at their most curable stage. To enable this vision, screening test of high specificity (low false positive rate) and high sensitivity (low false negative rate) are needed. Multi-marker approaches are here essential for success in such a heterogeneous disease as cancer. These tests will eventually lead to screening of high-risk populations or even populations of asymptomatic individuals for early detection. Other critical applications in need of powerful biomarkers are patient management through disease stratification, eventually leading to personalised medicine.
The technology development of the PROACTIVE-project has centred on the proximity ligation / extension assays (PLA/PEA), which have been invented and developed to aid in the discovery and validation of protein biomarkers present in blood. A number of unique features of these technologies make them especially suitable for these types of studies. The assays have high sensitivity, multiplex capability, low sample consumption, and potential for rapid assay development time and high throughput. At its present state of development, PLA/PEA can detect and quantify tens of putative biomarkers in a few micro-litres of a blood sample with femtomolar sensitivity at medium to low throughput. The PROACTIVE consortium has improved upon this technology through innovative research in order to dramatically increase its capacity to be able to detect hundreds of putative biomarkers in large sample collections.
Multivariate data analysis is a critical component of all biomarker research. It aims at studying in a potentially very large space of tens of variables (each one being a 'dimension'), which are the key relationships in the data set. In particular, much of the available experience gained from large-scale mRNA profiling available should be directly applicable to this project. However, the large number of putative biomarkers available in comparison to the often quite limited number of patient samples in mRNA and protein profiling makes the analyses very challenging. In particular, classical statistical tools and theorems developed for large sample sizes are not valid. This has resulted in a variety of novel and therefore still immature approaches based on ideas merged mainly from statistics, computer science and engineering. In this project, the current research front in data analysis has been translated and adapted to develop a high throughput protein profiling system.
By coordinated research and innovation, the four small and medium-sized enterprises (SMEs) have collaborated to drive the development of these technologies and analysis tools into high throughput systems. Together with the academic partners of the consortium, the respective expertise of the SMEs has synergistically enhanced the quality of this project. The development of this high content protein detection tool is targeting applications within the early phases of biomarker research. Naturally, it requires expert input from several areas such as statistics, clinical research, and leading diagnostics industry in order to build a valid analysis system. Indeed, all these components have been assembled within the PROACTIVE-consortium bringing together wet lab technology developers (Olink and Innova Bioscience), in silico scientist in biostatistics (UAH and Integromics), clinical cancer research specialist (KU), and the diagnostics industry (FDAB).
The project has been conducted in phases of incremental levels of development of the technologies as of sample capacity, biomarker content and analysis tools capability. Three pilot studies of successive system versions have been carried out during the development of this biomarker research pipeline. At these events all components of the system developed at that stage have been tested and the results verified by orthogonal / standard methods.
Project results:
Development and improvements of the conjugation protocol
Innova has continued to refine methods for attaching oligonucleotides to antibodies in the light of feedback from colleagues about the performance of the different types of conjugates in PEA/PLA assays. While there is now a choice of possible conjugation methods on the project, antibody reagents prepared from one of the more complex protocols (the 'four-step method'), which includes a chromatographic separation step, generally give the best performance in PLA/PEA assays, possibly because higher levels of oligonucleotide can be incorporated into the conjugate. For production purposes the four-step method is not particularly attractive because of its column separation step, which causes dilution of the sample and negatively impacts scalability. This step also introduces reagent losses that are hard to quantify when starting with small amounts of antibody (e.g. typically 100 micrograms from commercial sources). This means that the concentrations and ratios of reactants in the subsequent conjugation reactions are not known with the required certainty, and dilution of the sample inevitably reduces conjugation efficiency.
At the end of the last reporting period, we alluded to a promising three-step variant method, which does not contain the awkward column separation of the four-step method. The new three-step method has now been fully optimised and shown to be applicable to antibodies from mouse, goat, sheep, donkey, and rabbit. It is likely that the procedure will be applicable to many other species as well. The new three-step method allows tight control over the level of oligonucleotide incorporation into antibodies, thus conjugate performance to be fine-tuned for each antibody or for particular applications if required. By altering the reaction conditions the amount of attached oligonucleotide may be varied at will, as evidenced by the variable band shift on a sodium dodecyl sulfate polyacrylamide gel electrophoresis (SDS-PAGE) gel.
Based on PLA/PEA assay results a relatively high level of oligo incorporation is desirable. Conjugates are typically prepared with 50 ug of antibody (half of a vial), allowing the remaining 50 µg to be used as a back up or for further production if required. Typically twenty conjugates are made in parallel and then analysed by SDS-PAGE before the generation of the next batch is initiated. The new three-step method generates conjugates that exhibit good sensitivity in PEA/PLA assays. Compared with some of the alternative one-step methods the three-step method is more resistant to interfering substances found in some preparations of antibodies and can be used with as little as 25 µg of antibody (albeit somewhat higher than the 5-10 µg possible with one-step methods).
For ease of operation and to facilitate parallel working oligonucleotides preferably are first activated and freeze-dried. The high stability of material stored in this way permits the preparation of a large bank of 'conjugation-ready' oligonucleotides. For example the entire set of oligonucleotides prepared in early 2011 for milestone #3 were re-used several months later to prepare nearly 200 oligonucleotide conjugates in separate reactions with a donkey-anti-chicken antibody for assessing relative oligonucleotide-pair efficiency in PEA.
Despite the fact that free oligonucleotide is present in conjugates the PLA and PEA technology appears sufficiently robust that purification of the conjugates are not required as a matter of routine. The three-step method is ideal for antibodies that are available in purified form in a suitable formulation. An upstream purification module has been inserted in about 10 % of cases, either because antibodies are impure (e.g. crude serum, ascites fluid) or because undesirable additives are present (e.g. BSA, interfering compounds). Whilst methods of purification are well known we have reworked the various procedures so that only conjugation friendly reagents are employed.
In summary, a three-step method without any column separation steps has been employed for the production of hundreds of conjugates to support the development of PLA and especially the newer PEA assay technology. The three-step method has now become the method of choice for conjugate production on the project.
Development of the PEA
Previous work on PLA demonstrated that accurate protein quantifications were difficult to perform in complex biological samples, such as blood plasma as the DNA ligase activity seem to be inhibited to varying degree in different samples (recovery loss). In parallel with the work on milestone #3 we developed a new technology called the PEA. PEA is based on the same concept as PLA but uses a DNA polymerase to unite the two probe oligos instead of a DNA ligase.
As with PLA, antibodies (either two matched mAb, or one batch of pAb split in two fractions) are covalently linked with two different oligonucleotides at their 5'-ends. To one of the probes, a 56-mer DNA oligo (extension primer) comprising 40 nt that are complementary to probe A, a 7 nt spacer, and 9 nt complementary to the corresponding probe B is hybridised. Next, the hybridised probe pair is incubated with a sample containing the antigen of interest (e.g. blood plasma). This results in binding between the probe pair and the antigen, and as a result, the probe oligonucleotides come in close proximity and hybridize to each other. The addition of a DNA polymerase leads to an extension of the extension primer over probe oligo B. Finally, the generated DNA template can be detected and quantified by qPCR.
Recovery (the difference in measured signal between a complex and a non-complex matrix) reflects the ability of an assay to accurately quantify an analyte in biological materials. As mentioned above, the activity of DNA ligase utilised in these assays was impaired in blood plasma, sometimes resulting in poor recovery. The average recovery determined for 13 PLA assays was only 33 %, but with PEA the average recovery for the same analytes was 77 % (data not shown). PEA was found to perform equally well as PLA with regards to all other immunoassay parameters tested. Therefore, we decided to run all panel evaluations and the final biomarker study for milestone #3 with PEA.
Development of a 96-plex panel
Initially we planned to assess 180 markers (and 20 repeats) in 4 x 78 samples by multiplex PLA. After the successful completion of milestone #2, and with the successful development of a new technology, we decided to use PEA for all further optimisations, evaluations and the pilot biomarker study. Furthermore, we aimed at increasing the number of analytes to be assessed in multiplex to 96. By doing so, we would be generating a sensitive antibody-based immunoassay with the highest degree of multiplexing to date. All-in-all we generated 5 biomarker panels: four 24-plex and one 96-plex PEA panels.
In the 24-plex panels generated in milestone #2 we had included three non-human proteins, PE, GFP, APC, as well as a DNA oligo representing a complete PLA product as spike-in controls. As APC never worked well as a spike-in control in the 24-plex panels it was replaced with an 'extension control' in the 96-plex panel. The extension control was included as a way to measure the efficiency of the extension reaction in different samples.
In the work in milestone #1 and #2 we had some difficulties getting a robust pre-amplification of all amplicons for all analytes. In order to improve the pre-amplification, and thereby achieving better data quality, we redesigned all DNA sequences. About 2000 sequences were taken from the literature and used to build the new sequence sets including: probe oligos A and B, extension primers, unique and universal Frw and Rev primers, and a binding site for molecular beacon. All sequences were tested in silico for their tendency to form hairpin formation and ranked accordingly. The top 96 sequences were used for conjugations.
In pilot biomarker studies #1 and #2, we had problems with uneven pre-amplification in the sense that some sequences were amplified better than others, and that the precision was poor for some assays. In order to achieve a better pre-amplification step, new pre-amplification primer sites were added on both probe oligonucleotides. Primers were removed after the preamplification step. As in the previous design, unique primer sites are also present in each probe oligo for qPCR detection.
In order to evaluate the sequences used for conjugation, we conjugated all of the selected sequences to the same antibody (donkey anti-chicken) and measured their response to 10 pM chicken IgY. When plotting the signal-to-noise levels as a function of the background signal in buffer, it was striking that a lower buffer background level leads to higher signal-to-noise levels, and vv. When comparing the background levels for the anti-chicken conjugates to those of the 96-plex panel, we found that the signal and signal-to-noise levels are to a large part determined by which particular sequences are being used for the conjugations. These results are very informative for selecting sequences for future 96-plex assays.
Evaluations of preamplification and qPCR protocols
The pilot #1 and #2 biomarker studies used 48 x 48 microfluidic chips that were analysed with BioMark™ real time PCR system. Following that work, an upgraded chip version with a higher density (96 x 96) was made available. The 96 x 96 format suited our 96-plex format perfectly, since all analytes can be analysed on the same chip, the number of pipetting steps was even further reduced, and the number of chips run was reduced by 4 when comparing with the 48 x 48 system. Therefore we decided to run the pilot biomarker study #3 on this new system. We purchased the corresponding IFC Controller and chips. In addition, the risk of making mistake when pipetting is also lowered, and the data analysis is faster with the 96 x 96 system. We decided to use the 96 x 96 system also for the 24-plex panels, even though the 24-plex panels would fit on the 48 x 48 chips, mainly because the 96 x 96 system would give us quadruplicates for all data points, and thereby improve data quality.
To assess whether there were other assays that would run sub-optimally because of poor qPCR efficiency, the amplification efficiency was assessed for all primers. This was achieved by performing serial dilutions of a pre-amplified Ag-spiked plasma sample and measured the qPCR signal at each dilution. For some assays the number of qPCR templates was too low to get good estimations, and their data was excluded from the analysis. The median efficiency was always at least 70 % and on average 91 %.
Evaluation of immunoassay parameters
The sensitivity of the 96-plex PEA assay was estimated for 44 assays, for which antigen was available, using the following formula: [Sensitivity is the concentration of spike-in closest to background /2^dCt]. 12 assays had a sensitivity <1 pM, and the median sensitivity across the panel was around 3 pM. Previous evaluations of the 24-plex PLA and PEA assays demonstrated better sensitivities. All but 12 or 13 of 60 assays displayed a sensitivity of less than 1 pM. The median sensitivity was 0.2 and 0.3 pM, and down to 11 and 3 fM for PLA and PEA, respectively. However, these estimations were done using slightly different protocols for panel evaluations. To get a fair estimation, we calculated the median fold-change between healthy plasma samples and buffer background for the four markers present in both 24- and 96-plex panels. The fold-change was similar with the different panels suggesting that the different conjugate designs were equally good with respect to generating sensitive assays.
In addition, with more sequences being included in the panel the number of possible interactions and unspecific events are exponentially increased. Therefore, it is of certain importance to test the assay specificity. This was done by analysing samples in which only a subset of the antigens are present and at varying concentrations. For the antigen combinations tested there was no indication of unspecific signal for either of the conjugates.
The level of multiplexing of PLA/PEA had not previously exceeded 24. With more conjugates included in the panel there is a risk that interference between sequences/Abs might lead to a decrease in sensitivity. To test if this was the case, we compared the signal-to-noise levels for some conjugates using panels with increasing levels of multiplexing. For some conjugates, we compared their performance when included in a 16-, 32- or 96-plex panel. However, the signal-to-noise does not seem to be significantly affected by the different degree of multiplexing.
The 96 anti-chicken conjugate pairs from above were used to study the relationship between signal level and precision. As observed with the 24-plex PLA/PEA and with the 96-plex PEA panel, we found that the assay precision worsened (increased CV %) with decreased signal. The level of variation was further assessed for different parts of the PEA procedure by analysing replicate samples. For both pre-amplification and qPCR, there was a clear correlation between the signal level of the assay and variation; the higher the signal, the better precision.
The assay precision in plasma was determined for the spike-in proteins, for which the expected plasma levels are equal for all samples. Their signals across all samples on one PEA plate were used to compare precision (CV %) between the 24- and 96-plex assays, before or after normalisations. This demonstrated that, given a sufficient signal level of the assay, the precision was significantly improved in the 96-plex panel: CV of 25 % before and 10 % after normalisation. This is very good precision as the qPCR readout alone contributes to a variation of 5 %. The improved precision is likely due to a more robust pre-amplification and qPCR attained by using universal pre-amplification primers that are efficiently removed before qPCR. After summarising the panel evaluations, it is clear that one must ensure that the signal level for all conjugates is sufficient to obtain good overall assay performance.
24-plex versus 96-plex
Finally, data from immunoassay evaluations were summarised and compared for the 24-plex versus 96-plex panels. As discussed above, we found that the average sensitivity was better with the 24-plex and the fraction of analytes detectable in plasma was higher. This is partly explained by the use of lower probe concentration in the 24-plex assays. However, when comparing these two parameters on assays present in both PEA panel types, we found no significant difference. This suggested to us that the lower success rate obtained with the 96-plex panel was most likely due to the use of less well-studied analytes of which many had unknown, or low expected levels in plasma. When addressing precision for the different panel types by determining the variation of the PE signal across an entire PEA plate (93 plasma samples) before (37 to 70 % versus 25%) and after (17 to 28 % versus 10 %) GFP normalisations. This demonstrated that the new sequence design and new pre-amp protocol significantly improved precision, and that our normalisation strategy seems potent.
LIMS
A LIMS was developed by Integromics on top of an existing code base. The LIMS is built as a client-server application where the data is stored in a central repository (server) that is accessed using a web browser (client). This approach makes it very easy to share experiments between researchers and to roll out updates of the client. The data model was designed specifically for the needs of the PROACTIVE project, incorporating not only experiment parameters but also clinical information about the samples.
A prototype was implemented based on a requirements document produced by Olink and Copenhagen University, and the design was refined for each deliverable. A development system and a production system were hosted in parallel allowing testing of features without interfering with production at the address http://proactive.integromics.com/proactive/ Visualisations useful for quality control were implemented within the LIMS. Integration with more advanced data compression, visualisation and statistical analyses provided through an export feature. Raw data can be exported from the LIMS and normalised in a desktop software application developed by Uppsala University. The application outputs normalised data in the Microsoft Excel file format which the majority of statistical software applications can work with, including free open source suites such as R. This solution was deemed to produce the maximal amount of flexibility for downstream data analyses.
Multivariate analysis
For each release of PLA data, multivariate analysis work by Uppsala University within WP2 utilised principal component analysis and hierarchical clustering and other visualisation techniques for multidimensional data to identify subgroups in the data set. Then a variety of different machine learning techniques were evaluated for design of detectors of colorectal cancer, such as discriminant analysis in combination with forward and backward feature selection methods, k-nearest neighbour classification, support vector machine classification, nearest shrunken centroid classification and random forests. The main focus was the establishment of a procedure for demonstrating presence of a signal for classification of colorectal cancer samples based on the proximity ligand assay. Of the methods evaluated, random forest classification emerged as a good choice for finding a biomarker signature that can be used to predict colorectal cancer (CRC) status in plasma samples. Random forest is an ensemble method that builds many models that each casts a vote for the predicted group label. Hence interpretation of the model structure is difficult, but the value of each potential marker in predicting the label can be estimated using variable importance measures (data not shown). Performance and variable importance were estimated using re-sampling, i.e. by running the procedure on random subsamples of the data set to assess the variability of performance and variable importance measures in small data sets. Assays important for classification can then be further studied in a larger set of samples.
Classifications
In full agreement with the project plan, the research activity within WP2 has focussed on procedures for obtaining good performance estimates for small data sets. For instance, the cross-validation procedure in common use only produces a point estimate of the expected performance of a classifier built using datasets of a given size sampled from the population. However, the variability in performance is also of interest when choosing between different designing procedures. Hence a procedure using re-sampling for estimating the upper bound of the error rate was devised. Towards that end, a novel kernel density estimator was developed to be used for generating new samples to be used in estimating the distribution of error rates. Using simulations, it was shown, that except in rare cases, choosing between design procedures using expected error rate yields the same choice as the upper bound. Therefore, focus shifted towards computing confidence interval for the expected error rate. A procedure for producing a conservative estimate of the confidence interval for the expected performance was found in the literature and was subsequently used in analysing the performance. For instance, for PLA release #2, expected classification error rate of CRC versus healthy using random forest was estimated at 29 % with the conservative 95 % confidence interval (22 %, 37 %), for PLA release #3 the corresponding (preliminary) figures are 24 % (13 %, 36 %).
Plasma samples
The sample set was selected from a larger endoscopy study where subjects undergoing a sigmoidoscopy or colonoscopy either following symptoms consistent with CRC or patients attending surveillance programs due to hereditary CRC (HNPCC and FAP) were included in a cross sectional study. A total of 5165 subjects were included and according to the Helsinki II Declaration, oral and written consent was given from each subject. The study was approved by the Regional Ethical Committee of Greater Copenhagen, Denmark (KF 01-080/03). A case-control study was designed for the present EU project by randomly selecting 78 biobanked stage I-IV CRC samples. By matching age and gender of the 78 CRC patients, 78 individuals with no pathological findings by endoscopy and/or no self reported diseases or medication, 78 individuals with adenoma, and 78 individuals with non-cancer-disease were selected from the cohort. Due to size limitations in the current technical setup the analyses were restricted to 70 samples from each of the above mentioned groups resulting in a total of 280 samples.
Biomarker screening
For this screening, the basic proximity probing assay was shifted from PLA to PEA (described above). The 24-plex PLA assays (Lundberg et al., 2011) was applied as a model to transfer the PLA approach to the PEA. Screening #3 was divided into four panels in 24-plex and a single panel of 96-plex. The biomarkers in the 24-plex are identical to PLA release two and the 96-plex is composed of novel biomarkers and a few repeats from the 24-plex panels. For this screening all four groups (CRC, healthy, adenoma and non-cancer disease) were included.
In close collaboration with Olink, new putative biomarkers were selected for Biomarker screening #3. For the new PEA 96-plex, the biomarkers were carefully selected by thorough literature searches', including criteria's such as proteins present in pathways involved in CRC, proteins involved in inflammation and cancer, proteins found in screenings looking into the tumour tissue, and general cancer markers were included. Another important criterion was again the availability of appropriate antibodies. Based on the recommendations from the reviewers it was decided to focus on novel low abundance proteins, as the proximity assays possess a high sensitivity. To achieve this, we included markers with reported low plasma levels and several markers based on their increased expression in CRC tissue, with no available data on plasma levels.
Correlations between ELISA, PLA and PEA
At KU, six proteins have been quantified by ELISA in samples from healthy individuals and CRC patients. Correlations between data from PLA release #2 and PEA release #1 (WP3) and data from the different ELISAs has been made for these. Generally, the Spearman correlation (R) between PLA release#2 and PEA#3 was good. For all the analysed assays, except for one, we found that switching from PLA to PEA improved the Spearman correlation (R) to the ELISA data.
Identification of CRC-specific biomarkers
We focused on CRC specific markers by comparing CRC to each of the three other groups (healthy, non-cancer disease and adenoma). Applying these criteria we identified 35 assays which display specific changes only in CRC (data not shown). Altogether, we have identified 35 assays out of 150 individual assays (23 %) that specifically identify CRC from the other groups in the case-control study. Among the 96-plex assays, which comprised 86 assays, we identified six novel CRC specific assays. These markers are very promising for further investigation since they were identified among potential marker of low plasma abundance and tumour tissue biomarkers.
Potential impact:
The results generated within the PROACTIVE project have been presented in a several large scientific conferences, such as HUPO (Sidney/Geneva 2010/2011) and AACR (Philadelphia 2010), and published in two peer-reviewed journals (M. Lundberg et al., Nucleic Acids Res, 2011 and M. Lundberg et al., Mol Cell Proteomics, 2011).
The PROACTIVE project has enabled Olink to develop our proximity probing technologies both in terms of assay performance and at the multiplexing level. A plasma biomarker service based on the biomarker panels generated within the project have been released and resulted in a large interest amongst existing and new customers, generated significant revenue, as well as important collaborations for Olink. We are now continuing our work on the 96-plex protein detection technology and plan to develop and commercialise four panels within the coming year. This will make high-throughput biomarker screenings available for academic laboratories as well as biotech companies, and hopefully lead to the discovery of many new interesting biomarkers and/or biomarker panels.
The chosen strategy of developing panels of well selected and qualified analytes has several implications. It will be more indicative of the status of a disease than just analysing with any single marker, and help to reduce sample costs efficiently due to the high throughput capacity of the PLA platforms. Such a powerful tool will contribute to improve health care but also to significantly increase competitiveness and commercial success of the participating SMEs as well as to the PROACTIVE seven scientific recognitions for the academic partners. The intellectual property will be commercialised by the consortium participants as proprietary products and by out-licensing agreements.
CRC is the second most frequent cancer type second only to breast cancer. It is proven that colorectal cancer screening reduces disease-related mortality and is cost-effective. However, current compliance is poor. The potential use of the PROACTIVE high throughput technology enables biomarker screening to define a set of protein markers that potentially will identify individuals who should undergo colonoscopy by placing them in a certain risk group. CRC was chosen as the tumour type because of its enormous social and economic impact. In addition, the knowledge obtained from our studies in patients with CRC may be translated into the adjuvant treatment and perhaps into other cancer diseases and thereby be of benefit for a significant additional number of cancer patients. At the conclusion of this collaborative project, the research and innovation conducted within the consortium will enable the partners to position themselves at the very forefront of high-throughput biomarker research strengthening their competitiveness in the international arena.
Project website: http://www.olink.com/proactive
Simon Fredriksson
simon.fredriksson@olink.com via e-mail
Telephone: +46-184-443984
