Final Report Summary - HIPERDART (Development of High Performance Diagnostic Array Replication Technology)
Executive summary:
The aim of the HIPERDART project is to develop a higher standard clinical microarray technology platform, by proposing a highly innovative probe printing technology, called Supramolecular NanoStamping (SuNS). This microarray technology will be tested with a prognostic kit useful to predict recurrence in colorectal cancer.
The new microarray technology will decrease production time and costs, increase assay sensitivity, decrease hybridization time and move this technology to the workbench of pathology departments to easy personalized medicine. To fulfill these aims, the project has optimized four technological areas:
1) Rolling Cycle Amplification (RCA) technique is used to multiply the DNA probe specific regions. While typical microarrays use oligonucleotides as probes spotted or grown on the array surface, our proposal is to deposit a long sequence of probe DNA that has been repeatedly amplified by RCA, thus increasing sensitivity to detect sample target cDNA molecules.
2) Solid surface is covered with a brush polymer that keeps the long RCA probes attached to the glass and allows replication. Aldehyde functionalized brush was identified as the best solution for this coating.
3) Production time and cost of microarrays is reduced by using a duplication process by which a template DNA array is printed. This template is duplicated by polymerases and this second strand is transferred to a replica slide. The template can be reused for producing multiple replica arrays. A robotic printer simplifies array production and reduces human intervention errors.
4) A disposable cartridge hybridization chamber has been designed that uses microfluidic technology to minimize intervention and reduce processing time and cost.
Project Context and Objectives:
HIPERDART Targets and Objectives
In applications related to clinical diagnostics, DNA microarrays have enormous potential for multiplex analysis, yet issues related to technical challenges (low reproducibility, poor signal-to-noise, and lengthy workflows) have limited their practicality and slowed regulatory approval.
The aim of the HIPERDART project is to develop a higher standard clinical microarray technology platform, by proposing a highly innovative probe printing technology, called Supramolecular NanoStamping (SuNS). This microarray technology will be tested with a prognostic kit useful to predict recurrence in colorectal cancer.
We believe that the SuNS technology offers significant advantages in microarray manufacturing, such that these challenges may be overcome. By making a large up-front investment in the template and then replicating it many times, we can effectively dilute our production costs while maintaining high performance features in the array.
SuNS technology consists in fabricating a template array with single strands of DNA that self-replicate on a surface to form complementary DNA strands, which can be transferred to other replica surface. With this procedure, the template can be reused multiple times to decrease microarray fabrication time.
From a manufacturing standpoint, the project has addressed four fundamental features of microarrays, which have so far limited their fabrication and utility: probe density, surface coating, template replication infrastructure and hybridization device.
Probe Density: Signal-to-noise and assay sensitivity, key features of microarray utility, are both determined by probe density within a feature. The higher the target-capture capacity of the feature per unit area is, the better the performance of the microarray. There is a fundamental limit, however, to the oligo density, which can be achieved on a surface, due to the radii of gyration of the DNA strands. The approach here proposed is to maximize probe density by concatamerizing oligos in the 'z-direction' with respect to the array surface. The HIPERDART consortium envisions accomplishing this by utilizing rolling-circle amplification (RCA) technology. Extension from a common primer sequence will take place within the micro-wells, each with a different circularized DNA molecule.
Utilizing the common primer sequence, the resulting long-mer strands are themselves enzymatically replicated prior to being printed onto a replica surface. By increasing probe density using RCA technology, we can dramatically increase signal-to-noise and assay sensitivity to an unprecedented degree.
HydroSuNS: surface coating by hydrogel polymes. A compelling feature of the SuNS technology is the flexibility of substrate material onto which DNA molecules may be printed. A good substrate shall have properties such that it simultaneously (a) provides ideal conditions for SuNS printing, and (b) optimizes microarray assay performance. Scientists in our consortium recently have discovered that indeed such a surface exists, and that it is the surface of a hydrogel polymer.
There are essentially two technical challenges to consider with any SuNS approach. The first challenge is to achieve nanometric conformal contact between two surfaces over a macroscopic area, while the second is to minimize damage to the template DNA, which may result from repeated cycles of surface-to-surface contact. Molecular Stamping has overcome these two challenges of SuNS by printing onto the surface of a hydrogel. The project has explored multiple polymerization coatings to find the optimal chemistry for SuNS.
Template replication infrastructure. A device for slide replication using SuNS technology needed to be built and its quality standards determined for industrial production of arrays. The project has invested in such technology, introducing robotics to decrease human intervention in an attempt to attain maximum reproducibility and low error rate.
Array Hybridization device. Conventional microscope slides (75 x 25 mm2) are most commonly used as microarray substrates. Within the HIPERDART project, a disposable microarray hybridization cartridge and device that implements microfluidics has been developed which enables very low sample volumes, while simultaneously reducing the time required for hybridization and improving the reproducibility of assay results. These requirements are essential for clinical applications, in which nucleic acids derive from small biopsies. One key element of the design is that up to four independent samples can be hybridized simultaneously, and each chamber incorporates one-way channels so that its contents are guaranteed to remain isolated from the others during the assay implementation.
Prognostic prediction for personalized medicine in colorectal cancer. Colorectal cancer is the third most common cancer worldwide after lung and breast. Cumulative risk in European countries is near 6%, both in men and women. Five year survival estimates range from 90% in Stage I to less than 5% in stage IV, and is less accurate (45-80%) in stages II & III. Adjuvant chemotherapy is standard for stage III but not for stage II, where the challenge is to identify the 25% of patients not cured by surgery alone. Clinical and pathological risk factors (T4, G3, number of assessed lymph nodes, perforation or vascular invasion) have been identified but lack standardisation.
Molecular prognostic biomarkers that allow identifying patients at high-risk for relapse have been successful in breast cancer and within this project we aim to identify equivalent biomarkers for colorectal cancer, prove their utility for predicting recurrence and implement the profile in the SuNS microarray as a clinical kit. This will allow the pathologist and oncologist identify patients that require adjuvant therapy and possibly to spare it in the patients with low risk stage III.
Despite the large amount of literature on molecular biomarker candidates, none is routinely used in the clinic due to lack of proper standardisation and validation. In this project we aim to develop a prognostic predictor that has clinical usefulness in the management of colorectal cancer patients. We will use public available data on genomic expression analyzed with microarrays and our own data generated for a series of patients well characterized clinically and that have been followed up for more than 5 years. The information of genomic expression in tumors will be complemented with additional sources of data: genomic expression and germline polymorphisms in histologically normal colonic mucosa resected from patients with colorectal cancer and copy number variation. Also, novel genomic and epigenomic (methylation and miRNA) data is being analyzed in our series and will be used to identify the best prognostic biomarker candidates. These data will be measured in a limited sample, but large enough to ensure power to detect the most relevant signals. Genomic expression of normal mucosa and inherited genomic variation has been shown to have relevant prognostic value prior to the development of the tumor. In this study we aim to use a whole-genome approach, together with appropriate data mining techniques, to select the most informative prognostic markers.
Bioinformatics expertise from two of our partners (FBK and Crosslinks) will help with the data-mining process of selecting reliable and robust predictive models, which are essential to realize the promises of personalized medicine. Valid biomarkers derived from high-throughput technologies need to undergo a quality control process that can address the control of variability along all steps of the processing pipeline. In the HIPERDART project, we will evaluate and control variability of the new technology by adopting BioDCV, a state-of-the-art microarray profiling platform that implements a complete validation set and a rich family of machine learning algorithms for feature ranking. The platform can develop predictive signatures on genome-wide chips, and it has been recently extended with a set of algebraic indicators of stability for ranked gene lists. We will compare signatures produced with the HIPERDART or alternative platforms, and within the same platform, for different technical implementation of the upstream data production and preprocessing phases. Also novel network-based algorithms will be developed to identify the best candidates.
Project Results:
The project main results will be reported for these main areas:
1. Template Production
a.Probe design and RCA
b.Surface coating
c.Stamping
2. Array production and hybridization
a. Stamping device
b. Disposable hybridization cartridge device
3.Application to prognosis prediction in CRC
a.Identification of profile
b.Validation of profile
1.Template Production for SuNS microarray
The following steps are required for template production. The DNA products resulting of these steps are deposited on a glass surface appropriately coated before stamping and array replication.
i. Design, synthesis and analysis of the model system probe
ii. Circle synthesis and analysis of the ligation products
iii. Rolling circle amplification
iv. Second strand synthesis
a. Probe design and RCA.
The first step was to produce a single sequence DNA array. A model oligonucleotide was spotted at different concentrations on the array and analyzed by hybridization with a fluorescent-labeled oligonucleotide. The difference in color indicates different concentrations of the spotted oligonucleotide. This array has been used for experiments with rolling circle amplification (RCA) on the array. The bound spotted oligonucleotide functions as a primer for the RCA reaction. The oligonucleotide has also been used in connection with experiments performed in solution.
As model system an 80-mer region of the pUC19 vector was used. Both a synthetic and a PCR generated probe were analyzed by Pyrosequencing and the results are shown below. Sequencing was performed on the 5'-ends. The results show that the synthetic oligonucleotide contained frayed ends whereas the PCR fragment gave much purer signals indicating a much purer product.
Circle production and analysis of the ligation products. For circulization of the probe several different ligases were tested; both Circligase which utilize single-stranded DNA (ssDNA) and ligases that need a splint (double-stranded DNA ligases). For analysis of the ligation products a new assay was developed. In the new assay the ligation region was analyzed by sequencing a short stretch over the ligation site (a few bases before and a few bases after the ligation site were sequenced by Pyrosequencing). By the new assay a quantitative estimate of the efficiency of the ligation reaction could be obtained. The result was compared with traditional gel analysis.
The main difference between the procedures is the use of a blocking oligonucleotide to bind the splint DNA before the analysis. The blocking oligonucleotide will bind to the splint and thereby hinder it from disturbing the detection primer to bind to the template as well as hinder it from producing signals during the sequencing procedure.
From the results obtained from the different experiments performed with the different ligases and under different conditions it was concluded that the best result was obtained with Cirligase and ssDNA. The ligation can be performed without a splint and at high temperature (60 degree Celsius). Faster ligation was obtained with fewer problems with secondary structures. The method was cost effective as low amounts of enzyme can be used if lower ligation efficiency is accepted.
Rolling circle amplification (RCA). RCA reaction was set up and optimized in solution. The following points were of importance for the success of this step:
-Product detection
-Quantification/length of the product
-Quality
-Optimization of polymerase, primer-to-circle ratio, sequence design, purity, additives, temperature, time and cost
To be able to analyze the product from the RCA reaction a new assay was set up. The standard gel analysis method is hampered by the low resolution of large DNA fragments. The new assay is based on binding of primers to each of the amplified circle fragments. The product is then analyzed by Pyrosequencing. The observed signals are compared with an internal pyrophosphate standard. By this comparison and by analysis of the initial amount of circle used an estimate of the amplification efficiency can be obtained.
Rolling circle amplification on slides. After the RCA reaction was set up and optimized to work well in solution the procedure was evaluated on slides. Parameters such as time, additives and primer density were evaluated. The first experiments were performed to find good conditions and to get pure fluorescence signals. A protocol containing 1% BSA was found to give clear signals. Although pure signals were obtained no amplification was detected. Although up to 6 hours reaction time was tested with different concentrations of enzyme, nucleotides, primers and different additives no amplification was obtained. There was no reason to believe that the hybridization and detection steps wouldn't work as with similar conditions the reaction had worked well in solution. As many different additives were tested solubility and secondary structures should not be a problem. In a couple of experiments the effect of the primer density on the chip was analyzed. We found that the primer density was an important factor. At the high density normally used no amplification was obtained whereas at lower concentrations the amplification reaction worked. Although when the concentration was decreased the RCA reaction must be very efficient to obtain signal levels necessary for an efficient detection system. At 200 times lower primer concentration a RCA amplification of 50 fold was obtained. This amplification is not enough for our aims and after many fruitless tries to improve the amplification we decided for a contingency plan.
Contingency plan/Spotting of prototype. Due to the problem to obtain the desired signal intensity utilizing RCA directly on the arrays we decided to go for a contingency plan. The plan is based on the good results obtained for the RCA reaction in solution. By utilizing RCA in solution and then spot the amplified DNA on the array we could overcome the drawback with the inefficient RCA directly on the array. To speed up the project we decided to direct start the spotting with the selected probes. Factors to consider for this part of the project are optimization of concentrations, solubility, additives and time.
This method was used to produce RCA products for the 10 selected probes from WP5, that should be used for clinical application (see table).
The oligonucleotides have the complementary sequence to the original probe sequences. Underlined bases: restriction site. Extra bases (T)3' improve ligation. 5' phosphate. The 20 bases at 5' are used for detection and RCA reaction.
The designed circles for the 10 probes were amplified by the RCA reaction. The results show that all probes were amplified but that the efficiency differed between the different probes. The ligation efficiency did also differ for the different probes. The amplified DNA was used for spotting on arrays.
b. Optimal surface coating for RCA product stamping.
This area of work was responsibility of EPFL. The objective was to develop a method to modify replica surface with a thin hydrogel coating that allows a fast and reproducible nanostamping process. To this end, polymer brushes which are thin (20-200 nm) coatings consisting of densely packed polymer chains that are tethered with one of the chain ends to the substrate were proposed as good candidates to modify the replica surface. Since these polymer brush coatings contain reactive groups that would improve the efficiency in nanostamping process due to their potentially very high surface concentration of functional groups and an internal volume. In contrast to glass slides modified with a monolayer of functional compounds, these polymer brushes can be considered as 3D substrates.
Polymer brushes prepared specifically via the grafting from strategy in which the polymerization is directly initiated by the initiator functionalized surfaces; allows an accurate control over the brush thickness, composition and architecture particularly by controlled radical polymerization techniques. Using appropriate side chain functional monomers, surface-initiated polymerization leads to the polymer brushes that can present very high surface concentrations of functional groups to capture end-functionalized bio-molecules. Moreover, due to the relatively high thicknesses (20200 nm) compared to monolayer of functional groups, these polymer brushes can also possess an internal volume that can be used to capture and bind end-functionalized bio-molecules.
In order to develop a thin polymer brush coating for the modification of the replica surface, first approach involved the direct modification of the glass substrate with polymer brushes containing active ester groups to allow immobilization of end-functionalized bio-molecules. Successful synthesis and detailed characterization of these polymer brushes were demonstrated. Thereafter, these replica surfaces were investigated for their feasibility to bind end-functionalized bio-molecules; however the results were not satisfying to continue the studies with these active ester functionalized polymer brush coatings.
The poor results on the replica surface relative to the active ester chemistry had to be overcome by introducing an alternative surface chemistry onto the replica surface; which involves functional groups that show higher stability under nanostamping conditions and allow the immobilization of end-functionalized bio-molecules. To this end, the effort was focused to successfully introduce the alternative surface chemistry onto the replica surface; which provides aforementioned properties; and to optimize the conditions for the usage of this alternative surface chemistry to allow efficient covalent binding of end-functionalized bio-molecules.
The modification of replica surface with this alternative surface chemistry was investigated in detail and fully characterized. Thereafter, these replica surfaces containing alternative surface chemistry were investigated in terms of their feasibility to covalently bind end-functionalized bio-molecules. The good outcomes led to continue with transferring this alternative surface chemistry to PDMS polymer layer as a second approach in order to enhance the mechanical properties of the hydrogel coating on the replica surface.
Modification of PDMS polymer layer with the alternative surface chemistry was successfully demonstrated. Modified PDMS polymer layer with the alternative surface chemistry and particularly the presence of functional groups to bind end-functionalized bio-molecules on PDMS polymer layer were fully characterized by different techniques.
Surfaces prepared by direct modification of glass slides or by modification of PDMS polymer layer were compared in terms of their performances as replica surface. The outcomes of this study revealed that modified PDMS polymer layer containing functional groups was a good candidate to prepare replica surface with enhanced mechanical properties for nanostamping process.
The proposed strategy to modify the replica surface with a thin hydrogel coating based on polymer brushes may serve as an efficient substrate for the nanostamping process and can fulfill the requirements that are crucial to improve the efficiency in the process. This study revealed that major objectives have been achieved in development of chemistry-adapted polymer brushes to modify replica surfaces. Therefore, enhancements in both array replication and replica surface modification as a joint effort of HIPERDART collaborators may lead to improve the microarray diagnostic capability and performances; which would have a potential positive impact in fast and reliable clinical diagnostics.
c. Stamping of replicated RCA on EPFL surface.
KTH has produced more than 80 stamping experiments of RCA products using both EPFL made and MS made replica surfaces. All MS surfaces consist of PDMS layers on glass, while all of EPFL surfaces (except one) are PHEMA-Aldehyde polymer brush on glass. The different one is a PHEMA-Aldehyde polymer brush on glass/PDMS (As from a specific deliverable in WP3, D3.2). Templates used for stamping were replicated either by MS or Pal Nyren group at KTH.
The most promising result collected so far has been stamping of a KTH replicated template (Hi16 Protocol) on the EPFL PHEMA-Aldehyde polymer brush on glass/PDMS. It is worth it to underline that this stamp has been done with HIPERDART outcomes only: 10 probes selected by ICO, synthesized by KTH, spotted by MS, replicated by KTH, and stamped on a replica surface made by EPFL.
For experiments performed on 3 HIPERDART probes no DNA transfer occurred for the first set of trials except for 793-18 where some spots are visible. Another set of replica surfaces, partly produced by MS and partly from EPFL, was used for stamping of templates replicated by the Pål Nyrén group at KTH.
A first set of stamping was performed in the conditions described below. Array replication has been performed by KTH. Only MS800-05 gave detectable signals, although background signal is high.
The new 10-probes prototype template was used for stamping on different surfaces.
A last experiment has been performed using a new epoxide-coated template surface developed at MS, replicated by KTH and stamped on three different replica surfaces: MS PDMS-APTES on glass, EPFL PHEMA-Aldehyde on glass and EPFL o PHEMA-Aldehyde on glass/PDMS.
Array replication and stamping. On the spotted array second strand synthesis (replication) and stamping were performed.
Conclusions of template production. Since TB423_H10 is the most promising result overall, in the next future we should focus the stamping experiments on EPFL PHEMA-Aldehyde on glass/PDMS replica surfaces, using already available Aldehyde-coated templates to be replicated by KTH with protocol Hi-16. Stamping on EPFL PHEMA-Aldehyde on glass replica surfaces will not be discontinued but considered at lower priority. We do consider that major achievements have been reached on both array replication and replica surface development. Indeed, although we do not have statistical relevant numbers yet, the proof-of-concept of Second Strand Synthesis on the array, and the development of chemistry-adapted polymer brushes surfaces has been obtained.
2. Array production and hybridization
The template array is the basis for replica array production that can be used to hybridize nucleic acid samples. The HIPERDART project has an engineering section for designing and QC the stamping device, contributed by Molecular Stamping, and for designing and producing a disposable cartridge to ease the hybridization.
a. Stamping device.
The design of the stamping device was completed on schedule cooperating with Advance Automation engineers. Thanks to the excellent results of the cooperation, we have been able to anticipate the Factory Acceptance Test (FAT performed in Advance Automation facility in USA), the shipment to Italy, the equipment installation and the final Site Acceptance Test (SAT).
The equipment was installed in the facilities of Molecular Stamping (Trento, Italy) and Site acceptance test was conducted together with Advance Automation Engineers according to their AA standard procedure. Following the formal SAT test, the equipment was extensively checked, in order to verify its performances and to identify the best set of process parameters that could optimize the overall SuNS replica process.
Several automation hardware was designed and realized to improve the original model and decrease human intervention in the replica array fabrication process:
- Barcode Reader Integration
- Double Microarray Stamping
- Buffer Bottles Auto-Switch
- Tool for Slide Alignment before loading
- Tool for master-replica separation after stamping
- Process Data transferring for remote Monitoring
- Robotic load/download arm
Barcode reader Integration. For an automatic registration of the barcode of replicas and templates we have developed hardware and software interface for barcode reader. We added a hardware and software support to the Stamping Machine for the barcode reader; we here show both the Hardware and Software Interface:
Double Microarray Stamping Chuck. To improve the manufacturing capacity we have developed a Double Chuck; this allows to perform two DNA microarray stamps at once. Two templates are loaded on the upper chuck and two replica surfaces are stamped on the bottom chuck. A dedicated load unload tool has been designed and produced, to load/unload two slides at the same time.
Buffer Bottles Auto-Switch. A dedicated Switching Buffer Tool has been designed and realized; this tool allows to switch between six different buffer bottles automatically. Along with this, a remote control of the switching can be performed with a Software Interface.
Tool for Slide Alignment before Loading. For faster loading and to minimize any alignment operator error a tool has been designed and developed. This is composed by two parts: Picker Vacuum Tool that catches the slides by using vacuum, and Alignment Station, that trough a Washer and a Register Plate aligns the slide with the Picker Vacuum tool.
Tool for Master-Replica Separation after Stamping. For faster template-replica sandwich separation after stamp, and to avoid to brake any of these surfaces, we have designed and developed a dedicated tool and. This is composed by two parts.
- Picker Vacuum Tool that catches the sandwich by using vacuum
- Separation Station to divide the sandwich applying the vacuum on the master and replica and moving up the picker vacuum tool trough four pins
Process Data transferring for remote Monitoring. A software interface has been developed to allow the automatic data upload from the equipment to the controlling pc. Utilizing the features, giving the pc connected on line, all the process data can be monitored remotely.
Robotic load/download arm. An Industrial Manipulator is going to be designed in the future to complete the automatic loading and unloading of the slides. A possible implementation could be represented by the COMAU SMART NS 12-1.85 considered suitable for our application.
The activities performed in the last 12 months verified all mechanical and electrical compatibility with stamping equipment and microarray glass slides.
We double-checked that using this commercial manipulator we could get the best cost benefit ratio. The selected robot matches perfectly our need of low cost, dimension, operations (degree of freedom for each mechanical joint) and compatibility.
-Selected robot characteristics are:
-Height 1485 mm
-Width 506 mm
-Arm Max extension 1651 mm
-Base Rotation angle ± 180°
-Operating area and dimension allow a perfect match with stamping equipment
Robot coupled with Stamping Equipment
-Stamping equipment has been modeled by using 'SolidWork 3D' software
-SolidWork 3D was used to introduce the handling arm to validate its compatibility with the stamping equipment
-A complete 3D model has been prepared
-The manipulator COMAU SMART NS 12-1.85 can be easily combined with stamping equipment picker vacuum tool. In such a way, the load/unload operation will be completely automated, providing the slide holder is fixed in the robot operating range and utilizing the X,Y,Z movement and the ability to rotate clockwise and anti-clockwise.
Robot arm coupled with picker vacuum tool
By using this commercial solution, the following operations improvement will be achieved:
-Optimum slide alignment;
-Shorter slide load/download time;
-Softer slide handling and as consequence, minimization of slide breakage;
-Improved equipment and process troubleshooting by eliminating systemic errors caused by human
Quality Control of stamping device
A significant number of slides was ran through the machine to assess its stability and to confirm process repeatability. Results have been very good, even comparing the AA2 equipment performances with the original prototype AA1. The effects of the following process variables have been investigated with the instrument.
The QC thresholds and the related procedures to be used in production environment have been completely identified and tested by analyzing and summarizing all the measurement performed on each single slide. The best performing set of parameters have been selected.
The four main stamping variables can be considered:
I. Replica Sublayer (Thickness and Replica Signal Intensity)
II. Fluorophore type
III. Stamping Temperature
IV. Mechanical Parameters
I. Replica Sublayer Surface/Thickness
Polydimethylsiloxane (PDMS) was selected as best performing layer for stamping surfaces. Aminopropyltrimethoxysilane (APTES) has been used to functionalize spin-coated freshly oxidized PDMS layers. Both mentioned sublayers have been chosen according to our experiments. They benefit by a shorter production time, safer involved chemicals and the following better performances:
-Higher Homogeneity
-Higher peak of Transfer Level up to 40 %
-greater than 95 % spot with SBR greater than 10
The optimal thickness has been found to be ranging from 200 to 500 nm; Profilometer readings are used as standard QC of the coated polymer layers.
Replica's signal intensity. The stamping outputs have been used to define the QC limits of replicas; an internal parameter, ie Process Yield (PY), has been defined, being PY=1 an intensity of 2000 at 90/90 as Laser Power/PMT Gain settings. These QC thresholds can be summarized as follows:
-Intensity: Higher than PY = 5
-Intra-slide Uniformity %: Lower than 40 %
-% of spots with SB greater than 10: At least 97 %
II. Fluorophore
In order to improve the stability of the signal, we tested ATTO 647 N fluorophore as new dye and we compared it with ALEXA 647 that was our reference point during the first project period.
-ATTO 647 N templates show double stability of the signal
-ATTO 647 N replicas show Higher stability of the signal and 1/3 background intensity
III. Stamping Temperature(s)
Different stamping temperature tests have been done in the last months.
We performed tests at 30, 45, 60 and 80 °C, hybridizing with both primer and with gDNA sequences.
The result is that 80°C is the best stamping temperature; this temperature leads to
-% of spots with SBR greater than 10 near 100%
-High transfer level
Ratio of average Signal-Background on the replica after hybridization over average Signal-Background on the related template. The unloading temperature was set to 48°C because higher temperatures could remove both strands of DNA from the template surface.
IV. Mechanical Parameters
The following set of Mechanical Parameters lead to the best stamping outputs when using Super Slow Approach and has been chosen as standard operating procedure for HIPERDART slide production:
-Soak position 60 sec
-2 position 60 sec
-3 position 60 sec
-4 position 60 sec
-5 position 60 sec
-Float position 0 sec
-Contact position 10 sec
-Full force 85kg
Quality Control procedures
Two standard Operating Procedures have been issued to formalize the QC step.
First procedure, named OPN-HYB, is applicable to QC slide Hybridization.
Second procedure, named OPN-STQC, is applicable to actual slide Quality Control step.
In conclusion, stamping equipment has been successfully designed, manufactured and installed. Stamping process has been developed and optimized for each single process variable. QC procedures and threshold values have been set and validated. Automated equipment for production has been designed and tested.
b. Development of the disposable hybridization cartridge
Within this task, a novel hybridization cartridge was developed. The new hybridization chamber targets small sample volumes. The goal is to reduce both the reaction time and the sample consumption to improve hybridization efficiency. Microfluidics offers a prerequisite for reduced sample volumes. Diffusion distances are greatly reduced in microfluidic systems and therefore reaction times are reduced in diffusion limited systems. Also, microfluidic system offer excellent controllability of fluids. Networks of microfluidic channels can be constructed to allow to process different samples at the same time while eliminating any crosstalk and therefore greatly simplify the experimental protocol for multiple experiments. While the experimental set-up needed to run the cartridge will allow fast and reproducible hybridization experiments. The cartridge is intended to be a disposable.
Prototyping can be done by milling or laser ablation of polymer substrates. However, the surface properties (e.g. surface roughness) are depending on the applied manufacturing technique. Therefore it is important to make tests in a very early project stage using structures manufactured using the method which will also be applied later on for series production. In most cases this will be micro injection molding.
Within the project Development of the disposable hybridization cartridge, the following deliverables have been addressed:
-Report on the specifications of the hybridization set-up and concept of cartridge and set-up
-First generation of cartridge and test set-up
-Second generation of cartridge
The collection of specifications and the cartridge concept have been reported. Three different concepts have been presented:
1)Poylmer substrate & disposable assembly
2)Glass slide substrate & disposable housing
3)Glass slide substrate & disposable housing with integrated fluid control
The first generation of the hybridization cartridge and experimental test setup has been realized according to concept 3. The design of the cartridge incorporates beside the hybridization chamber the fluidic storage and waste deposition on chip as well as the fluidic transport and fluidic management by means of a pump and a multitude of valves. Since volumes of the washing solutions from the original hybridization protocol are much too large for on chip storage and waste deposition, the approach for the cartridge design was to minimize that volume by pumping the washing solutions back and forth to achieve adequate results. Five washing solutions with a total volume of 20 ml (4ml each) can be operated with the cartridge (original protocol requires 70 ml). Therefore special fluidic management was applied to handle the waste volume with using the volume of the storage containers of the washing solutions for the waste once they are empty.
The first generation cartridge was milled from polycarbonate material. A pressure sensitive adhesive tape was used for sealing the fluidic channels. A double sided pressure sensitive tape was used to apply the glass slide with the microarray to the hybridization chamber of the cartridge. The pump membrane and the valves have been made from elastomeric material. The pneumatic actuated pump and valve membranes are clamped to the cartridge with a pneumatic manifold part, which was also milled from polycarbonate.
The experimental test setup includes membrane pumps for generating the pneumatic pressure and vacuum, electromagnetic pneumatic valves for switching the pump and valve membranes on the cartridge, PCB boards with driver electronics for the electromagnetic valves and temperature control and two data acquisition boards with USB interfaces to the PC.
First tests with spotted microarray glass slides show that the double sided adhesive tape for the sealing and assembly of the spotted microarray slide was not performing under heat (slide was not removable without damage after hybridization). Within tests with an alternate sealing method, the hybridization of the microarray was successful. Analysis shows that the hybridization signals are very good, but signal uniformity is subject for further improvement (poor wash efficiency).
For the design of the second generation hybridization cartridge the following main objectives have been targeted:
-Hybridization of four independent microarrays
-Injection molded disposable cartridge
-Fluid storage (wash solutions and waste) off chip
-Flow control on chip as well as off chip. Pump off chip to provide more flow rate flexibility for wash sequence optimization.
-Optimization of glass slide handling (attachment/removal).
For the second generation of the cartridge, a concept for four independent hybridization chambers and sample lines has been elaborated according to the project proposal and an equivalent design for injection molded components has been made. From the design of the hybridization chamber (50 μm depth), a sample volume of 25 μl will be required for each sample line, which is less than half of the volume compared to other hybridization devices.
The cartridge will be delivered as an assembly of the card with a laser welded cover film (both made from polycarbonate) on the back and a thermoplastic elastomeric sealing for the microarrays. For the sealing of the sample ports, additional plugs from thermoplastic elastomeric material are supplied. In a first step, the microarray glass slide is clicked on the cartridge with an integrated hook. The microarrays are finally sealed by application of force towards the seal in the instrument. In a second step, the sample volumes are applied on the sample ports with a pipette tip. The samples are now loaded in the cartridge sample containers and agitation containers. After sealing the sample ports with the plugs, the cartridge is ready to be loaded into the instrument.
The instrument itself provides beside the temperature control a set of parallel driven plunger actuators to manage the hybridization protocol as well as fluidic interfaces for the waste and wash fluids. The wash fluids are driven with external pumps.
The cartridges have been injection molded from black polycarbonate. The sample port plugs have been injection molded from a thermoplastic elastomeric material. The chamber sealing parts were laser cut from an elastomeric material.
A breadboard was built up for disposable cartridge testing. The user interface was implemented on a Netbook size computer. Communication between computer and breadboard was implemented over USB connection.
The breadboard and a set of cartridges (R2 card) have been tested. The temperature stability of the system was given and hybridization was successfully shown on a 4 fold microarray. Nevertheless, optimization steps have been necessary for the cartridge and the breadboard. For the cartridge, the sample filling of the sample container was not uniform. A second set of cartridges (R3 card) have been assembled with post machined flow stops structures in the sample containers.
Further hybridization tests show that the sample loading improved. DNA hybridization (spike-in oligonucleotide) results showed a big improvement of the hybridization signal inside the same slide, both in terms of slide uniformity and fluorescence signal. Molecular Stamping reports that overall results for the cartridge testing are in general very good.
3. Application to prognosis prediction in colorectal cancer
This last area of the project aimed to apply the developed technology to a specific clinical problem, aiming to design a clinical prognostic tool to help personalize colorectal cancer (CRC) therapy according to a prognostic prediction. The tasks in this section are
a.Identification of an expression profile useful to predict CRC prognosis and provide a set of targets for the SuNS array fabrication
b.Technical validation of the SuNS microarray
c.Clinical validation of the profile
These tasks have been jointly performed by ICO as clinical group and FBK and Crosslinks as bioinformatics groups.
a.Development of a series of predictors for clinical outcomes and selection of probes for the prototype SuNS microarray
i) Build the predictors from existing data
The HIPERDART WP5 goal is to develop a prognostic predictor for colorectal cancer patients. In the first phase, we analyzed genomic expression data from both international archives (GEO datasets 2-6, and ArrayExpress dataset 1) and the MECC unpublished data (dataset 7) provided by ICO to the WP5 partners to explore gene panels to identify a short list of clinical candidate biomarkers. A total of 1482 expression arrays from 9 studies has been collected.
The data were used both as originally provided on the Web and after normalization and quality control procedures performed within the WP5 activities (CrossLink). Many diagnostic and prognostic outcomes were taken into account according to the available clinical information (which was not the same for all datasets): Dukes Stage A vs D , B vs C, OS and EFS status 0 and follow-up time >1000 vs status 1 (on the B+C, on MSS, on B+C and MSS), and other clinical phenotypes. Moreover, the choice of the binary classification problem addressed for each dataset depended on the number of samples available for each class (too unbalanced tasks were avoided). As a result, the following 10 genes were selected: ITM2A, COL5A1, LIF, KLF4, VDR, CHD2, CALR, NR5A2, SEC62, RPS5.
With this proposed signature (leave one out cross validation) we obtain the following values: Accuracy: 0.75 Sensitivity: 0.35 , Specificity: 0.96 MCC: 0.41 AUC: 0.78
Our conclusions are that these genes aren't very useful in practice for prediction because when cross validation is applied, the accuracy decreases fast. So they aren't very good for our purposes.
ii)Detailed analysis of the published prognosis signatures in CRC.
The comprehensive study of published prognostic signatures on CRC was published in the open access journal PLoS ONE: Sanz-Pamplona R, et al. Clinical Value of Prognosis Gene Expression Signatures in Colorectal Cancer: A Systematic Review. PLoS ONE 2012; 7(11): e48877. doi:10.1371/journal.pone.0048877. This paper had been written jointly by partners from ICO and FBK. The paper underlines the difficulties in finding good prognosis predictors in CRC. The conclusion of the study was that, though some of the 33 signatures studied show significant association with prognosis, the predictive ability is limited and many of the published signatures only perform well in the datasets used to develop them. This is a typical problem of over-fitting and limited validation, which has been recognized in many signatures . Anyway, the clinical utility of some expression profiles analyzed was reasonable, with average 3 yr separation of survival curves around 20%, which may be useful in practice.
iii)Work to identify better prognostic candidates for our validation in the SuNS microarray.
In this line of work we have pursued five parallel analyses of our data aiming to identify a set of interesting candidates for extensive validation in clinical samples. The analyses have used data generated within the ICO COLONOMICS series (see http://www.colonomics.org online), that has analyzed 100 tumors, 100 paired normal mucosa from stage II CRC patients (22 with bad prognosis). An effort to find prognosis-associated biomarkers was done in both RNA and DNA molecules.
1.SNPs and CNV data from Affymetrix SNP arrays: This work was subcontracted to the Spanish Genotyping Center (CeGen) in Santiago de Compostela. The DNA samples were extracted at ICO, analyzed for good quality DNA and sent to CeGen. The provided CEL files were tested at ICO for QC. A few samples were found to have problems (unpaired N/T and gender problems). DNAs from these samples were re-extracted and submitted to CeGen. The complete dataset was obtained by the end of September 2011. After a new QC analysis, the raw data was submitted to FBK and Crosslinks in November 2011. CRMAv2 normalization method implemented in aroma.affymetrix package was used to pre-process the 99 paired T-N samples with valid genotyping data. This method performs the following steps: calibration for offset and crosstalk between alleles, probe sequence normalization, probe-level summarization, fragment-length normalization and GC-content normalization. Raw Total Copy Number (TCN) for each tumor at a locus level was estimated using their paired normal tissue as a reference. B allele frequencies (BAF) were estimated and calibrated using the TumorBoost algorithm and the genotypes of the pool of normal samples. Samples showing an erratic TCN profile were considered defective and removed from further analysis (n=14). After pre-processing, paired Parent Specific Circular Binary Segmentation (PSCBS) algorithm was used to identify consecutive regions, which shared Copy Number Variations (CNV). Chromosomal copy number losses and gains of TCN were then identified. Moreover, LOH was assessed by comparing the genotypes of a SNP in paired normal and tumor samples. Segmentation identified regions with shared CNV using paired PSCBS (R package PSCBS). Gistic 2.0 software was used to identify gains (peaks in red) and deletions (peaks in blue) that appear repeatedly in CRC samples. Some regions were found to be associated with prognosis after comparison of recurrent chromosomic aberrations between the two tumor groups. In addition to the analysis of CN regions, PLINK software was used to look for SNPs significantly associated with bad-prognosis using the trend test Cochran-Armitage.
2. DNA methylation: Illumina 450K array was used to profile DNA methylation of these 100 paired samples of tumor and normal mucosa. Quality hybridization was assessed by inspection of the control probes included in the Infinium 450K chip and graphics showing overall Beta densities within samples. Assays with a detection p-value more than 0.01 were considered as failed and consequently coded to missing. Genotyping probes were use to detect samples with identification errors. Samples highlighted as outliers in a Principal Component Analysis (PCA) were also excluded. Data were normalized using the Subset-quantile Within Array Normalization (SWAN) method in order to reduce differences between probe types that translate into technical variation within and between arrays. Raw data was imported and processed using the R package minfi. In order to identify non-informative loci, we fitted a mixture of Gaussian distributions to the standard deviations (sd) of the beta values at each locus in logarithm scale using the mclust R package. A total of 275.181 probes, which mapped to 19.681 genes were used for the further analysis. A PCA was performed in logit-transformed methylation Beta values. First two components clearly discriminated among sample types. No methylation differences were observed between healthy colon and adjacent normal tissue from colorectal cancer patients at this level. Pairwise two-sided t-tests were used to assess methylation differences at probe level between Tumor (T), Normal mucosa (N) and healthy Mucosa (M). Tumor methylation data were used to fit a Cox hazard regression model to find prognosis-associated probes. The survival estimates were calculated using Kaplan-Meier method.
3. Micro-RNA data from Solid small RNA Next Generation Sequencing: RNA of the 100 paired samples of tumor and normal mucosa were sequenced using SOLiD technology to search for those miRNAs implicated in colon tumorogenesis and tumor progression. Briefly, after sequencing and quality control assessment, reads were mapped in the miRBase and quantified. Finally, a differential expression analysis was performed. The analysis pipeline for miRNAs was developed by the FBK group. The analysis of miRNAs associated to prognosis allowed identifying probes with elevated expression in tumors of bad prognosis. These miRNAs may be regulating other genes relating to progression.
4. Next Generation Sequencing data of the DNA exomes: Due to budget restrictions, a total of 42 subjects were selected for exome analysis. The mutations identified will be compared between the 21 tumors with bad prognosis and 21 tumors with good prognosis.The DNAs from tumor and paired normal mucosa were analyzed using NimbleGen Sequence Capture technology, that allows capturing the entire human exome on a single array including 180,000 coding exons and 551 miRNA exons. The captured DNA was sequenced on Illumina Genome Analyzer II machines. All tumor samples were sequenced with a average of 60X coverage and paired normal mucosas were sequenced at 40x coverage. The analysis pipeline of these data was developed by the FBK group and run in parallel at FBK and ICO.Periodic teleconferences were set to bring variants and pipeline of analysis together. In a first step, a quality control of the sequencing raw data using fastq software was done (per base and per sequence quality and GC content). As a result, all samples reached a good quality so were used in posterior analyses. Reads were mapped to reference genome (hg19) using GEM software (Marco-Sola S et al, Nature Methods 2012) and bowtie2. PCR duplicates were removed (samtools) and a local realignment was done with GATK software. Finally, samtools was used to make the variant calling and annovar software was used to annotate it. To avoid false positive variants in subsequent analyses, astringent quality control filters were applied (a minimum read depth of 10x, variant mapping quality > 30 and a minimum variant allele frequency of 10%). Variants in normal tissue were used to filter SNPs. As a result, a mean of 212 somatic mutations per sample were found in tumors (periodic reports sent to all partners). An alternative analysis comparing all filtered variants (both germline and somatic) between two groups of samples (good and bad prognosis) was also done. As a result, 509 variants had a Fisher p-value less than 0.01.
5. Expression data from Affymetrix U219 arrays: For testing in the SuNS array in a timely manner, however, the experience with DNA-based probes was low since all the technical validations had been done to show good performance for RNA assays. Thus, we decided to select the 10-probeRNA expression profile for the new SuNS array. A new search of prognosis-associated genes was performed focused in the targeted study of two pathways that we have identified with potential prognostic value: epithelial to mesenchymal transition (EMT) and genes expressed by carcinoma-associated fibroblasts (CAF). A detailed analysis of the genes in these pathways allowed the identification of 10 genes with high predictive ability, selected from a larger cluster of genes related to prognosis. CAFs are important contributors of microenvironment in determining the behavior of colorectal carcinoma (Berdiel-Acer M et al, Neoplasia 2011 Oct;13(10):931-46). Epithelial to mesenchymal transition signature (EMT) has been reported to play a pivotal role in many different kinds of tumors spread and metastasis, including colon (Loboda et al, BMC Med Genomics. 2011 Jan 20;4:9). To verify the interest of validating this signature, an independent in silico analysis was undertaken using 11 public datasets. A significant association with prognosis was found in 6 of them. Moreover, the MCC value, sensitivity and specificity were calculated for each of the datasets. These are the results in one of the significant datasets: MCC=0.27 Sensitivity=0.72 Specificity=0.63. The 10 new probes were designed by ICO and Crosslinks verified the correct design and interpretation of the strand that should be synthesized.
b. Technical validation of the SuNS microarray
After some discussion among MS, Crosslinks and ICO, it was agreed that the technical validation would be done by MS and Crosslinks in parallel using two approaches. First, RNA extracted from 10 well-characterized CRC cell lines provided by ICO would be hybridized in the SuNS slides and tested by qPCR by Crosslinks. Secondly, Crosslinks would produce synthetic RNA corresponding to the designed probes, which will be also hybridized in the SuNS arrays. RNAs from colorectal cancer cell lines with known expression profile have been extracted and sent to MS for validating the spotted arrays. Crosslinks provided Molecular Stamping with a set of workbenching tools to assess the performances of prototype HIPERDART microarrays. Specifically, a set of synthetic RNA transcripts encoding the SunS probes target transcripts sequence.
Linear High Density arrays (LHD arrays), derived from spotting of old and new sets of rollicon-probes produced by Rolling Circle Amplification in solution, have been subjected to hybridization tests in order to evaluate probe-specificity and sensitivity and the overall LHD array behavior upon clinical sample hybridization. The transcripts provided by Crosslinks have been purified using Millipore Amicon Ultracel-50 kDaMembrane, diluted in ddWater and Cy3-labeled using Kreatech Universal Linkage System. The hybridization of COL5A1C and ITM2AC targets on LHD arrays evidenced a substantial specificity of the probe sets upon transcript hybridization, confirming the good probe design and the developed protocols of sample preparation and LHD array hybridization. The hybridization of different target quantities on the arrays showed also a very good quantitative array response to decreasing target concentrations.
c. Clinical validation of the SuNS microarray
Next step was to test the hybridization with real human samples. ICO provided cDNA retro-transcribed from RNA extracted from ten tumor tissue samples. A quantitative PCR was done as a gold standard to assess cDNA quality. Despite the good cDNA quality, technical difficulties arose and these samples could not be labeled for hybridization. A second fresh RNA extraction and retro-transcription was attempted, again with good quality for qPCR but labeling was inhibited by some unknown factor. This technical difficulty has impeded testing real samples in the LHD array so far and further work is needed to set up the protocol for human samples analysis in HIPERDART chip.
The collection of RNAs from CRC tumors is being done at ICO. Patients diagnosed of CRC during the last 15 years have been systematically invited to participate in research studies. In agreement with the approved protocol by the Ethics Committee of the Hospital, for patients that provided informed consent tumor samples were collected after surgery and deposited in the tumor biobank. These samples have been retrieved and RNA has been extracted and kept frozen in preparation for clinical studies. Currently, the required sample size is available for the validation of the definitive set of probes done in stamped arrays (Table 2). Some of these samples have also being used in other collaborative studies of prognostic profiles (development of ColoPrint predictor by Agendia, Amsterdam). This is a continuous task at ICO and the recruitment and RNA extraction continues to increase the sample collection, since it is recognized that large sample sizes are needed to assure the best quality any clinical study. The RNA quality of these samples is routinely tested with Quiagen kits. These sample collection is well clinically annotated, with basic variables for all subjects: age, gender, tumor location, stage at diagnosis, pathology variables including histology, grade, pT, pN, vascular invasion, lymphocyte infiltration and resection margins. Therapies are recorded (radiotherapy for rectal cancer and chemotherapy regimens). Finally, follow-up is regularly updated in our database as patients attend periodic control visits. Our oncologists devote important time to provide this information and revise the quality.
Potential Impact:
The HIPERDART project offers great immediate and long-term health benefits as well as opportunities for SME's in Europe. This collaborative research project has shown how SME's and research institutes can cooperate to enable a new high performance microarray technology, which is well suited for clinical applications. Key elements for the new microarray technology are reduction in production time and cost, microarray feature uniformity, increased assay sensitivity, reduced assay workflow time, and decreased hybridization time.
More specifically, the project has advanced knowledge in these areas:
a) Technology
1. SuNS technology has proven a viable option for cheaper and faster microarray production using a reusable template.
This technology, that had proof of principle at the beginning of the project, has been implemented successfully to the industrial production of microarrays. Its advantages, as mentioned, are decrease time production and cost through the reuse of a microarray template and profit DNA replication ability to ensure fidelity and homogeneity of the produced arrays.
2. Rolling Cycle Amplification (RCA) technology has been successfully applied to amplify the signal sensitivity of microarray probes. This, coupled with SuNS technology provides an alternative to standard spotted microarrays.
3. Coating slide surface with brush polymers has been essential to the SuNS success. The expertise gathered in this area is part of a PhD thesis at EPFL and its protection will result in an exploitable knowledge.
The proposed strategy to modify the replica surface with a thin hydrogel coating based on polymer brushes may serve as an efficient substrate for the nano-stamping process and can fulfill the requirements that are crucial to improve the efficiency in the process. This study revealed that major objectives have been achieved in development of chemistry-adapted polymer brushes to modify replica surfaces. Therefore, enhancements in both array replication and replica surface modification as a joint effort of HIPERDART collaborators may lead to improve the microarray performances; which would have a potential positive impact in fast and reliable clinical diagnostics.
b) Engineering devices
1.Stamping device. The design of this machine has been essential for SuNS array fabrication and may be the model for future versions that can be installed in companies producing SuNS arrays.
2. Disposable hybridization device. This device will help ease the application of microarray technology out of genomic core facilities. Its small format and versatility should be attractive for any lab working with microarrays and facilitate the access to the technology to pathology departments.
c) Bioinformatics for discovery of useful profiles and clinical application to prognostic prediction
1.Data mining software, algorithms and pipelines. The work in this area has generated a collection of pipelines and routines that will help the scientific community doing research on genomic profile search for biomarker identification.
The HIPERDART WP5 activities required the development of innovative methods for biomarker profiling. A set of tools for the systematic comparison of gene signatures was derived, and implementation by software protocols completed. All software is being reachable from the European Open Source Machine Learning repository (MLPY project). In particular, methods for performance assessment can now include a measure of gene list stability (based on Algebraic Comparison of biomarker Lists) also available for partial lists, and thus different platforms with partially overlapping list of features can now be compared.
2 Clinical application: Prognostic profile for colorectal cancer. The identified candidates with prognostic value in colorectal cancer will be protected as prediction tool for personalized medicine. The extensive validation of this profile, expected to be successful, will result in a better use of adjuvant therapy in colorectal cancer.
Dissemination strategies
The strategy to disseminate the knowledge generated in the project has been to develop a web site that collects relevant information about the consortium, and outcomes oriented to the scientific community, clinical community and general public. More specific for the scientific community, presentations of results to scientific meetings in form of poster or oral communication and publication of scientific papers have been taken care during the development of the project and will continue since many results are still being drafting into papers.
Project website: http://HIPERDART.eu
Contact details:
MS
Giovanni De Ceglia gdeceglia@twof.com
Alberto Durante adurante@twof.com
Barbara Pisanelli bpisanelli@twof.com
KTH
Pål Nyren nyren@kth.se
EPFL
Harm-Anton Klok harm-anton.klok@epfl.ch
Bilgic Tugba tugba.bilgic@epfl.ch
thinXXs
Michel Neumeier neumeier@thinxxs.de
FBK
Cesare Furlanello furlan@fbk.eu
ICO
Victor Moreno v.moreno@iconcologia.net
Monica Sanchis msanchis@iconcologia.net
Crosslinks
Ronald Nanninga ronald@crosslinks-it.com
Mario Pescatori mario@crosslinks-it.com
The aim of the HIPERDART project is to develop a higher standard clinical microarray technology platform, by proposing a highly innovative probe printing technology, called Supramolecular NanoStamping (SuNS). This microarray technology will be tested with a prognostic kit useful to predict recurrence in colorectal cancer.
The new microarray technology will decrease production time and costs, increase assay sensitivity, decrease hybridization time and move this technology to the workbench of pathology departments to easy personalized medicine. To fulfill these aims, the project has optimized four technological areas:
1) Rolling Cycle Amplification (RCA) technique is used to multiply the DNA probe specific regions. While typical microarrays use oligonucleotides as probes spotted or grown on the array surface, our proposal is to deposit a long sequence of probe DNA that has been repeatedly amplified by RCA, thus increasing sensitivity to detect sample target cDNA molecules.
2) Solid surface is covered with a brush polymer that keeps the long RCA probes attached to the glass and allows replication. Aldehyde functionalized brush was identified as the best solution for this coating.
3) Production time and cost of microarrays is reduced by using a duplication process by which a template DNA array is printed. This template is duplicated by polymerases and this second strand is transferred to a replica slide. The template can be reused for producing multiple replica arrays. A robotic printer simplifies array production and reduces human intervention errors.
4) A disposable cartridge hybridization chamber has been designed that uses microfluidic technology to minimize intervention and reduce processing time and cost.
Project Context and Objectives:
HIPERDART Targets and Objectives
In applications related to clinical diagnostics, DNA microarrays have enormous potential for multiplex analysis, yet issues related to technical challenges (low reproducibility, poor signal-to-noise, and lengthy workflows) have limited their practicality and slowed regulatory approval.
The aim of the HIPERDART project is to develop a higher standard clinical microarray technology platform, by proposing a highly innovative probe printing technology, called Supramolecular NanoStamping (SuNS). This microarray technology will be tested with a prognostic kit useful to predict recurrence in colorectal cancer.
We believe that the SuNS technology offers significant advantages in microarray manufacturing, such that these challenges may be overcome. By making a large up-front investment in the template and then replicating it many times, we can effectively dilute our production costs while maintaining high performance features in the array.
SuNS technology consists in fabricating a template array with single strands of DNA that self-replicate on a surface to form complementary DNA strands, which can be transferred to other replica surface. With this procedure, the template can be reused multiple times to decrease microarray fabrication time.
From a manufacturing standpoint, the project has addressed four fundamental features of microarrays, which have so far limited their fabrication and utility: probe density, surface coating, template replication infrastructure and hybridization device.
Probe Density: Signal-to-noise and assay sensitivity, key features of microarray utility, are both determined by probe density within a feature. The higher the target-capture capacity of the feature per unit area is, the better the performance of the microarray. There is a fundamental limit, however, to the oligo density, which can be achieved on a surface, due to the radii of gyration of the DNA strands. The approach here proposed is to maximize probe density by concatamerizing oligos in the 'z-direction' with respect to the array surface. The HIPERDART consortium envisions accomplishing this by utilizing rolling-circle amplification (RCA) technology. Extension from a common primer sequence will take place within the micro-wells, each with a different circularized DNA molecule.
Utilizing the common primer sequence, the resulting long-mer strands are themselves enzymatically replicated prior to being printed onto a replica surface. By increasing probe density using RCA technology, we can dramatically increase signal-to-noise and assay sensitivity to an unprecedented degree.
HydroSuNS: surface coating by hydrogel polymes. A compelling feature of the SuNS technology is the flexibility of substrate material onto which DNA molecules may be printed. A good substrate shall have properties such that it simultaneously (a) provides ideal conditions for SuNS printing, and (b) optimizes microarray assay performance. Scientists in our consortium recently have discovered that indeed such a surface exists, and that it is the surface of a hydrogel polymer.
There are essentially two technical challenges to consider with any SuNS approach. The first challenge is to achieve nanometric conformal contact between two surfaces over a macroscopic area, while the second is to minimize damage to the template DNA, which may result from repeated cycles of surface-to-surface contact. Molecular Stamping has overcome these two challenges of SuNS by printing onto the surface of a hydrogel. The project has explored multiple polymerization coatings to find the optimal chemistry for SuNS.
Template replication infrastructure. A device for slide replication using SuNS technology needed to be built and its quality standards determined for industrial production of arrays. The project has invested in such technology, introducing robotics to decrease human intervention in an attempt to attain maximum reproducibility and low error rate.
Array Hybridization device. Conventional microscope slides (75 x 25 mm2) are most commonly used as microarray substrates. Within the HIPERDART project, a disposable microarray hybridization cartridge and device that implements microfluidics has been developed which enables very low sample volumes, while simultaneously reducing the time required for hybridization and improving the reproducibility of assay results. These requirements are essential for clinical applications, in which nucleic acids derive from small biopsies. One key element of the design is that up to four independent samples can be hybridized simultaneously, and each chamber incorporates one-way channels so that its contents are guaranteed to remain isolated from the others during the assay implementation.
Prognostic prediction for personalized medicine in colorectal cancer. Colorectal cancer is the third most common cancer worldwide after lung and breast. Cumulative risk in European countries is near 6%, both in men and women. Five year survival estimates range from 90% in Stage I to less than 5% in stage IV, and is less accurate (45-80%) in stages II & III. Adjuvant chemotherapy is standard for stage III but not for stage II, where the challenge is to identify the 25% of patients not cured by surgery alone. Clinical and pathological risk factors (T4, G3, number of assessed lymph nodes, perforation or vascular invasion) have been identified but lack standardisation.
Molecular prognostic biomarkers that allow identifying patients at high-risk for relapse have been successful in breast cancer and within this project we aim to identify equivalent biomarkers for colorectal cancer, prove their utility for predicting recurrence and implement the profile in the SuNS microarray as a clinical kit. This will allow the pathologist and oncologist identify patients that require adjuvant therapy and possibly to spare it in the patients with low risk stage III.
Despite the large amount of literature on molecular biomarker candidates, none is routinely used in the clinic due to lack of proper standardisation and validation. In this project we aim to develop a prognostic predictor that has clinical usefulness in the management of colorectal cancer patients. We will use public available data on genomic expression analyzed with microarrays and our own data generated for a series of patients well characterized clinically and that have been followed up for more than 5 years. The information of genomic expression in tumors will be complemented with additional sources of data: genomic expression and germline polymorphisms in histologically normal colonic mucosa resected from patients with colorectal cancer and copy number variation. Also, novel genomic and epigenomic (methylation and miRNA) data is being analyzed in our series and will be used to identify the best prognostic biomarker candidates. These data will be measured in a limited sample, but large enough to ensure power to detect the most relevant signals. Genomic expression of normal mucosa and inherited genomic variation has been shown to have relevant prognostic value prior to the development of the tumor. In this study we aim to use a whole-genome approach, together with appropriate data mining techniques, to select the most informative prognostic markers.
Bioinformatics expertise from two of our partners (FBK and Crosslinks) will help with the data-mining process of selecting reliable and robust predictive models, which are essential to realize the promises of personalized medicine. Valid biomarkers derived from high-throughput technologies need to undergo a quality control process that can address the control of variability along all steps of the processing pipeline. In the HIPERDART project, we will evaluate and control variability of the new technology by adopting BioDCV, a state-of-the-art microarray profiling platform that implements a complete validation set and a rich family of machine learning algorithms for feature ranking. The platform can develop predictive signatures on genome-wide chips, and it has been recently extended with a set of algebraic indicators of stability for ranked gene lists. We will compare signatures produced with the HIPERDART or alternative platforms, and within the same platform, for different technical implementation of the upstream data production and preprocessing phases. Also novel network-based algorithms will be developed to identify the best candidates.
Project Results:
The project main results will be reported for these main areas:
1. Template Production
a.Probe design and RCA
b.Surface coating
c.Stamping
2. Array production and hybridization
a. Stamping device
b. Disposable hybridization cartridge device
3.Application to prognosis prediction in CRC
a.Identification of profile
b.Validation of profile
1.Template Production for SuNS microarray
The following steps are required for template production. The DNA products resulting of these steps are deposited on a glass surface appropriately coated before stamping and array replication.
i. Design, synthesis and analysis of the model system probe
ii. Circle synthesis and analysis of the ligation products
iii. Rolling circle amplification
iv. Second strand synthesis
a. Probe design and RCA.
The first step was to produce a single sequence DNA array. A model oligonucleotide was spotted at different concentrations on the array and analyzed by hybridization with a fluorescent-labeled oligonucleotide. The difference in color indicates different concentrations of the spotted oligonucleotide. This array has been used for experiments with rolling circle amplification (RCA) on the array. The bound spotted oligonucleotide functions as a primer for the RCA reaction. The oligonucleotide has also been used in connection with experiments performed in solution.
As model system an 80-mer region of the pUC19 vector was used. Both a synthetic and a PCR generated probe were analyzed by Pyrosequencing and the results are shown below. Sequencing was performed on the 5'-ends. The results show that the synthetic oligonucleotide contained frayed ends whereas the PCR fragment gave much purer signals indicating a much purer product.
Circle production and analysis of the ligation products. For circulization of the probe several different ligases were tested; both Circligase which utilize single-stranded DNA (ssDNA) and ligases that need a splint (double-stranded DNA ligases). For analysis of the ligation products a new assay was developed. In the new assay the ligation region was analyzed by sequencing a short stretch over the ligation site (a few bases before and a few bases after the ligation site were sequenced by Pyrosequencing). By the new assay a quantitative estimate of the efficiency of the ligation reaction could be obtained. The result was compared with traditional gel analysis.
The main difference between the procedures is the use of a blocking oligonucleotide to bind the splint DNA before the analysis. The blocking oligonucleotide will bind to the splint and thereby hinder it from disturbing the detection primer to bind to the template as well as hinder it from producing signals during the sequencing procedure.
From the results obtained from the different experiments performed with the different ligases and under different conditions it was concluded that the best result was obtained with Cirligase and ssDNA. The ligation can be performed without a splint and at high temperature (60 degree Celsius). Faster ligation was obtained with fewer problems with secondary structures. The method was cost effective as low amounts of enzyme can be used if lower ligation efficiency is accepted.
Rolling circle amplification (RCA). RCA reaction was set up and optimized in solution. The following points were of importance for the success of this step:
-Product detection
-Quantification/length of the product
-Quality
-Optimization of polymerase, primer-to-circle ratio, sequence design, purity, additives, temperature, time and cost
To be able to analyze the product from the RCA reaction a new assay was set up. The standard gel analysis method is hampered by the low resolution of large DNA fragments. The new assay is based on binding of primers to each of the amplified circle fragments. The product is then analyzed by Pyrosequencing. The observed signals are compared with an internal pyrophosphate standard. By this comparison and by analysis of the initial amount of circle used an estimate of the amplification efficiency can be obtained.
Rolling circle amplification on slides. After the RCA reaction was set up and optimized to work well in solution the procedure was evaluated on slides. Parameters such as time, additives and primer density were evaluated. The first experiments were performed to find good conditions and to get pure fluorescence signals. A protocol containing 1% BSA was found to give clear signals. Although pure signals were obtained no amplification was detected. Although up to 6 hours reaction time was tested with different concentrations of enzyme, nucleotides, primers and different additives no amplification was obtained. There was no reason to believe that the hybridization and detection steps wouldn't work as with similar conditions the reaction had worked well in solution. As many different additives were tested solubility and secondary structures should not be a problem. In a couple of experiments the effect of the primer density on the chip was analyzed. We found that the primer density was an important factor. At the high density normally used no amplification was obtained whereas at lower concentrations the amplification reaction worked. Although when the concentration was decreased the RCA reaction must be very efficient to obtain signal levels necessary for an efficient detection system. At 200 times lower primer concentration a RCA amplification of 50 fold was obtained. This amplification is not enough for our aims and after many fruitless tries to improve the amplification we decided for a contingency plan.
Contingency plan/Spotting of prototype. Due to the problem to obtain the desired signal intensity utilizing RCA directly on the arrays we decided to go for a contingency plan. The plan is based on the good results obtained for the RCA reaction in solution. By utilizing RCA in solution and then spot the amplified DNA on the array we could overcome the drawback with the inefficient RCA directly on the array. To speed up the project we decided to direct start the spotting with the selected probes. Factors to consider for this part of the project are optimization of concentrations, solubility, additives and time.
This method was used to produce RCA products for the 10 selected probes from WP5, that should be used for clinical application (see table).
The oligonucleotides have the complementary sequence to the original probe sequences. Underlined bases: restriction site. Extra bases (T)3' improve ligation. 5' phosphate. The 20 bases at 5' are used for detection and RCA reaction.
The designed circles for the 10 probes were amplified by the RCA reaction. The results show that all probes were amplified but that the efficiency differed between the different probes. The ligation efficiency did also differ for the different probes. The amplified DNA was used for spotting on arrays.
b. Optimal surface coating for RCA product stamping.
This area of work was responsibility of EPFL. The objective was to develop a method to modify replica surface with a thin hydrogel coating that allows a fast and reproducible nanostamping process. To this end, polymer brushes which are thin (20-200 nm) coatings consisting of densely packed polymer chains that are tethered with one of the chain ends to the substrate were proposed as good candidates to modify the replica surface. Since these polymer brush coatings contain reactive groups that would improve the efficiency in nanostamping process due to their potentially very high surface concentration of functional groups and an internal volume. In contrast to glass slides modified with a monolayer of functional compounds, these polymer brushes can be considered as 3D substrates.
Polymer brushes prepared specifically via the grafting from strategy in which the polymerization is directly initiated by the initiator functionalized surfaces; allows an accurate control over the brush thickness, composition and architecture particularly by controlled radical polymerization techniques. Using appropriate side chain functional monomers, surface-initiated polymerization leads to the polymer brushes that can present very high surface concentrations of functional groups to capture end-functionalized bio-molecules. Moreover, due to the relatively high thicknesses (20200 nm) compared to monolayer of functional groups, these polymer brushes can also possess an internal volume that can be used to capture and bind end-functionalized bio-molecules.
In order to develop a thin polymer brush coating for the modification of the replica surface, first approach involved the direct modification of the glass substrate with polymer brushes containing active ester groups to allow immobilization of end-functionalized bio-molecules. Successful synthesis and detailed characterization of these polymer brushes were demonstrated. Thereafter, these replica surfaces were investigated for their feasibility to bind end-functionalized bio-molecules; however the results were not satisfying to continue the studies with these active ester functionalized polymer brush coatings.
The poor results on the replica surface relative to the active ester chemistry had to be overcome by introducing an alternative surface chemistry onto the replica surface; which involves functional groups that show higher stability under nanostamping conditions and allow the immobilization of end-functionalized bio-molecules. To this end, the effort was focused to successfully introduce the alternative surface chemistry onto the replica surface; which provides aforementioned properties; and to optimize the conditions for the usage of this alternative surface chemistry to allow efficient covalent binding of end-functionalized bio-molecules.
The modification of replica surface with this alternative surface chemistry was investigated in detail and fully characterized. Thereafter, these replica surfaces containing alternative surface chemistry were investigated in terms of their feasibility to covalently bind end-functionalized bio-molecules. The good outcomes led to continue with transferring this alternative surface chemistry to PDMS polymer layer as a second approach in order to enhance the mechanical properties of the hydrogel coating on the replica surface.
Modification of PDMS polymer layer with the alternative surface chemistry was successfully demonstrated. Modified PDMS polymer layer with the alternative surface chemistry and particularly the presence of functional groups to bind end-functionalized bio-molecules on PDMS polymer layer were fully characterized by different techniques.
Surfaces prepared by direct modification of glass slides or by modification of PDMS polymer layer were compared in terms of their performances as replica surface. The outcomes of this study revealed that modified PDMS polymer layer containing functional groups was a good candidate to prepare replica surface with enhanced mechanical properties for nanostamping process.
The proposed strategy to modify the replica surface with a thin hydrogel coating based on polymer brushes may serve as an efficient substrate for the nanostamping process and can fulfill the requirements that are crucial to improve the efficiency in the process. This study revealed that major objectives have been achieved in development of chemistry-adapted polymer brushes to modify replica surfaces. Therefore, enhancements in both array replication and replica surface modification as a joint effort of HIPERDART collaborators may lead to improve the microarray diagnostic capability and performances; which would have a potential positive impact in fast and reliable clinical diagnostics.
c. Stamping of replicated RCA on EPFL surface.
KTH has produced more than 80 stamping experiments of RCA products using both EPFL made and MS made replica surfaces. All MS surfaces consist of PDMS layers on glass, while all of EPFL surfaces (except one) are PHEMA-Aldehyde polymer brush on glass. The different one is a PHEMA-Aldehyde polymer brush on glass/PDMS (As from a specific deliverable in WP3, D3.2). Templates used for stamping were replicated either by MS or Pal Nyren group at KTH.
The most promising result collected so far has been stamping of a KTH replicated template (Hi16 Protocol) on the EPFL PHEMA-Aldehyde polymer brush on glass/PDMS. It is worth it to underline that this stamp has been done with HIPERDART outcomes only: 10 probes selected by ICO, synthesized by KTH, spotted by MS, replicated by KTH, and stamped on a replica surface made by EPFL.
For experiments performed on 3 HIPERDART probes no DNA transfer occurred for the first set of trials except for 793-18 where some spots are visible. Another set of replica surfaces, partly produced by MS and partly from EPFL, was used for stamping of templates replicated by the Pål Nyrén group at KTH.
A first set of stamping was performed in the conditions described below. Array replication has been performed by KTH. Only MS800-05 gave detectable signals, although background signal is high.
The new 10-probes prototype template was used for stamping on different surfaces.
A last experiment has been performed using a new epoxide-coated template surface developed at MS, replicated by KTH and stamped on three different replica surfaces: MS PDMS-APTES on glass, EPFL PHEMA-Aldehyde on glass and EPFL o PHEMA-Aldehyde on glass/PDMS.
Array replication and stamping. On the spotted array second strand synthesis (replication) and stamping were performed.
Conclusions of template production. Since TB423_H10 is the most promising result overall, in the next future we should focus the stamping experiments on EPFL PHEMA-Aldehyde on glass/PDMS replica surfaces, using already available Aldehyde-coated templates to be replicated by KTH with protocol Hi-16. Stamping on EPFL PHEMA-Aldehyde on glass replica surfaces will not be discontinued but considered at lower priority. We do consider that major achievements have been reached on both array replication and replica surface development. Indeed, although we do not have statistical relevant numbers yet, the proof-of-concept of Second Strand Synthesis on the array, and the development of chemistry-adapted polymer brushes surfaces has been obtained.
2. Array production and hybridization
The template array is the basis for replica array production that can be used to hybridize nucleic acid samples. The HIPERDART project has an engineering section for designing and QC the stamping device, contributed by Molecular Stamping, and for designing and producing a disposable cartridge to ease the hybridization.
a. Stamping device.
The design of the stamping device was completed on schedule cooperating with Advance Automation engineers. Thanks to the excellent results of the cooperation, we have been able to anticipate the Factory Acceptance Test (FAT performed in Advance Automation facility in USA), the shipment to Italy, the equipment installation and the final Site Acceptance Test (SAT).
The equipment was installed in the facilities of Molecular Stamping (Trento, Italy) and Site acceptance test was conducted together with Advance Automation Engineers according to their AA standard procedure. Following the formal SAT test, the equipment was extensively checked, in order to verify its performances and to identify the best set of process parameters that could optimize the overall SuNS replica process.
Several automation hardware was designed and realized to improve the original model and decrease human intervention in the replica array fabrication process:
- Barcode Reader Integration
- Double Microarray Stamping
- Buffer Bottles Auto-Switch
- Tool for Slide Alignment before loading
- Tool for master-replica separation after stamping
- Process Data transferring for remote Monitoring
- Robotic load/download arm
Barcode reader Integration. For an automatic registration of the barcode of replicas and templates we have developed hardware and software interface for barcode reader. We added a hardware and software support to the Stamping Machine for the barcode reader; we here show both the Hardware and Software Interface:
Double Microarray Stamping Chuck. To improve the manufacturing capacity we have developed a Double Chuck; this allows to perform two DNA microarray stamps at once. Two templates are loaded on the upper chuck and two replica surfaces are stamped on the bottom chuck. A dedicated load unload tool has been designed and produced, to load/unload two slides at the same time.
Buffer Bottles Auto-Switch. A dedicated Switching Buffer Tool has been designed and realized; this tool allows to switch between six different buffer bottles automatically. Along with this, a remote control of the switching can be performed with a Software Interface.
Tool for Slide Alignment before Loading. For faster loading and to minimize any alignment operator error a tool has been designed and developed. This is composed by two parts: Picker Vacuum Tool that catches the slides by using vacuum, and Alignment Station, that trough a Washer and a Register Plate aligns the slide with the Picker Vacuum tool.
Tool for Master-Replica Separation after Stamping. For faster template-replica sandwich separation after stamp, and to avoid to brake any of these surfaces, we have designed and developed a dedicated tool and. This is composed by two parts.
- Picker Vacuum Tool that catches the sandwich by using vacuum
- Separation Station to divide the sandwich applying the vacuum on the master and replica and moving up the picker vacuum tool trough four pins
Process Data transferring for remote Monitoring. A software interface has been developed to allow the automatic data upload from the equipment to the controlling pc. Utilizing the features, giving the pc connected on line, all the process data can be monitored remotely.
Robotic load/download arm. An Industrial Manipulator is going to be designed in the future to complete the automatic loading and unloading of the slides. A possible implementation could be represented by the COMAU SMART NS 12-1.85 considered suitable for our application.
The activities performed in the last 12 months verified all mechanical and electrical compatibility with stamping equipment and microarray glass slides.
We double-checked that using this commercial manipulator we could get the best cost benefit ratio. The selected robot matches perfectly our need of low cost, dimension, operations (degree of freedom for each mechanical joint) and compatibility.
-Selected robot characteristics are:
-Height 1485 mm
-Width 506 mm
-Arm Max extension 1651 mm
-Base Rotation angle ± 180°
-Operating area and dimension allow a perfect match with stamping equipment
Robot coupled with Stamping Equipment
-Stamping equipment has been modeled by using 'SolidWork 3D' software
-SolidWork 3D was used to introduce the handling arm to validate its compatibility with the stamping equipment
-A complete 3D model has been prepared
-The manipulator COMAU SMART NS 12-1.85 can be easily combined with stamping equipment picker vacuum tool. In such a way, the load/unload operation will be completely automated, providing the slide holder is fixed in the robot operating range and utilizing the X,Y,Z movement and the ability to rotate clockwise and anti-clockwise.
Robot arm coupled with picker vacuum tool
By using this commercial solution, the following operations improvement will be achieved:
-Optimum slide alignment;
-Shorter slide load/download time;
-Softer slide handling and as consequence, minimization of slide breakage;
-Improved equipment and process troubleshooting by eliminating systemic errors caused by human
Quality Control of stamping device
A significant number of slides was ran through the machine to assess its stability and to confirm process repeatability. Results have been very good, even comparing the AA2 equipment performances with the original prototype AA1. The effects of the following process variables have been investigated with the instrument.
The QC thresholds and the related procedures to be used in production environment have been completely identified and tested by analyzing and summarizing all the measurement performed on each single slide. The best performing set of parameters have been selected.
The four main stamping variables can be considered:
I. Replica Sublayer (Thickness and Replica Signal Intensity)
II. Fluorophore type
III. Stamping Temperature
IV. Mechanical Parameters
I. Replica Sublayer Surface/Thickness
Polydimethylsiloxane (PDMS) was selected as best performing layer for stamping surfaces. Aminopropyltrimethoxysilane (APTES) has been used to functionalize spin-coated freshly oxidized PDMS layers. Both mentioned sublayers have been chosen according to our experiments. They benefit by a shorter production time, safer involved chemicals and the following better performances:
-Higher Homogeneity
-Higher peak of Transfer Level up to 40 %
-greater than 95 % spot with SBR greater than 10
The optimal thickness has been found to be ranging from 200 to 500 nm; Profilometer readings are used as standard QC of the coated polymer layers.
Replica's signal intensity. The stamping outputs have been used to define the QC limits of replicas; an internal parameter, ie Process Yield (PY), has been defined, being PY=1 an intensity of 2000 at 90/90 as Laser Power/PMT Gain settings. These QC thresholds can be summarized as follows:
-Intensity: Higher than PY = 5
-Intra-slide Uniformity %: Lower than 40 %
-% of spots with SB greater than 10: At least 97 %
II. Fluorophore
In order to improve the stability of the signal, we tested ATTO 647 N fluorophore as new dye and we compared it with ALEXA 647 that was our reference point during the first project period.
-ATTO 647 N templates show double stability of the signal
-ATTO 647 N replicas show Higher stability of the signal and 1/3 background intensity
III. Stamping Temperature(s)
Different stamping temperature tests have been done in the last months.
We performed tests at 30, 45, 60 and 80 °C, hybridizing with both primer and with gDNA sequences.
The result is that 80°C is the best stamping temperature; this temperature leads to
-% of spots with SBR greater than 10 near 100%
-High transfer level
Ratio of average Signal-Background on the replica after hybridization over average Signal-Background on the related template. The unloading temperature was set to 48°C because higher temperatures could remove both strands of DNA from the template surface.
IV. Mechanical Parameters
The following set of Mechanical Parameters lead to the best stamping outputs when using Super Slow Approach and has been chosen as standard operating procedure for HIPERDART slide production:
-Soak position 60 sec
-2 position 60 sec
-3 position 60 sec
-4 position 60 sec
-5 position 60 sec
-Float position 0 sec
-Contact position 10 sec
-Full force 85kg
Quality Control procedures
Two standard Operating Procedures have been issued to formalize the QC step.
First procedure, named OPN-HYB, is applicable to QC slide Hybridization.
Second procedure, named OPN-STQC, is applicable to actual slide Quality Control step.
In conclusion, stamping equipment has been successfully designed, manufactured and installed. Stamping process has been developed and optimized for each single process variable. QC procedures and threshold values have been set and validated. Automated equipment for production has been designed and tested.
b. Development of the disposable hybridization cartridge
Within this task, a novel hybridization cartridge was developed. The new hybridization chamber targets small sample volumes. The goal is to reduce both the reaction time and the sample consumption to improve hybridization efficiency. Microfluidics offers a prerequisite for reduced sample volumes. Diffusion distances are greatly reduced in microfluidic systems and therefore reaction times are reduced in diffusion limited systems. Also, microfluidic system offer excellent controllability of fluids. Networks of microfluidic channels can be constructed to allow to process different samples at the same time while eliminating any crosstalk and therefore greatly simplify the experimental protocol for multiple experiments. While the experimental set-up needed to run the cartridge will allow fast and reproducible hybridization experiments. The cartridge is intended to be a disposable.
Prototyping can be done by milling or laser ablation of polymer substrates. However, the surface properties (e.g. surface roughness) are depending on the applied manufacturing technique. Therefore it is important to make tests in a very early project stage using structures manufactured using the method which will also be applied later on for series production. In most cases this will be micro injection molding.
Within the project Development of the disposable hybridization cartridge, the following deliverables have been addressed:
-Report on the specifications of the hybridization set-up and concept of cartridge and set-up
-First generation of cartridge and test set-up
-Second generation of cartridge
The collection of specifications and the cartridge concept have been reported. Three different concepts have been presented:
1)Poylmer substrate & disposable assembly
2)Glass slide substrate & disposable housing
3)Glass slide substrate & disposable housing with integrated fluid control
The first generation of the hybridization cartridge and experimental test setup has been realized according to concept 3. The design of the cartridge incorporates beside the hybridization chamber the fluidic storage and waste deposition on chip as well as the fluidic transport and fluidic management by means of a pump and a multitude of valves. Since volumes of the washing solutions from the original hybridization protocol are much too large for on chip storage and waste deposition, the approach for the cartridge design was to minimize that volume by pumping the washing solutions back and forth to achieve adequate results. Five washing solutions with a total volume of 20 ml (4ml each) can be operated with the cartridge (original protocol requires 70 ml). Therefore special fluidic management was applied to handle the waste volume with using the volume of the storage containers of the washing solutions for the waste once they are empty.
The first generation cartridge was milled from polycarbonate material. A pressure sensitive adhesive tape was used for sealing the fluidic channels. A double sided pressure sensitive tape was used to apply the glass slide with the microarray to the hybridization chamber of the cartridge. The pump membrane and the valves have been made from elastomeric material. The pneumatic actuated pump and valve membranes are clamped to the cartridge with a pneumatic manifold part, which was also milled from polycarbonate.
The experimental test setup includes membrane pumps for generating the pneumatic pressure and vacuum, electromagnetic pneumatic valves for switching the pump and valve membranes on the cartridge, PCB boards with driver electronics for the electromagnetic valves and temperature control and two data acquisition boards with USB interfaces to the PC.
First tests with spotted microarray glass slides show that the double sided adhesive tape for the sealing and assembly of the spotted microarray slide was not performing under heat (slide was not removable without damage after hybridization). Within tests with an alternate sealing method, the hybridization of the microarray was successful. Analysis shows that the hybridization signals are very good, but signal uniformity is subject for further improvement (poor wash efficiency).
For the design of the second generation hybridization cartridge the following main objectives have been targeted:
-Hybridization of four independent microarrays
-Injection molded disposable cartridge
-Fluid storage (wash solutions and waste) off chip
-Flow control on chip as well as off chip. Pump off chip to provide more flow rate flexibility for wash sequence optimization.
-Optimization of glass slide handling (attachment/removal).
For the second generation of the cartridge, a concept for four independent hybridization chambers and sample lines has been elaborated according to the project proposal and an equivalent design for injection molded components has been made. From the design of the hybridization chamber (50 μm depth), a sample volume of 25 μl will be required for each sample line, which is less than half of the volume compared to other hybridization devices.
The cartridge will be delivered as an assembly of the card with a laser welded cover film (both made from polycarbonate) on the back and a thermoplastic elastomeric sealing for the microarrays. For the sealing of the sample ports, additional plugs from thermoplastic elastomeric material are supplied. In a first step, the microarray glass slide is clicked on the cartridge with an integrated hook. The microarrays are finally sealed by application of force towards the seal in the instrument. In a second step, the sample volumes are applied on the sample ports with a pipette tip. The samples are now loaded in the cartridge sample containers and agitation containers. After sealing the sample ports with the plugs, the cartridge is ready to be loaded into the instrument.
The instrument itself provides beside the temperature control a set of parallel driven plunger actuators to manage the hybridization protocol as well as fluidic interfaces for the waste and wash fluids. The wash fluids are driven with external pumps.
The cartridges have been injection molded from black polycarbonate. The sample port plugs have been injection molded from a thermoplastic elastomeric material. The chamber sealing parts were laser cut from an elastomeric material.
A breadboard was built up for disposable cartridge testing. The user interface was implemented on a Netbook size computer. Communication between computer and breadboard was implemented over USB connection.
The breadboard and a set of cartridges (R2 card) have been tested. The temperature stability of the system was given and hybridization was successfully shown on a 4 fold microarray. Nevertheless, optimization steps have been necessary for the cartridge and the breadboard. For the cartridge, the sample filling of the sample container was not uniform. A second set of cartridges (R3 card) have been assembled with post machined flow stops structures in the sample containers.
Further hybridization tests show that the sample loading improved. DNA hybridization (spike-in oligonucleotide) results showed a big improvement of the hybridization signal inside the same slide, both in terms of slide uniformity and fluorescence signal. Molecular Stamping reports that overall results for the cartridge testing are in general very good.
3. Application to prognosis prediction in colorectal cancer
This last area of the project aimed to apply the developed technology to a specific clinical problem, aiming to design a clinical prognostic tool to help personalize colorectal cancer (CRC) therapy according to a prognostic prediction. The tasks in this section are
a.Identification of an expression profile useful to predict CRC prognosis and provide a set of targets for the SuNS array fabrication
b.Technical validation of the SuNS microarray
c.Clinical validation of the profile
These tasks have been jointly performed by ICO as clinical group and FBK and Crosslinks as bioinformatics groups.
a.Development of a series of predictors for clinical outcomes and selection of probes for the prototype SuNS microarray
i) Build the predictors from existing data
The HIPERDART WP5 goal is to develop a prognostic predictor for colorectal cancer patients. In the first phase, we analyzed genomic expression data from both international archives (GEO datasets 2-6, and ArrayExpress dataset 1) and the MECC unpublished data (dataset 7) provided by ICO to the WP5 partners to explore gene panels to identify a short list of clinical candidate biomarkers. A total of 1482 expression arrays from 9 studies has been collected.
The data were used both as originally provided on the Web and after normalization and quality control procedures performed within the WP5 activities (CrossLink). Many diagnostic and prognostic outcomes were taken into account according to the available clinical information (which was not the same for all datasets): Dukes Stage A vs D , B vs C, OS and EFS status 0 and follow-up time >1000 vs status 1 (on the B+C, on MSS, on B+C and MSS), and other clinical phenotypes. Moreover, the choice of the binary classification problem addressed for each dataset depended on the number of samples available for each class (too unbalanced tasks were avoided). As a result, the following 10 genes were selected: ITM2A, COL5A1, LIF, KLF4, VDR, CHD2, CALR, NR5A2, SEC62, RPS5.
With this proposed signature (leave one out cross validation) we obtain the following values: Accuracy: 0.75 Sensitivity: 0.35 , Specificity: 0.96 MCC: 0.41 AUC: 0.78
Our conclusions are that these genes aren't very useful in practice for prediction because when cross validation is applied, the accuracy decreases fast. So they aren't very good for our purposes.
ii)Detailed analysis of the published prognosis signatures in CRC.
The comprehensive study of published prognostic signatures on CRC was published in the open access journal PLoS ONE: Sanz-Pamplona R, et al. Clinical Value of Prognosis Gene Expression Signatures in Colorectal Cancer: A Systematic Review. PLoS ONE 2012; 7(11): e48877. doi:10.1371/journal.pone.0048877. This paper had been written jointly by partners from ICO and FBK. The paper underlines the difficulties in finding good prognosis predictors in CRC. The conclusion of the study was that, though some of the 33 signatures studied show significant association with prognosis, the predictive ability is limited and many of the published signatures only perform well in the datasets used to develop them. This is a typical problem of over-fitting and limited validation, which has been recognized in many signatures . Anyway, the clinical utility of some expression profiles analyzed was reasonable, with average 3 yr separation of survival curves around 20%, which may be useful in practice.
iii)Work to identify better prognostic candidates for our validation in the SuNS microarray.
In this line of work we have pursued five parallel analyses of our data aiming to identify a set of interesting candidates for extensive validation in clinical samples. The analyses have used data generated within the ICO COLONOMICS series (see http://www.colonomics.org online), that has analyzed 100 tumors, 100 paired normal mucosa from stage II CRC patients (22 with bad prognosis). An effort to find prognosis-associated biomarkers was done in both RNA and DNA molecules.
1.SNPs and CNV data from Affymetrix SNP arrays: This work was subcontracted to the Spanish Genotyping Center (CeGen) in Santiago de Compostela. The DNA samples were extracted at ICO, analyzed for good quality DNA and sent to CeGen. The provided CEL files were tested at ICO for QC. A few samples were found to have problems (unpaired N/T and gender problems). DNAs from these samples were re-extracted and submitted to CeGen. The complete dataset was obtained by the end of September 2011. After a new QC analysis, the raw data was submitted to FBK and Crosslinks in November 2011. CRMAv2 normalization method implemented in aroma.affymetrix package was used to pre-process the 99 paired T-N samples with valid genotyping data. This method performs the following steps: calibration for offset and crosstalk between alleles, probe sequence normalization, probe-level summarization, fragment-length normalization and GC-content normalization. Raw Total Copy Number (TCN) for each tumor at a locus level was estimated using their paired normal tissue as a reference. B allele frequencies (BAF) were estimated and calibrated using the TumorBoost algorithm and the genotypes of the pool of normal samples. Samples showing an erratic TCN profile were considered defective and removed from further analysis (n=14). After pre-processing, paired Parent Specific Circular Binary Segmentation (PSCBS) algorithm was used to identify consecutive regions, which shared Copy Number Variations (CNV). Chromosomal copy number losses and gains of TCN were then identified. Moreover, LOH was assessed by comparing the genotypes of a SNP in paired normal and tumor samples. Segmentation identified regions with shared CNV using paired PSCBS (R package PSCBS). Gistic 2.0 software was used to identify gains (peaks in red) and deletions (peaks in blue) that appear repeatedly in CRC samples. Some regions were found to be associated with prognosis after comparison of recurrent chromosomic aberrations between the two tumor groups. In addition to the analysis of CN regions, PLINK software was used to look for SNPs significantly associated with bad-prognosis using the trend test Cochran-Armitage.
2. DNA methylation: Illumina 450K array was used to profile DNA methylation of these 100 paired samples of tumor and normal mucosa. Quality hybridization was assessed by inspection of the control probes included in the Infinium 450K chip and graphics showing overall Beta densities within samples. Assays with a detection p-value more than 0.01 were considered as failed and consequently coded to missing. Genotyping probes were use to detect samples with identification errors. Samples highlighted as outliers in a Principal Component Analysis (PCA) were also excluded. Data were normalized using the Subset-quantile Within Array Normalization (SWAN) method in order to reduce differences between probe types that translate into technical variation within and between arrays. Raw data was imported and processed using the R package minfi. In order to identify non-informative loci, we fitted a mixture of Gaussian distributions to the standard deviations (sd) of the beta values at each locus in logarithm scale using the mclust R package. A total of 275.181 probes, which mapped to 19.681 genes were used for the further analysis. A PCA was performed in logit-transformed methylation Beta values. First two components clearly discriminated among sample types. No methylation differences were observed between healthy colon and adjacent normal tissue from colorectal cancer patients at this level. Pairwise two-sided t-tests were used to assess methylation differences at probe level between Tumor (T), Normal mucosa (N) and healthy Mucosa (M). Tumor methylation data were used to fit a Cox hazard regression model to find prognosis-associated probes. The survival estimates were calculated using Kaplan-Meier method.
3. Micro-RNA data from Solid small RNA Next Generation Sequencing: RNA of the 100 paired samples of tumor and normal mucosa were sequenced using SOLiD technology to search for those miRNAs implicated in colon tumorogenesis and tumor progression. Briefly, after sequencing and quality control assessment, reads were mapped in the miRBase and quantified. Finally, a differential expression analysis was performed. The analysis pipeline for miRNAs was developed by the FBK group. The analysis of miRNAs associated to prognosis allowed identifying probes with elevated expression in tumors of bad prognosis. These miRNAs may be regulating other genes relating to progression.
4. Next Generation Sequencing data of the DNA exomes: Due to budget restrictions, a total of 42 subjects were selected for exome analysis. The mutations identified will be compared between the 21 tumors with bad prognosis and 21 tumors with good prognosis.The DNAs from tumor and paired normal mucosa were analyzed using NimbleGen Sequence Capture technology, that allows capturing the entire human exome on a single array including 180,000 coding exons and 551 miRNA exons. The captured DNA was sequenced on Illumina Genome Analyzer II machines. All tumor samples were sequenced with a average of 60X coverage and paired normal mucosas were sequenced at 40x coverage. The analysis pipeline of these data was developed by the FBK group and run in parallel at FBK and ICO.Periodic teleconferences were set to bring variants and pipeline of analysis together. In a first step, a quality control of the sequencing raw data using fastq software was done (per base and per sequence quality and GC content). As a result, all samples reached a good quality so were used in posterior analyses. Reads were mapped to reference genome (hg19) using GEM software (Marco-Sola S et al, Nature Methods 2012) and bowtie2. PCR duplicates were removed (samtools) and a local realignment was done with GATK software. Finally, samtools was used to make the variant calling and annovar software was used to annotate it. To avoid false positive variants in subsequent analyses, astringent quality control filters were applied (a minimum read depth of 10x, variant mapping quality > 30 and a minimum variant allele frequency of 10%). Variants in normal tissue were used to filter SNPs. As a result, a mean of 212 somatic mutations per sample were found in tumors (periodic reports sent to all partners). An alternative analysis comparing all filtered variants (both germline and somatic) between two groups of samples (good and bad prognosis) was also done. As a result, 509 variants had a Fisher p-value less than 0.01.
5. Expression data from Affymetrix U219 arrays: For testing in the SuNS array in a timely manner, however, the experience with DNA-based probes was low since all the technical validations had been done to show good performance for RNA assays. Thus, we decided to select the 10-probeRNA expression profile for the new SuNS array. A new search of prognosis-associated genes was performed focused in the targeted study of two pathways that we have identified with potential prognostic value: epithelial to mesenchymal transition (EMT) and genes expressed by carcinoma-associated fibroblasts (CAF). A detailed analysis of the genes in these pathways allowed the identification of 10 genes with high predictive ability, selected from a larger cluster of genes related to prognosis. CAFs are important contributors of microenvironment in determining the behavior of colorectal carcinoma (Berdiel-Acer M et al, Neoplasia 2011 Oct;13(10):931-46). Epithelial to mesenchymal transition signature (EMT) has been reported to play a pivotal role in many different kinds of tumors spread and metastasis, including colon (Loboda et al, BMC Med Genomics. 2011 Jan 20;4:9). To verify the interest of validating this signature, an independent in silico analysis was undertaken using 11 public datasets. A significant association with prognosis was found in 6 of them. Moreover, the MCC value, sensitivity and specificity were calculated for each of the datasets. These are the results in one of the significant datasets: MCC=0.27 Sensitivity=0.72 Specificity=0.63. The 10 new probes were designed by ICO and Crosslinks verified the correct design and interpretation of the strand that should be synthesized.
b. Technical validation of the SuNS microarray
After some discussion among MS, Crosslinks and ICO, it was agreed that the technical validation would be done by MS and Crosslinks in parallel using two approaches. First, RNA extracted from 10 well-characterized CRC cell lines provided by ICO would be hybridized in the SuNS slides and tested by qPCR by Crosslinks. Secondly, Crosslinks would produce synthetic RNA corresponding to the designed probes, which will be also hybridized in the SuNS arrays. RNAs from colorectal cancer cell lines with known expression profile have been extracted and sent to MS for validating the spotted arrays. Crosslinks provided Molecular Stamping with a set of workbenching tools to assess the performances of prototype HIPERDART microarrays. Specifically, a set of synthetic RNA transcripts encoding the SunS probes target transcripts sequence.
Linear High Density arrays (LHD arrays), derived from spotting of old and new sets of rollicon-probes produced by Rolling Circle Amplification in solution, have been subjected to hybridization tests in order to evaluate probe-specificity and sensitivity and the overall LHD array behavior upon clinical sample hybridization. The transcripts provided by Crosslinks have been purified using Millipore Amicon Ultracel-50 kDaMembrane, diluted in ddWater and Cy3-labeled using Kreatech Universal Linkage System. The hybridization of COL5A1C and ITM2AC targets on LHD arrays evidenced a substantial specificity of the probe sets upon transcript hybridization, confirming the good probe design and the developed protocols of sample preparation and LHD array hybridization. The hybridization of different target quantities on the arrays showed also a very good quantitative array response to decreasing target concentrations.
c. Clinical validation of the SuNS microarray
Next step was to test the hybridization with real human samples. ICO provided cDNA retro-transcribed from RNA extracted from ten tumor tissue samples. A quantitative PCR was done as a gold standard to assess cDNA quality. Despite the good cDNA quality, technical difficulties arose and these samples could not be labeled for hybridization. A second fresh RNA extraction and retro-transcription was attempted, again with good quality for qPCR but labeling was inhibited by some unknown factor. This technical difficulty has impeded testing real samples in the LHD array so far and further work is needed to set up the protocol for human samples analysis in HIPERDART chip.
The collection of RNAs from CRC tumors is being done at ICO. Patients diagnosed of CRC during the last 15 years have been systematically invited to participate in research studies. In agreement with the approved protocol by the Ethics Committee of the Hospital, for patients that provided informed consent tumor samples were collected after surgery and deposited in the tumor biobank. These samples have been retrieved and RNA has been extracted and kept frozen in preparation for clinical studies. Currently, the required sample size is available for the validation of the definitive set of probes done in stamped arrays (Table 2). Some of these samples have also being used in other collaborative studies of prognostic profiles (development of ColoPrint predictor by Agendia, Amsterdam). This is a continuous task at ICO and the recruitment and RNA extraction continues to increase the sample collection, since it is recognized that large sample sizes are needed to assure the best quality any clinical study. The RNA quality of these samples is routinely tested with Quiagen kits. These sample collection is well clinically annotated, with basic variables for all subjects: age, gender, tumor location, stage at diagnosis, pathology variables including histology, grade, pT, pN, vascular invasion, lymphocyte infiltration and resection margins. Therapies are recorded (radiotherapy for rectal cancer and chemotherapy regimens). Finally, follow-up is regularly updated in our database as patients attend periodic control visits. Our oncologists devote important time to provide this information and revise the quality.
Potential Impact:
The HIPERDART project offers great immediate and long-term health benefits as well as opportunities for SME's in Europe. This collaborative research project has shown how SME's and research institutes can cooperate to enable a new high performance microarray technology, which is well suited for clinical applications. Key elements for the new microarray technology are reduction in production time and cost, microarray feature uniformity, increased assay sensitivity, reduced assay workflow time, and decreased hybridization time.
More specifically, the project has advanced knowledge in these areas:
a) Technology
1. SuNS technology has proven a viable option for cheaper and faster microarray production using a reusable template.
This technology, that had proof of principle at the beginning of the project, has been implemented successfully to the industrial production of microarrays. Its advantages, as mentioned, are decrease time production and cost through the reuse of a microarray template and profit DNA replication ability to ensure fidelity and homogeneity of the produced arrays.
2. Rolling Cycle Amplification (RCA) technology has been successfully applied to amplify the signal sensitivity of microarray probes. This, coupled with SuNS technology provides an alternative to standard spotted microarrays.
3. Coating slide surface with brush polymers has been essential to the SuNS success. The expertise gathered in this area is part of a PhD thesis at EPFL and its protection will result in an exploitable knowledge.
The proposed strategy to modify the replica surface with a thin hydrogel coating based on polymer brushes may serve as an efficient substrate for the nano-stamping process and can fulfill the requirements that are crucial to improve the efficiency in the process. This study revealed that major objectives have been achieved in development of chemistry-adapted polymer brushes to modify replica surfaces. Therefore, enhancements in both array replication and replica surface modification as a joint effort of HIPERDART collaborators may lead to improve the microarray performances; which would have a potential positive impact in fast and reliable clinical diagnostics.
b) Engineering devices
1.Stamping device. The design of this machine has been essential for SuNS array fabrication and may be the model for future versions that can be installed in companies producing SuNS arrays.
2. Disposable hybridization device. This device will help ease the application of microarray technology out of genomic core facilities. Its small format and versatility should be attractive for any lab working with microarrays and facilitate the access to the technology to pathology departments.
c) Bioinformatics for discovery of useful profiles and clinical application to prognostic prediction
1.Data mining software, algorithms and pipelines. The work in this area has generated a collection of pipelines and routines that will help the scientific community doing research on genomic profile search for biomarker identification.
The HIPERDART WP5 activities required the development of innovative methods for biomarker profiling. A set of tools for the systematic comparison of gene signatures was derived, and implementation by software protocols completed. All software is being reachable from the European Open Source Machine Learning repository (MLPY project). In particular, methods for performance assessment can now include a measure of gene list stability (based on Algebraic Comparison of biomarker Lists) also available for partial lists, and thus different platforms with partially overlapping list of features can now be compared.
2 Clinical application: Prognostic profile for colorectal cancer. The identified candidates with prognostic value in colorectal cancer will be protected as prediction tool for personalized medicine. The extensive validation of this profile, expected to be successful, will result in a better use of adjuvant therapy in colorectal cancer.
Dissemination strategies
The strategy to disseminate the knowledge generated in the project has been to develop a web site that collects relevant information about the consortium, and outcomes oriented to the scientific community, clinical community and general public. More specific for the scientific community, presentations of results to scientific meetings in form of poster or oral communication and publication of scientific papers have been taken care during the development of the project and will continue since many results are still being drafting into papers.
Project website: http://HIPERDART.eu
Contact details:
MS
Giovanni De Ceglia gdeceglia@twof.com
Alberto Durante adurante@twof.com
Barbara Pisanelli bpisanelli@twof.com
KTH
Pål Nyren nyren@kth.se
EPFL
Harm-Anton Klok harm-anton.klok@epfl.ch
Bilgic Tugba tugba.bilgic@epfl.ch
thinXXs
Michel Neumeier neumeier@thinxxs.de
FBK
Cesare Furlanello furlan@fbk.eu
ICO
Victor Moreno v.moreno@iconcologia.net
Monica Sanchis msanchis@iconcologia.net
Crosslinks
Ronald Nanninga ronald@crosslinks-it.com
Mario Pescatori mario@crosslinks-it.com