WO2008043040A2 - Integrated microfluidic device for preconcentration and detection of multiple biomarkers - Google Patents

Abstract

Micro fluidic device including a nanofluidic concentrator for amplifying a sample containing a biomarker until the biomarker concentration approaches the disassociation constant of a biomarker/antibody complex. A suspended microchannel resonator receives the amplified sample and generates a signal related to the number of biomarkers contained in the sample.

Description

Integrated Microfluidic Device for Preconcentration and Detection of Multiple Biomarkers

This application claims priority to Provisional Application Serial Number 60/849,581 filed October 5, 2006, the contents of which are incorporated herein by reference.

Background of the Invention

This invention relates to a system for cancer biomarker analysis and more particularly to such a system that includes a nanofluidic concentrator and a suspended microchannel resonator detector.

Despite progress in the development of new therapeutic agents for the treatment of cancer, there has been very little progress in the development of molecular markers for the early detection of cancer. With the exception of the Prostate Specific Antigen (PSA) which is currently used to screen men for the presence of prostate cancer, most cancers have no molecular marker in clinical use. There are several possible reasons for this lack of new cancer biomarkers: i) To be useful screening markers, cancer-associated markers must be released into the circulation so that they can be detected in blood or urine; ii) Early tumors are quite small, often under 10 mm in diameter. It is clear that the amount of protein secreted by such tumors will be quite small, requiring sensitive assays able to detect proteins in biological fluids at concentrations of 0.1 -1.0 ng/ml or less; iii) Individual markers must be specific for a particular type of tumor. If one is trying to diagnose breast cancer, it is not sufficient to have a marker that is produced by several other tumor types; iv) Most of our current assays for biomarkers are based on immunoassays and therefore require antibodies. Thus for any biomarker to be developed for clinical use, successful antibodies must be produced for assay development; v) New diagnostic assays must be approved by the FDA and adopted by the medical community; and, vi) Most diagnostic companies will not develop or market assays unless there is strong intellectual property tied to a particular marker.

The two hallmarks of a diagnostic marker are sensitivity and specificity. Sensitivity refers to the percentage of patients with a disease who will test positive in the assay. False negative results dilute the sensitivity of an assay. Specificity refers to the percentage of patients without disease who test as negative in the assay. False positive results dilute the specificity of a diagnostic assay. Although both are extremely important, it is apparent that low sensitivity in a cancer test can be life threatening if false negative results prevent individuals with cancer from receiving timely treatment.

To date, the only tumor biomarker approved by the FDA for widespread cancer screening is the PSA test.¹ Superscript numbers refer to the appended references, the contents of which are incorporated herein by reference. PSA testing is frequently performed on men over the age of 50. Between 1989 and 1996 prostate cancer incidence rates increased steadily with a parallel decrease in mortality from the disease.² A serum PSA value greater than 4ng/ml has routinely been used as the criterion for suspicion of prostate cancer and for further testing by biopsy. Recently, however, there has been modification to the screening protocols and some insurers will not reimburse PSA testing because of problems related to low specificity and sensitivity. As many as 1/3 of the men with a PSA value of 4ng/ml will already have metastatic disease. Consequently it was recommended to reduce the cutoff for diagnosis of prostate cancer to 2.5 ng/ml. Yet at this cutoff, the number of false-positive tests rises dramatically. The increase in sensitivity comes, therefore, with a sharp decrease in specificity. The result is overdiagnosis of prostate cancer with a concomitant increase in the number of unnecessary biopsies. Thus, despite being an accepted diagnostic test in widespread use, the PSA test does not succeed as a robust test for the diagnosis of prostate cancer.

Recently, other methods have been discovered that along with the PSA test method has enhanced the predictive power of the PSA test. These include:

1. PSA Velocity: PSA velocity is the change in PSA level over time. A steep rise in PSA level increases the likelihood of malignant prostate cancer. A recent study demonstrated a correlation between the PSA velocity and time of death from prostate cancer after radical prostatectomy. Patients whose PSA level increased by more than 2.0ng per milliliter during the year prior to diagnosis of prostate cancer were shown to be at higher risk of dying from the disease despite undergoing radical prostatectomy.⁴

2. PSA Density: PSA density considers the relationship of the PSA level to the size of the prostate. An elevated PSA might not arouse suspicion in a patient with a pre-existing enlarged prostate. Thus, consideration of PSA density may avoid unnecessary biopsy in men with elevated PSA due to benign prostate hypertrophy. The method has a drawback, however, in that some aggressive cancers may be missed in this same cohort. 3. Free versus bound PSA: Circulating PSA in the serum has been identified in two forms, free PSA or PSA bound to protein. The ratio of free to bound PSA decreases from benign to cancer i.e. there is more free PSA in benign conditions while more bound PSA in cancer. Therefore the ratio of bound/free PSA can be used as an adjunct to the total PSA level to give an additional indication of the presence of clinically relevant prostate cancer.⁵

Nevertheless, the experience with PSA and the lack of other successful individual cancer markers has led many in the cancer diagnostics field to conclude that a single biomarker cannot succeed as a stand-alone diagnostic screening tool for virtually any cancer. At the same time, improvements in genomic and proteomic technologies, especially the development of sensitive and accurate mass spectrometry tools, have led to the discovery of hundreds of potential new cancer markers. There is evidence to suggest that the next generation of cancer screening tests may employ not just one, but a small panel of less than ten biomarkers that together add statistical power to the detection of specific cancers.

Current practice suggests that the ELISA (Enzyme-Linked Immunosorbent Assay) method is the most common approach utilized for the detection of biomarkers in biological fluids followed by radioimmunoassay and, for detection in tissues, immunohistochemistry. As the names suggest, all three depend on the prior production of high affinity antibodies that recognize the desired biomarker. Immunohistochemistry will not be discussed here because it is useful principally for detecting biomarkers in stained tissue sections and is not applicable for the detection of markers in biological fluids.

An ELISA is an assay where the abundance of an antigen (e.g. a biomarker) is quantified by measuring the amount of antibody that binds to the antigen. In its simplest form, the antigen is adsorbed to a surface (often the bottom of a 96 well plate) and labeled antibody is then allowed to bind to the antigen. Selectivity can be increased if the first antibody is then recognized by a second anti-antibody (such as anti IgG) which binds to the antigen/antibody complex. Selectivity can also be increased if the antigen is first adsorbed onto a specific antibody that has been used to coat the plate. In this case, two distinct and specific antibodies must be available for the antigen of interest. For either approach, the amount of binding is typically quantified by colorimetry, luminescence or fluorescence with extreme sensitivity. For example, Ward and coworkers⁶, and Mirkin and coworkers⁷ developed chemical amplification strategies based on PCR amplification of DNA (immuno-PCR) and silver amplification (bio-bar-code) to enhance the signal from the antigen-antibody binding event. These new strategies can achieve detection limits at the aM to fJVl level. In addition to sensitivity, the capability of automation and the non- involvement of radioisotopes make ELISA-based assays versatile and well-suited for routine use.

However, an inherent disadvantage of the ELISA and related immunoassays is that the fidelity is primarily governed by the disassociation constant, IQ, of the antibody-antigen complex (Figure 1). If the target antigen concentration is significantly below the IQ, then the binding kinetics are slow and readout precision of the antigen-antibody complex is degraded by noise. The IQ depends on the properties of the antibody (e.g. monoclonal versus polyclonal) and typically ranges from 10^~8 to 10^~12 M. This limitation results in several impediments for advancing the effectiveness of cancer biomarker detection: i) protein-based biomarkers in urine and serum can be present at concentration levels that are significantly below the IQ of the antibody-antigen complex; ii) high affinity antibodies are not always available for a given biomarker. Lower affinity molecules such as protein fragments or peptides can be used instead, however ELISA performance is degraded, iii) biomarker concentrations can vary over many orders of magnitude and ELISA has a limited dynamic range of ~10³. Aside from impediments that are associated with dependence on IQ, there are additional limitations: ELISA is generally designed to measure only one analyte and consequently is not easily amenable to the simultaneous detection of multiple markers as we propose here. In some cases, ELISA performance can also be degraded by high background readings, or from toxicity of the enzyme reagents. Although immunoassays have been commonly used for detection of biomarkers in urine or serum, the remarkably small number of approved biomarkers in clinical use (only PSA) suggests that inherent technical limitations are preventing the true diagnostic and predictive power of biomarkers for cancer.

We are proposing that a method for controllably preconcentrating biomarkers to the vicinity of the IQ can alleviate many of the inherent limitations of immunoassays such as ELISA. Once a biomarker has been concentrated to the IQ, the antibody-antigen binding reaction will be faster and more complete. Provided the amplification (or gain) of the preconcentrator is adjustable, the dynamic range and detection limit of the assay will ultimately by governed by the properties of the preconcentrator and not the value of IQ. We have developed and validated a nanochannel-based preconcentrator that increases sample concentration linearly with time and can enhance the initial sample concentration by 10⁷ in approximately one hour. Since the total volume of the concentrate is 10-100 pL, we are proposing to integrate the preconcentrator with a detector of similar volume in order to avoid dilution. The detector is conceptually similar to ELISA however the readout of the antigen- antibody binding is based on the direct detection of biomarker mass with picogram resolution within a 10 pL volume. We next review prior work in areas that are related to the preconcentration and detection devices that we will present below.

The fact that there is no PCR (Polymerase Chain Reaction) equivalent for proteins severely limits technology development of tools for proteomic analysis. A practical alternative to a PCR-like technique for a specific protein is to combine sample purification/fractionation with high-efficiency preconcentration. This is a viable strategy for cancer biomarker detection since it is possible to start with a relatively large volume of sample (~1 mL or more). Highly efficient sample preconcentration techniques will allow one to use more aggressive sample separation steps (such as repeated removal of majority protein species by immunoaffmity capturing⁸), to increase the detection specificity and sensitivity.

Given the importance of sample preconcentration, several techniques, including field- amplified sample stacking (FAS)⁹, isotachophoresis(ITP)¹⁰, electrokinetic trapping¹¹, micellar electrokinetic sweeping¹², chromatographic preconcentration¹³, and membrane filter preconcentration¹⁴ have been developed. However, these techniques all fall short of achieving both high concentration factors and the flexibility required for integration with detection systems. Many techniques, such as FAS or ITP are originally developed for capillary electrophoresis, and require special arrangement of buffers with different ionic concentrations, which makes the system integration challenging. Techniques such as micellar electrokinetic sweeping rely on a detergent additive (sodium dodecyl sulfate), which has a negative impact on the downstream detection. Chromatographic preconcentration schemes can capture proteins or peptides by the hydrophobic interaction, which tends to favor larger, more hydrophobic proteins (albumins and globulins, e.g.) over smaller, more hydrophilic signaling molecules, hormones and biomarkers. Also, the concentration factor in the chromatographic preconcentration is limited by the total binding surface area in the system, and washing steps (sometimes with high-salt solution) for the elution of trapped molecules could cause dilution of the sample bolus in addition to incompatibility issues with detectors. Filtration-based preconcentration, which becomes progressively more difficult for smaller peptides, is also limited by the ambiguity of the molecular weight cut-off of the nanoporous filter membrane materials. Electrokinetic trapping techniques have been recently studied as an efficient way of concentrating protein samples, but the linearity and stability of the trapping has been an issue. Most importantly, all of the above techniques have so far demonstrated maximum concentration factors of -1000, which is not sufficient for the given problem of detecting low-abundance biomarkers out of high background of serum majority proteins.

While ELISA and radioimmunoassays are generally regarded as the gold standards in terms of sensitivity and selectivity, a number of research groups are directing efforts towards implementing such assays with microfabricated devices. The concept is that immobilized affinity capture molecules can selectivity bind biomarkers directly to the device surface and either mechanical, electrical, or optical properties of the device can provide a direct, or label- free, readout of the binding. The approach is motivated by the scalability, robustness and scales of economy associated with microfabricated devices.

Most notable is the work by Wu et al¹⁵ from UC Berkeley on using microcantilevers to detect PSA from 0.2 ng/ml to 60 μg/ml in the background of human serum albumin. The detection process stems from a previous discovery that when biomolecular binding occurs on only one surface of a microcantilever, molecular interactions between the adsorbed biomarkers can induce stress that bends the cantilever. The degree of bending, which is detected optically, is related to the concentration of the adsorbed biomolecules. This result is encouraging because it demonstrates that microcantilevers can detect a clinically relevant concentration of PSA. However, there is an increased complexity associated with the surface functionalization process since one side of the cantilever must be blocked while the other side must be made to be specific. As a result, an approach for integrating the cantilever within microfluidic channels while allowing access for functionalization has not yet been established. To date, cantilevers are individually functionalized by either immersing it a micro-capillary, or by injecting drops of the analyte onto the cantilever surface. They are subsequently immersed in a flow cell for the detection assay. Thus, there remain several innovations that must occur before the microcantilever stress sensor can be readily integrated with upstream microfluidic concentrators or separators and detect within sub-nanoliter sample volumes.

Biomolecules can also be detected by their intrinsic charge with charge sensitive devices such as silicon field-effect sensors. For example, Cui et al. 16 at Harvard demonstrated the detection of streptavidin with a biotin-functionalized nanowire. In other work, the Manalis lab 17 has detected the hybridization of DNA by silicon field-effect. While electronic readout has the advantage of providing a simple and direct interface to the digital world, we (and others¹⁸) have found that its application to protein detection is limited. This is explained by two reasons: i) the charge to mass ratio for proteins is significantly lower than for molecules such as DNA. Furthermore, not all proteins are charged, and ii) the electric field from the target protein is screened by the counterions of the buffer and since the capture antibody is typically a few nanometers in size, the electric field in the silicon is severely reduced. While sensitivity can be increased by lowering the ionic strength to reduce screening, specificity towards the target protein can be degraded.

As is the case with all immuno-based detection approaches, the dynamic range and detection limit of micro- and nano fabricated sensors will ultimately by governed by the disassociation constant of the receptor-target complex. We believe such limitations can be eliminated by integrating a preconcentrator with the detector.

Experience in the Zetter lab supports the premise that using multiple biomarkers can increase the statistical power to detect specific cancers. Their group has identified a novel prostate cancer biomarker called Thymosin βl5 (Tβl5).¹⁹ Tβl5 has a restricted expression profile, being limited to mammalian embryos and is virtually absent from normal adult tissues.²⁰ By itself, elevation of thymosin βl5 in the tumor or in patient serum or urine, identifies patients with prostate cancer who have a higher risk of going on to metastatic disease. However, when the two tests were combined, the specificity of the test rose from 55% for PSA alone to 71% for the combination test.²¹

A similar approach has been employed by Landers and colleagues.²² This group applied a panel of four markers to a gene expression screen of prostate cancer patient tissues using realtime PCR to detect gene transcripts. The markers employed were UDP-N- Acetyl-alpha- Dgalactosamine transferase 3 (GalNAc-T3), Prostate-specific membrane antigen PSMA, Hepsin, and DD3PCA3. When the gene expression data for these 4 biomarkers was combined in a logistic regression model, a predictive index was obtained that distinguished 100% of the prostate cancer samples from control samples or samples from patients with benign prostatic hyperplasia. This result indicates that the addition of multiple biomarkers to a screen can increase the accuracy of tumor detection. The implication of this test is that the use of additional markers beyond two could allow the sensitivity and specificity of a screening assay to approach 100% when no single biomarker could achieve that result.

The Zetter laboratory has experience in the development of clinically relevant assays for these markers. Most recently they have developed a competitive ELISA for detection of the prostate cancer marker thymosin βl5 and are aware of the problems and pitfalls in setting up such an assay.²³ Development of a single ELISA can take several months to more than a year. Sensitivity is rarely less than 1 ng/ml and interference by other components present in the samples is very common. Because of plate to plate and day to day variation, standard curves must be generated with purified antigen for every individual assay. A representative standard curve of a Tβl5 ELISA is shown in Figure 2. Multiple antibodies often have to be generated to find the one or two that are useful in ELISA.

Recently, the Han group developed a novel nanofluidic device that can achieve more than a million-fold sample preconcentration within an hour.²⁴ Preconcentration using this device has been demonstrated in the Han laboratory for peptides, proteins, and DNA molecules. The schematic diagram of the nanofluidic concentration device is shown in Figures 3 and 4. The entire system consists of two micro fluidic channels (a few tens of μm in dimension) bridged by a nanofluidic channel as thin as 40nm in depth. The uniformity and regularity of the 40nm channel has been confirmed by cross-sectional SEM imaging.²⁵ At moderate buffer concentrations (-1OmM), the Debye layer thickness within a nanofluidic channel is not negligible, and the nanofluidic channel becomes perm-selective when an electric field (E_n) is applied across the nanochannel. For a negatively charged surface such as SiO₂, the resulting ion current will preferentially transfer positively charged counterions over the negatively charged co-ions. This will create an extended space charge layer within the microchannel (near the nanochannel), which acts as an energy barrier for negatively charged biomolecules. Summary of the Invention

In one aspect, the invention is a microfluidic device including a nanofluidic concentrator for amplifying a sample containing a biomarker until the biomarker concentration approaches the disassociation constant of a biomarker/antibody complex. A suspended microchannel resonator receives the amplified sample and generates a signal related to the number of biomarkers contained in the sample. In a preferred embodiment the integrated system includes a concentrator channel, a nanochannel filter and a suspended microchannel resonator detector.

Brief Description of the Drawing

Figure Ia is a schematic illustration of quantitative measurement of PSA concentration.

Figure Ib is a graph of signal versus concentration showing a typical binding curve for an antibody-antigen reaction.

Figure 2 is a graph showing a standard ELISA curve for Tβl5.

Figure 3 a is a top view of a schematic diagram of a nanofluidic concentrator.

Figure 3b is a cross-sectional view of the nanofluidic concentrator along the dotted line in Figure 3 a.

Figure 3c is a schematic illustration of device layout along with dimensions.

Figures 4a-d are schematic illustrations and micrographs showing the mechanism of a nanofluidic concentrator.

Figure 5a is a fluorescence image of focused proteins (GFP) in a channel.

Figure 5b is an illustration of channel fluorescence signal profile at an initial concentration.

Figure 5 c is an illustration of channel fluorescence signal profile at a concentration of 0.33 μM GFP.

Figure 5 d is an illustration of fluorescence signal profile of concentrated GFP in a channel.

Figure 6a is a graph showing concentration of GFP solution. Figure 6b is a detailed graph of GFP solution. Figure 7a is a schematic drawing showing voltages applied to reservoirs during concentration and release (capillary electrophoresis) steps.

Figure 7b is a capillary electrophoresis (CE) electropherogram of fluorescence-labeled peptide.

Figure 7c is a capillary electrophoresis electropherogram of two simultaneously collected and launched proteins.

Figure 8 is a photomicrograph illustrating nanochannels made of silicon nitride.

Figure 9a is a micrograph showing an untreated PDMS device after contact with fluorescent markers.

Figure 9b is a micrograph showing a device coated with PEG-di (tryethoxy) silane. Figure 10a is a perspective illustration of a suspended microchannel resonator. Figure 10b is a cross-sectional view of a vibrating SMR.

Figure 10c is a cross-sectional view showing a target analyte entering the SMR without altering resonant frequency.

Figure 1Od is a cross-sectional view showing that targets bind to immobilized receptors and the high surface concentration lowers resonant frequency.

Figure 11 is an electron micrograph of three suspended microchannel resonators.

Figure 12 is a schematic illustration of a six-inch wafer containing approximately 135 SMRs.

Figure 13a is a graph of frequency shift versus time showing a shift in SMR resonant frequency to injections of NaCl at various concentrations.

Figure 13b is a graph of frequency shift versus density change showing a linear response of frequency shift versus volumetric mass density.

Figure 13c are graphs showing the response to surface mass density of avidin and biotinylated-BSA binding to interior channel walls.

Figure 14a is a perspective illustration of an integrated nanofluidic concentrator (NC) and suspended microchannel resonator (SMR) detector.

Figure 14b is a close-up view showing trapped biomolecules near the boundary of the extended space charge region.

Figure 14c is a perspective illustration showing that the concentrate is transported into the SMR by applying a negative pressure on its outlet. Figure 15a is a cross-sectional view of etched silicon channels.

Figure 15b is a cross-sectional view showing the deposition by LPCVD of silicon nitride. Figure 15c is a cross-sectional view showing the deposition of sacrificial poly-silicon. Figure 15d is a cross-sectional view showing chemical mechanical polishing of poly-Si.

Figure 15e is a cross-sectional view showing the deposit of 40 nm sacrificial poly-Si for nanochannels.

Figure 15f is a cross sectional view showing the etching of poly-Si to define nanochannels.

Figure 15g is a cross-sectional view showing the deposit of LPCVD silicon nitride.

Figure 15h is a cross-sectional view showing the etching of nitride.

Figure 15i is a cross-sectional view showing the etching of poly-Si in hot KOH.

Figure 15j is a cross-sectional view illustrating glass lid bonding.

Figure 16 illustrates fluorescent intensity (saturated) versus concentration of Akt target.

Figure 17 is a graph of surface coverage versus distance from inlet showing the percent surface coverage of SMR channel versus SMR channel length.

Figure 18 illustrates sample concentration for various times.

Figure 19 is a graph showing frequency response from anti-GFP binding to an avidin- functionalized SMR.

Figure 20a is a perspective view showing optical readout for a vibrating suspended channel.

Figure 20b is a perspective view showing an electrical readout for vibration by capacitance detection.

Figure 21 is a schematic block diagram showing separation of signal spectrum from low frequency noise.

Figure 22 comprises graphs showing signal modulation/demodulation commonly used to separate signal from noise.

Figure 23 is a circuit diagram showing an additional feedback stage used to reduce signal degradation from parasitic capacitances.

Figure 24 is a schematic illustration showing parallel detection of four biomarkers. Description of the Preferred Embodiment

The mechanism of the nanofluidic concentrator can be explained by nonlinear electrokinetic phenomena. As the electric field across the nanofilter is increased, this perm- selective current will first generate an ion depletion region near the nanofilter, as predicted by standard concentration polarization theory of ion-selective membrane (Figure 4a,b). When the E_n is increased further, the ion transport in nanochannel enters a nonlinear regime where the space charge layer (double layer) is extended into the microfluidic channel near the nanofilter, due to the strong E_n (Figure 4c). A similar phenomenon has been observed in the charged gel bead system.²⁶ Within the extended space charge layer, electroneutrality is broken (just as in the primary Debye layer), and the co-ion is prohibited from this region due to the non-zero potential. Therefore, this will become an energy barrier for anionic (negatively charged) biomolecules. Finally, when a tangential electric field in the microchannel (E_τ) is applied in addition to E_n, a strong secondary electroosmotic flow (electroosmotic flow of the second kind²⁶) will be induced in the microchannel, bringing the molecules from the sample reservoir to the biomolecule trap (Figure 4d). This induced electrokinetic flow is generally much stronger than the primary electroosmotic flow (generated by Debye layer charges), because its strength scales as the product of E_T with E_n. Therefore, this device will bring the molecules to the trap with a high speed and will trap them at the boundary between the normal and extended space charge regions. Such a process can be initiated even at buffer ionic strength as high as 1OmM and the nanochannel depth as large as 40nm since the concentration polarization, once initiated, decreases the ionic strength near the nanochannel. This will further increase the Debye length within the nanochannel, which will push the system toward the non-linear regime.

To test the operation of the device, a dilute protein or peptide solution (fluorescently labeled) was loaded into the sample reservoir of the device, and the electric field was applied to collect the molecules at the electrokinetic trap generated near the nanofilter in the microfluidic channel. By controlling and optimizing the driving potentials carefully, one could achieve continuous sample preconcentration and stacking, which was stable for several hours. Due to this stability of the trapping, one could achieve very high preconcentration factors, typically larger than 10⁶. As an illustrative example, the fluorescent images of trapped and collected proteins are shown in Figure 5a. This sample plug was collected from 33pM green fluorescent protein (GFP) solution, which was not detectable by the fluorescence microscopy detection setup used (Figure 5b). After preconcentration for ~50 minutes, the fluorescence signal from the collected sample plug (Figure 5d) was much higher than that of 0.33μM GFP solution, which was barely detectable by the same detection condition (Figure 5c). During the quantification of the preconcentration factor, protein molecules could adsorb to the surface of the device due to nonspecific binding, especially when the sample plug concentration is high (above ~μM). Before the experiment, we photobleached any GFP molecules adsorbed to the surface to make sure that any signal detected is purely due to the fresh molecules coming from the reservoir.

The nanofluidic concentration process, once established, can be stably maintained for a long time in order to achieve higher preconcentration factors. Figure 6 shows the result of 3- hour-long preconcentration from 33nM, 33pM, and 33fM GFP solutions. It can be seen that after 2-3 hours of preconcentration, the plug concentration reached well above the 0.3 μM. This is equivalent to more than 10⁷ fold preconcentration, which has never been demonstrated by any method so far, at least to the best of our knowledge.

The concentration process can be stopped by switching off the field (E_n), and the collected biomolecules can be released by either electroosmotic or pressure-driven flow as shown in Figure 7a. The nanofluidic concentrator works both for small peptides and larger proteins, as long as they are charged (Figure 7b). In Figure 7b, the nanofluidic concentrator was used as a sample injector for CE. Two proteins were collected simultaneously and launched into a microchannel for successful CE separation (Figure 7c).

There are many characteristics that make this device ideal as a component for integrated sample preparation: i) the concentration factors achieved in this device are exceptionally high, probably due to the fact that one can concentrate the dilute sample for a long time. The stability of the system is partly due to the mechanical robustness of the solid-state nanofluidic filter membrane, ii) The operation of the device is not dependent on the specific kind of buffer solution or any reagents used. We have used several different buffers (phosphate, Tris-EDTA) at several different pH values (pH 6~9). The one parameter that is important is the ionic strength of the solution (the lower the ionic strength is, the larger Debye layer would be, therefore preconcentration would be more efficient). However, even with different buffer ionic strengths, one could adjust the operation parameters (field values and the nanofilter thickness) to retain the preconcentration capability. This suggests that this device could handle a wide variety of sample buffer conditions, including organic buffers as well as unknown solutions such as serum. Even when the sample solutions have relatively high ionic strength (10OmM physiological solution, for example), one could dilute them to 1OmM for processing in the preconcentration device, iii) The concentration occurs in a micro fluidic channel that is comparable to the size of a capillary system, without using any membranes or filters blocking the flow of concentrated solution. Therefore, the nano fluidic channel does not limit the flow rate or the capacity of the device since it is simply providing an energy barrier, iv) The preconcentration device is based on the biomolecule trap generated by the extended space charge layer, and the delivery of the molecules can either be achieved by induced electroosmotic flow (as in this demonstration) or by pressure- driven flow (instead of E_T in the Figure 7. This means that the preconcentrator could be coupled both with electric-field driven and pressure-driven micro fluidic devices. Therefore, the nanofluidic concentrator could be used as an adapter between pressure-driven and electrokinetic driven micro fluidic components with different flow and field requirements, v) While the previous results were achieved with SiO₂ nanochannels, we have recently demonstrated that silicon nitride nanochannels will also concentrate biomolecules (Figure 8). This property is important since the integrated system disclosed herein will be based on a silicon nitride fabrication process. The fact that a similar extended space-charge layer was observed even from the silicon nitride nanochannels clearly validates the possibility of seamless integration between the nanofluidic concentrator and the suspended microchannel resonator (SMR) detector described below.

The large surface area-to-volume in biological microsystems necessitates precise control of surface characteristics, such as wettability, surface topology, and interfacial charge. In particular, it is necessary to modulate protein adsorption through surface modifications. For example, non-specific protein adsorption must be blocked in the proposed device. The high surface-to-volume ratios in nanofluidic devices further imply that slight inhomogeneities in the surface will cause device malfunction.²⁷ Surface modification of glass substrates via silane chemistry is well-established²⁸, but the techniques for polymer (e.g., PDMS) devices are still being developed.²⁹ Several approaches have been described for surface modification of polymer microfluidic devices, such as plasma treatment³⁰, silanization of oxidized PDMS³¹, polymer grafting³², adsorption of polyelectrolytes³³, adsorption of detergents or quaternary amines³⁴, and precoating with proteins.³⁵

One of the most frequently used technique for preventing non-specific binding is the coating by poly(ethylene glycol) (PEG), and procedures for both Si and plastic (PDMS) surfaces have been well known.³⁶ These techniques increase the (effective) binding co³⁷nstant for the nonspecific binding of proteins over ~10 μg/mL levels. We have tested the efficiency of the PEG- silane coating method in the PDMS devices, as shown in Figure 9. Similar coating methods can be used for SiZSiO₂ surfaces, with better resistance to non-specific binding.

The Manalis lab has demonstrated a fundamentally new approach for detecting biomolecular mass in the aqueous environment. Known as the suspended microchannel resonator (SMR), target molecules flow through a suspended microchannel and are captured by receptor molecules attached to the interior channel walls (Figure 1O).³⁸ As with other resonant mass sensors, the SMR detects the amount of captured target molecules via the change in resonance frequency of the channel during the adsorption. However what separates the SMR from the myriad of existing resonant mass sensors is that the receptors, targets, and their aqueous environment are confined inside the resonator, while the resonator itself can oscillate at high Q in an external vacuum environment, thus yielding extraordinarily high mass resolution.

There are three key properties that enable SMR detection: First, the mass density of biomolecules is greater than the density of the water. For example, proteins have a mass density in the range of 1.3-1.4 g/cm³.³⁹ Thus, the net mass of the fluid-filled resonator depends on the total number of biomolecules that are contained within the resonator. Second, the energy loss of the resonator due to viscous drag is negligible. As a result, the presence of fluid inside the suspended microchannel resonator does not create a measurable change in the quality factor, Q. This is not the case for the quartz crystal microbalance (QCM) and other acoustic detectors. Third, the surface to volume ratio of the microchannel is sufficiently large that the number of surface-bound molecules is generally much larger than the number of molecules contained within the microchannel volume. Thus, the binding of target biomolecules to the microchannel walls can be monitored in real-time. SMR devices can be manufactured to be compact, robust, and cost-effective by using well-established micro fabrication processes. Initially, SMR devices were fabricated at MIT facilities and packaged at the level of individual devices with PDMS micro fluidics. For sensitive detection, the suspended microchannels must be sufficiently thin so as to be effective resonators, and they must be configured for continuous fluidic delivery for real-time measurements. To address both of these requirements, the Manalis lab combined a polysilicon Damascene process, sacrificial layer etching in hot potassium hydroxide,⁴⁰ and bulk micromachining to fabricate suspended microchannels with a wall thickness of 800 nm and a fluid layer thickness of 1.2 μm. Channels of nearly 1 mm in length were completely released in less than 2Oh with a yield of 80%. An electron micrograph of three early - prototype suspended microchannels is shown in Figure 11. While this approach led to a successful demonstration, the PDMS packaging process was tedious and the overall system was delicate, unstable, and difficult to reproduce.

To address these limitations, the Manalis lab has established a partnership with Innovative Micro Technology (IMT) to implement a packaging process based on full-wafer, bonded glass micro fluidics. The SMR devices are fabricated at MIT and then sent to IMT for packaging and dicing. Packaging involves the fabrication of a capping wafer which is a glass wafer containing etched channels for fluidic delivery to the SMR (bypass channel), etched cavity for isolating the SMR in vacuum, and patterned metal electrodes to electrostatically drive the SMR. Glass has been chosen for the microfluidics capping wafer since it is optically transparent, chemically inert, and highly robust. The capping wafer also contains ~10 μm tall standoffs such that when it is bonded to the device wafer in vacuum with a glass frit sealing, the device - capping wafer separation is well controlled. Figure 12 shows completed devices that were made with this process. Once bonded, the dies are robust and can be handled without special care. All bond pads for electrical contacts are placed on the glass lid and are exposed when the silicon wafer is diced. We found that the quality factor for fluid filled vacuum encapsulated devices ranged from 300-700 which indicates an ambient pressure of a few Torr. We also found that the Q did not depend on whether the microchannel was filled with air, water, or alcohol. This indicates that energy loss due to viscous damping from fluid is negligible and that sensitivity will not be degraded by detection in the aqueous environment. To measure the SMR resonant frequency in real-time, the output of the optical lever sensor is amplified, filtered, and connected to the electrostatic drive electrode. This feedback loop ensures that the resonator is continuously driven at its resonant frequency which can be readily measured with a standard frequency counter. In this configuration, an Agilent HPLC pump and autosampler is used to maintain continuous buffer flow through the resonator and to provide systematic injections of reagents for both functionalizing and delivering target analytes. To calibrate the resonator, we varied the volumetric mass density by injecting solutions with different concentrations of NaCl (Figure 13a) and plotted the resulting frequency shift as a function of the concentration (Figure 13b). This measurement revealed a volumetric mass resolution in the range of ~10^~6 g/cm³ which is comparable to the best commercially available densitometer from Mettler Toledo.

To demonstrate detection of specific binding, we first functionalized the surfaces with avidin through an approximately 1 minute injection with the autosampler; the resonant frequency subsequently dropped by about 2 Hz from an initial resonant frequency of 32.8 kHz (Figure 13c). During the avidin injection, the frequency shift revealed not only nonspecific binding of avidin to the surface, but also the volumetric mass density change from the avidin. Once the avidin was rinsed out with buffer, the resonant frequency increased slightly and the net shift revealed only the addition of avidin to the surface. Subsequent injections of avidin revealed only transient frequency shifts due to the volumetric mass change before returning to the initial base line. Next, we injected biotinylated Bovine Serum Albumin (b-BSA) and observed a further decrease in resonant frequency due to specific binding of the b-BSA to the avidin functionalized surface; similar controls were subsequently performed. Finally, we injected avidin and again observed a further decrease in resonant frequency as it attached to the immobilized b-BSA. We have found that the results shown in Figure 4 are repeatable, consistent and straightforward to achieve.

A distinguishing quality of the suspended microchannel resonator (SMR) with respect to other "label-free" sensors such as the quartz crystal microbalance (QCM) and surface plasmon resonance (SPR) is its ability to achieve ultra-high intrinsic sensitivity. To date, our measurements and calculations are based on the metric of mass per area that is adsorbed to the sensor surface. Recent experiments with the MIT/IMT devices yield a sensitivity that approaches 10^~18 g/μm² (equivalent to 10 proteins/μm² assuming 100 kD molecular weight). This corresponds to a total mass resolution of -100 femtograms for a surface area of 10⁵ μm². We predict that the sensitivity will improve by an order of magnitude with further refinements in the device fabrication, displacement sensor and frequency detection circuitry. Since this metric is based only on the intrinsic properties of the sensor and is independent of assay conditions, it is therefore very useful for comparing the SMR sensitivity to other label-free and label-dependent platforms: the SMR is two orders of magnitude better than the QCM, one order of magnitude better than the SPR and approximately equivalent to fluorescent readers for microarrays.

Of equal importance to a biosensor's sensitivity is its selectivity. For biomarker detection, the SMR will be required to operate in the presence of complex mixtures and will therefore be subjected to nonspecific binding. However, since very few antibodies are truly monospecific when used to probe biomarkers in complex mixtures, all affinity-based biosensors are susceptible to nonspecific binding. The magnitude depends on assay parameters such as surface attachment chemistry, IQ of the antibody-biomarker complex, and complexity of the sample. To reduce signal degradation from nonspecific binding, most immunoassays use both primary and secondary antibodies to detect the biomarker through a sandwich assay. Such assays achieve exquisite selectivity because the specificities of two different antibodies are exploited, and very rarely will the capture and detection antibodies bind to the same extraneous protein. In our program, we will also use the sandwich assay in order to achieve a selectivity that is comparable to immunoassays that are routinely used today.

Our primary objective is to integrate the nanofluidic concentrator (NC) with the suspended microchannel resonant (SMR) detector and apply the system to cancer biomarker analysis. The resulting system will enable fundamentally new types of measurements that would not be possible to achieve with the isolated components. As described above, the NC is several orders of magnitude more efficient than conventional concentrators, and the SMR is an order of magnitude more sensitive than conventional label-free detectors. However, each component, as an isolated device, has a critical limitation when it comes to detecting biomarkers. For the NC, the volume of the concentrating region is -10-100 picoliters and this makes off-chip affinity detection difficult to achieve without diluting the sample. For the SMR (as well as other affinity biosensors), the capture rate for biomarker concentrations significantly below the IQ can be very slow and the resulting time required for detection can be prohibitively long. However, these limitations are entirely eliminated when the NC is integrated with the SMR. With an integrated device, a sample containing a dilute biomarker concentration can be concentrated to well above the Kd at a rate that is orders of magnitude faster than the surface adsorption rate. The resulting concentrate will be complex in that it will contain many other types of proteins besides the biomarker. When the concentrate is transported through the SMR, the biomarker along with some of the other proteins will be quickly captured by the surface and the output of the SMR will reflect a combination of specific and nonspecific binding. To obtain high specificity, a secondary antibody (similar to ELISA) will be transported through the SMR. Since the SMR provides a direct measure of mass, the resulting output will indicate the number of adsorbed secondary antibodies, and hence, the number of biomarkers. Since the concentrator is linear with respect to time, the initial biomarker concentration can be determined directly. Based on our preliminary results, a concentration increase of 10⁷ in a 50 pico liter volume can be achieved in 1 hour and the SMR can resolve a mass near 100 femtograms. Therefore, it should be possible to detect an initial biomarker concentration below 1 pg/mL. We anticipate that lower concentrations can be detected by concentrating for a longer period.

In addition to high sensitivity, there are two major advantages of the integrated system over existing methods. First is increased selectivity: the sample can be purified extensively with off-chip methods such as affinity columns and gels in order to remove biomolecules that will degrade affinity detection. Although the biomarker concentration of the purified sample will be diluted, the effective "signal" will be recovered by using a high gain in the concentration stage. Second is increased dynamic range: for many situations, the initial biomarker concentration from patients can vary by orders of magnitude. For existing methods, the dynamic range is limited by the sensitive range of the dose-response curve. However in the case of the integrated system, it is possible to use the SMR to directly monitor the output of the concentrator in real-time and essentially achieve a closed-loop detection system. In contrast, chemical amplification strategies used by ELISA or related assays have to be carefully optimized for each binding step in order to preserve the linearity of the detected signal.

The design for the integrated system consists of a concentrator channel, nanochannel filter, and suspended microchannel resonant detector. The channel height of the concentrator and suspended channel is ~1 μm while the height of the nanochannels will be ~40 nm. All channels will be fabricated will a sacrificial process as described in Reference 37. Fluids can be delivered to these channels with high flow rates by tall U-shaped bypass channels. A series of 3D illustrations of the integrated system is shown in Figure 14 and 2D illustrations of the fabrication process are shown in Figure 15.

The walls of the concentrator channel will be passivated by either BSA or PEG. We have already demonstrated that the concentration process can be achieved even when the nanochannels are coated with polyacrylamide, which is commonly used to prevent nonspecific binding. However, the operating potential, buffer ionic strength, and nanochannel height must be optimized in order to obtain the most efficient concentration.

Once the channels are passivated, a series of reagents will be transported through the inlet and outlet of the "SMR functionalization" channel in order to attach the capture molecules to the sensor surface. Although the channel surface is silicon nitride, it has been shown that the outer layer of silicon nitride becomes partially oxidized, thus presenting silanol groups with a density similar to that of a SiO₂ surface.⁴¹ This will allow us to adapt pre-existing functionalization processes that are routinely used by DNA and protein microarrays. We will generally follow a three-step procedure for validating the surface attachment chemistry for a particular biomarker: i) obtain a dose-response curve in a glass capillary with fluorescent readout and find assay conditions for maximum sensitivity and selectivity, ii) obtain a dose-response curve in the SMR device (without the concentrator, see Figure 12) with mass readout and verify that assay conditions are optimal, and iii) implement assay in the integrated device shown in Figure 15.

Our specific procedure for surface attachment of biomarkers will be based on methods already developed by the Manalis and Sorger labs for detecting Akt, a protein involved in signaling events in the ErbB signaling network. The continuous flow procedure for silanizing and covalently attaching antibodies to the glass capillary is as follows: plain capillaries are cleaned in 100% ethanol for 10 minutes and then treated with a 3% solution of 3- aminopropyltriethoxysilane in 95% ethanol for 1 hour. The capillaries are then briefly washed in 100% ethanol and dried with nitrogen to remove excess silanol. The absorbed silane layer is cured at 115 C for 1 hour. After cooling to room temperature, the capillaries are washed for 10 minutes in 95% ethanol to remove uncoupled reagent. BSA is immobilized on the surface of the aminosilane-coated capillaries using 100 mM N,Nl-disuccinimidyl carbonate and 100 mM N ,N- diisopropylethylamine. Carboxylates on the surface of BSA are subsequently activated with N- hydroxysuccinamide, as described previously.⁴² The activated BSA capillaries can be stored in a desiccator under vacuum at room temperature for up to one month without noticeable loss of activity.

The activated BSA passivates the glass surface and is used to covalently capture antibodies or other proteins via its activated aspartate and glutamate residues. In the example of Akt detection, a recombinant human anti-Akt antibody was diluted in PBS at 0.5 mg/ml, introduced into a series of capillaries, left at room temperature to dry overnight, and then placed in the cold room in a humidified chamber. Following protein immobilization, the glass surface was quenched with 1% ethanolamine in dF^O for 5 minutes and then passivated for an additional 30 minutes with a standard blocking agent.

To test the derivatized capillaries, the target protein (in this case, human protein Aktl) is serially diluted from 1000 ng/ml to 1 ng/ml. The antigen is fluorescently labeled by incubating for 30 minutes with 1 : 100 polyclonal secondary rabbit anti-Akt antibody and with 1 :200 fluorescent goat anti-rabbit IgG-PE. The mixture is introduced into the capillary and rinsed with 0.1 % Tween to remove unbound reagents and was visualized using an inverted fluorescent microscope (Figure 16). For the quantification of the fluorescent intensity in the capillaries, the mean intensity inside each capillary is calculated.

Our procedure for functionalizing the SMR will follow the same protocol as described above. The main difference is that liquids will be introduced using a constant pressure controlled fluidic system rather than a constant flow rate controlled system, as in the capillaries. This pressure induced fluidic system has been tested using a custom-made system of pistons that allows for the precise monitoring of the pressure of the inlet and outlet to the suspended channel. The outlet can be used to deliver different reagents whereas the inlet can be used to deliver washing buffers. By simply differentiating the pressures between the inlet and outlet, the suspended channel can be serially exposed to different reagents and washing media. To test the fluidic system, functionalization of our existing SMR devices have been carried out under an upright microscope. The flow of the reagents could easily be observed and controlled, and no clogging was observed even when cell lysate was used.

We will use either electro osmotic or pressure driven flow to drive the sample through the concentrator. When a voltage is applied across the nanochannel filter, the ionic conductivity of the concentrator channel is significantly reduced in the vicinity of the nanochannels. As a result, incoming sample molecules are trapped at the high/low conductivity boundary and the concentration in this region increases. We have found from previous experiments that this region is -100 μm upstream of the nanochannels and that its exact location along the concentrator channel can be adjusted by tuning the voltage across the nanochannels. Once the sample has been concentrated for the desired time, it is directly transported to the SMR by applying a negative pressure to the SMR output channel. The negative pressure will ensure that the concentrate is transported only into to the SMR and not through the concentrator channel.

The concentrator will need to accumulate a total biomarker mass of approximately 1 pg (10⁷ biomarkers assuming 10OkD molecular weight) in order to provide the SMR with a detectable amount of mass. A critical parameter for the detection process will be the flow velocity (as determined by the negative pressure on the SMR output) of the concentrate within the suspended channel. If the velocity is too slow, the majority of the concentrate will be captured by the receptors located at the input to the SMR and only a small portion will adsorb at the apex of the SMR where the mass sensitivity is greatest. If the velocity is too fast, the majority of the concentrate will not have enough time to be captured before it exits the SMR. Thus, there will be a velocity for which the mass resolution of the SMR will be optimal and it will depend on the IQ, surface density of capture probes, and SMR channel height. Through collaboration with the Jensen group (MIT ChemEng), the transport dependent binding kinetics in the SMR has been modeled extensively and we have used these models to verify that a high collection efficiency can be achieved from flow velocities that result from pressures below 1 atm (Figure 17). Since the capture probe density is not always a known parameter, we will need to conduct calibration measurements to empirically determine the optimal velocity and related pressure. Given that the concentration process will produce ~10⁷ molecules, we do not expect the surface binding sites to be saturated (as verified by simulation shown in Figure 17). One approach is to analyze a sample several times in either a serial or parallel format (Figure 18). In each analysis, the sample will be concentrated for an increasing time period (e.g. 1, 10, and 100 seconds). For the shorter times, the concentrate will not contain enough molecules to be resolved by the SMR; thus the SMR output will remain unchanged after the concentrate is injected into the suspended channel. However, once the NC has increased the sample concentration to a value near the IQ, the subsequent injection of the concentrate will be detected as it rapidly adsorbs to the SMR sensor surface. The resulting signal will reflect the biomarker adsorption plus nonspecific binding from other proteins in the concentrate. To isolate the specific signal, the secondary antibody for the biomarker can be injected through the suspended channel output. If the secondary antibody is not available, then it will be necessary to make a differential measurement in order to reduce signal degradation from nonspecific binding. In this scenario, the sample will be delivered to a reference system where the SMR sensor surface is either passivated, or functionalized with capture molecules that are known to not be specific with the biomarker. The signal from nonspecific molecules that have equivalent affinity for the reference and active sensors will not appear in the differential output. We have already successfully demonstrated this approach with other types of label-free microsensors. For example, we have shown that a single base mismatch within 12mer oligonucleotides can be distinguished electronically by using a differential pair of silicon field-effect sensors.⁴³ While the single-ended output revealed large signals from nonspecific binding, the signals were completely canceled in the differential output. In another example, we showed that a micromechanical stress sensor that inherently suppresses background effects can achieve a differential detection limit that is up to an order of magnitude lower than the single-ended limit in the low-frequency range of 0.0003-1 Hz where many types of biologically relevant reactions occur.⁴⁴'⁴⁵ This allowed us to measure the concentration of a specific protein in the presence of a cell lysate.

We plan to investigate the use of gold nanoparticle labeling of the secondary antibody for improving the biomarker sensitivity of the SMR. Since the mass density of gold is significantly higher than fluid, the effective mass change from a nanoparticle labeled antibody will be higher than from a unlabeled antibody. For instance, labeling with a 10 nm nanoparticle diameter would improve the SMR sensitivity by two orders of magnitude. In this case, the concentrator would need to accumulate a biomarker mass of only 10 femtograms, or 10⁵ biomarkers. In addition to sensitivity enhancement through gold labeling, it should be possible to use any chemical amplification strategy that is currently used by ELISA-related assays. Thus, this approach is not central to the aims of this invention and will therefore not be pursued with high priority.

We will validate the system sensitivity and dynamic range by concentrating and detecting pure samples of GFP. The concentrator channel will initially be passivated with either PEG or BSA. As we have previously demonstrated (Figure 19), the SMR will be functionalized with avidin in order to bind biotinylated anti-GFP for the affinity capture of GFP. Initial GFP concentrations ranging from 0.1 pg/mL to 100 ng/mL will be concentrated for periods ranging from ~1 to 10⁴ seconds by the protocol illustrated in Figure 18. While in principle it should be possible to obtain a direct count of bound GFP molecules based purely on the resonant frequency shift of the SMR, the spatially non-uniform absorption of GFP (as revealed in Figure 17) will require that a scaling factor be known. This factor can be determined by conducting a series of absorption measurements from samples of known concentration. Since the pressure driven flow that transports the concentrate into the SMR will result in a highly repeatable flow velocity, the calibration curve can be used for all subsequent assays with GFP. In general, we plan to acquire a systems-level calibration curve for biomarker assays that we discuss below.

While GFP validation will provide sensitivity and dynamic range metrics that are necessary for device optimization, it will not be useful for assessing selectivity and performance for real-world assays where the biomarker must be detected within a complex mixture such as serum. In order to validate these aspects, we will determine the sensitivity and dynamic range for the detection of prostate specific antigen (PSA) in the presence of human and mouse serum. We will develop sandwich assays to detect both free and complexed PSA. Using the previously described procedure for silanating the SMR sensor surface, we will covalently attach mouse monoclonal antibodies for the affinity capture of PSA. Samples will contain various concentrations of PSA within a background of 10% serum. Once the sample has been completely concentrated, transported through the SMR and rinsed with buffer, polyclonal anit- PSA antibodies will be delivered to the SMR sensor surface through the suspended channel output. We will measure absorption for both PSA (specific + nonspecific binding) and the polyclonal anti-PSA (specific binding) as a function of the serum concentration in order to determine how sensitivity and dynamic range depend on sample complexity. The presence of background components at high initial levels will likely limit the attainable concentration gain. We will therefore investigate the effectiveness of sample purification methods such as gel electrophoresis and affinity separation for improving the sensitivity and dynamic range of the integrated detection system.

Before conducting PSA assays with the integrated system, we will validate the concentrator and detector as individual components. For the detector, we will obtain a dose- response curve for the direct binding of PSA and for the binding of the polyclonal secondary anti-PSA antibody. The dose-response curve will be measured for both purified PSA and for PSA spiked in samples containing various concentrations of serum. Our goal for these measurements will be to determine the limit of detection and verify the PSA binding kinetics as a function of concentration. For the concentrator, visualization of PSA aggregation in the nanofluidic concentrator device will be achieved by labeling PSA with fluorescence dyes (such as Cy-series), which are known not to change the pi and the molecular weight value of the target significantly) in order to characterize the PSA accumulation in the nanofluidic concentrator. PSA-spiked mouse serum will be processed by standard sample preparation protocols in order to obtain a relatively purified sample. However, we expect that even after 2D gel separation and/or additional purification steps, the sample might contain many many different molecular species. The standard gel electrophoresis (either ID or 2D) typically generates the sample volume of ~20μl, which will be used as our starting sample. Since the clinical concentration range for PSA is relatively high (more than IpM) compared with other, more challenging biomarkers, the main focus will be set on verifying the linearity of the PSA collection in the nanofluidic concentrator in realistic situation (serum background), Concentrated molecules can be easily detected when they are fluorescently labeled, and the quantification of the collected PSA can be done as a function of collection time and other experimental parameters. This test will clearly validate the usability of the nanofluidic concentrator for analyzing biomarkers in serum samples, even over a background of several other molecules with similar pi and size. While it is possible that the nanofluidic concentrator might concentrate both the background molecules and the target biomarkers, the kinetics of the biomarker-antibody is much more favorable when the concentration of biomarker is increased near the IQ value. Verifying linearity of the nanofluidic concentrator is important for detecting multiple biomarkers with different gain settings (collection times). Therefore, PSA-spiked mouse serum with PSA concentrations down to -IfM (~lpg/mL) will be made, and will be used to verify the linearity of the concentration, and the long-term stability. While the -IfM PSA concentration might not be relevant in diagnosing prostate cancer, this will verify the linearity of the nanofluidic concentration process with realistic samples which we will encounter in subsequent years when detecting other biomarkers.

The goal of this aim is to develop electronic readout for the SMR arrays and use the system to detect four biomarkers in parallel. The electronic readout system will consist of low- noise capacitance circuitry that will be integrated with the NC/SMR system. In the first section of this aim, we provide a detailed description of the readout circuitry design, and in the second section, we describe the multi-biomarker assay.

Displacements of microcantilevers (such as those used with the atomic force microscope) are typically detected optically by bouncing a laser beam off the cantilever and detecting its position with a photo-sensitive detector (Figure 20a). However it is difficult to use this approach for cantilever arrays since multiple laser beams and detectors must be carefully aligned for each cantilever. Electronic readout is highly scalable, suitable for mass production, and extremely robust. Since we are already using an integrated electrode to electrostatically resonate the suspended channel, electronic readout can be readily achieved by connecting the electrode to low-noise circuitry that is designed for detecting tiny changes in capacitance (Figure 20b).

Given the dimensions of the SMR device structure, the capacitance of the metalized suspended channel and adjacent drive electrode is 100 fF. Our frequency detection limit, as established by optical readout, is approximately 1 mHz for a 40 kHz resonance. To achieve this detection limit by electrical readout, we will need to develop low noise circuitry that detects the channel capacitance with a precision of 1 part in 10⁵. Such a detection level has already been achieved in industry and academic research groups. For instance, integrated circuitry for low- cost Analog Devices micro-gyroscopes can resolve capacitance changes as small as 10 zeptoFarads which corresponds to a position resolution of 10^"4 Angstroms. At MIT, the Sarpeshkar group has designed and validated circuitry that achieves similar metrics for detecting displacement of micro-accelerometers. The Manalis group is currently collaborating with the Sarpeshkar group to implement such circuitry for the SMR. The circuitry will be fabricated in a standard CMOS process by MOSIS integrated circuit fabrication services.

The capacitance readout will be incorporated in our existing scheme that uses feedback to maintain a constant oscillation frequency of the suspended channel. With feedback, the output is altered by taking a specific portion of the system's forward transfer characteristics back to the input. This eliminates the need for an input signal and can greatly increase the performance of an oscillator since it is always driven exactly at its resonant frequency. Using feedback with optical readout, we can measure a 40 kHz SMR with a resolution of 1-10 mHz in a 1 second averaging time. Our goal for capacitance readout will be to achieve a similar metric.

Since only one electrode will be used to both excite and sense the suspended channel at resonance, we will modulate the channel and on-chip reference capacitor at 1 MHz. The resulting signal at the 'sense node' will be demodulated with an analog multiplier in order to reveal the relative low-frequency vibration (40 kHz) of the channel. This signal will then be added to the 1 MHz modulation and delivered back to the channel and reference capacitor in order to close the feedback loop (as depicted in Figure 21). A second analog multiplier will be used outside of the feedback loop in order to demodulate the channel vibration and produce a signal that reveals the vibration frequency. A key attribute of demodulation, or lock-in, approaches is that low-frequency noise from electrical coupling and 1/f noise from the circuitry is eliminated; Figure 22 illustrates this attribute by showing the signals in the frequency domain.

In order for our circuitry to achieve optimal performance, we will address two main design issues: i) CMOS multipliers can be driven into a non- linear regime by the presence of a DC offset on either positive or negative input. Any slight inaccuracy in one of the inputs can result in an unpredicted change in the output. This could result from DC offset in the previous stage or even mismatches internal to the multiplier. We will address this problem by implementing a nonlinear feedback technique previously validated by the Sarpeshkar lab. ii) the low capacitance signal from the suspended channel is effectively reduced by parasitic capacitances between the sensing node and ground. By placing the electronics on a chip connected to the SMR, we estimate a parasitic capacitance of approximately 200 fF will exist at the sensing node. To reduce signal degradation from the parasitic capacitance, additional feedback (see Figure 23) will be used at the front-end to ensure that the loop gain is independent of the parasitic capacitance.

In addition to PSA and thymosin βl5 which are described above, we will also employ the Prostate Specific Membrane Antigen (PSMA)⁴⁶ and chromogranin A, both of which have been shown to be elevated in both serum and urine of prostate cancer patients compared to normal controls.⁴⁷ These have been chosen because each of these markers circulates in the bloodstream and can be detected in prostate cancer patient serum and urine. They have also been selected in a recent study that evaluated 91 potential prostate cancer markers and selected the five most promising for clinical development.⁴⁸ According to these authors, an ideal biomarker for the detection of prostate cancer would be able to be "prostate-specific, detectable in an easily accessible biological fluid such as human serum, urine, or prostatic fluid, and able to distinguish between normal, BPH, prostatic intreaepithelial neoplasia [PIN] and cancerous prostate tissues." In addition, each of these four markers has a dedicated ELISA assay developed which can be used to confirm the levels of analyte present in the biological samples. The ELISAs for each of these markers are described in the following references.⁴⁹ Chromogranin A has the additional benefit of being able to detect the presence of the neuroendocrine form of prostate cancer which is not always well detected by other markers. In most cases the detection sensitivity of these assays is on the order of lng/ml. To implement the assay in NC/SMR array system, we will use similar antibodies and assay conditions as are currently used in the ELISA. The performance of the integrated system can then be benchmarked against the ELISA. In later iterations, it may be possible to use lower affinity capture methods such as peptides or single chain antibodies that are generated by phage display methods. As we do not expect our sensor to require the same high affinity found in an antibody/antigen reaction, it may be possible to use peptides which are considerably faster and easier to generate than high affinity antibodies.

Our ultimate goal is to the use the parallel system with the feedback approach (Figure 18) for measuring the abundance of the four biomarkers. Based on the design and operation described above, the system would be capable of detecting biomarkers over a wide range of initial concentrations using receptor molecules that have either high or low binding affinity constants. This concept is illustrated in Figure 24. References

1. M. R. Cooperberg, D. P. Lubeck, S. S. Jehta, P. R. Carroll, "Time trends in clinical risk stratification for prostate canceπlmplications for outcomes." J. Urol; 170 :S21-25 (2003).

2. M.J. Barry "Prostate-specific antigen testing for early diagnosis of prostate cancer." N. Engl J Med, 344, 1373 (2001).

3. F.H. Schroder and R. Kranse, "Verification bias and the prostate-specific antigen test - Is there a case for a lower threshold for biopsy?" N Engl J Med;349:393-395 (2003).

4. A.V. D'Amico, M.H. Chen, K.A. Roehl and W.J. Catalona, "Preoperative PSA velocity and the risk of death from prostate cancer after radical prostatectomy" N Engl J Med 351,12-135 (2004).

5. Jung K, Elgeti U, Lein M, Brux B, Sinha P, Rudoplh B, et al: "Ratio of free or complexed prostate-specific antigen (PSA) to total PSA: which ratio improves differentiation between benign prostate hyperplasia and prostate cancer?" Clin Chem. 46, 55-62 (2000)

6. B. Schweitzer, S. Wiltshire, J. Lambert, S. O'Malley, K. Kukanskis, Z. Zhu, S. F. Kingsmore, P. M. Lizardi, and D. C. Ward, "Immunoassays with rolling circle DNA amplification: A versatile platform for ultrasensitive antigen detection," Proc. Natl. Acad. ScL U. S. A., vol. 97, pp. 10113-10119, 2000. http://dx.doi.org/10.1073/pnas.170237197

7. J.-M. Nam, C. S. Thaxton, and C. A. Mirkin, "Nanoparticle-Based Bio-Bar Codes for the Ultrasensitive Detection of Proteins," Science, vol. 301, pp. 1884-1886, 2003;

D. G. Georganopoulou, L. Chang, J.-M. Nam, C. S. Thaxton, C. E. Mufson, W. L. Klein, and C.

A. Mirkin, "Nanoparticle-based detectioni in cerebral spinal fluid of a soluble pathogenic biomarker for Alzheimer's disease," Proc. Natl. Acad. Sci. U. S. A., vol. 102, pp. 2273-2276, 2005. http://dx.d0i.0rg/l 0.1073/pnas.0409336102

8. J. Travis and P. Pannell, "Selective removal of albumin from plasma by affinity chromatography.," Clin. Chim. Acta, vol. 49, pp. 49-52, 1973;

Bjorck and G. Kronvall, "Purification and some properties of streptococcal protein G, a novel IgG-binding reagent.," J. Immunol., vol. 133, pp. 969-974, 1984. IgG removal;

B. Akerstrom, T. Brodin, K. Reis, and L. Bjorck, "Protein G: a powerful tool for binding and detection of monoclonal and polyclonal antibodies.," J. Immunol., vol. 135, pp. 2589-2592, 1985;

B. Guss, M. Eliasson, A. Olsson, M. Uhlen, A. K. Frej, H. Jornvall, J. I. Flock, and M. Lindberg, "Structure of the IgG-binding regions of streptococcal protein G.," EMBO J., vol. 5, pp. 1567- 1575, 1986; N. I. Govorukhina, A. Keizer-Gunnink, A. G. J. van der Zee, S. de Jong, H. W. A. de Bruijn, and R. Bischoff, "Sample preparation of human serum for the analysis of tumor markers: Comparison of different approaches for albumin and [gamma] -globulin depletion," J. Chromatogr. A, vol. 1009, pp. 171-178, 2003.

9. A. G. Palmer and N. L. Thompson, "Molecular aggregation characterized by high order autocorrelation in fluorescence correlation spectroscopy," Biophys. J., vol. 52, pp. 257- 270, 1987. http://www.biophysj.Org/cgi/content/abstract/52/2/257

D. S. Burgi and R. -L. Chien, "Optimization in Sample Stacking for High-Performance Capillary Electrophoresis," Anal. Chem., vol. 63, pp. 2042-2047, 1991

R. -L. Chien and D. S. Burgi, "Sample Stacking of an Extremely Large Injection Volume in High-Performance Capillary Electrophoresis," Anal. Chem., vol. 64, pp. 1046-1050, 1992

C-X. Zhang and W. Thormann, "Head-Column Fielid-Amplified Sample Stacking in Binary System Capillary Electrophoresis: A Robust Approach Providing over 100-Fold Sensitivity Enhancement," Anal. Chem., vol. 68, pp. 2523-2532, 1996

J. Lichtenberg, E. Verpoorte, and N. F. d. Rooij, "Sample preconcentration by field amplification stacking for microchip-based capillary electrophoresis," Electrophoresis, vol. 22, pp. 258-271, 2001

10. P. Gebauer and P. Bocek, "Recent progress in capillary isotachophoresis," Electrophoresis, vol. 23, pp. 3858-3864, 2002

11. A. K. Singh, D. J. Throckmorton, B. J. Kirby, and A. P. Thompson, "A Novel Miniaturized Protein Preconcentrator Based on Electric Filedaddressable Retention and Release," presented at Micro Total Analysis Systems, 2002.

J. Astorga- Wells and H. Swerdlow, "Fluidic Preconcentrator Device for Capillary Electrophoresis of Proteins," Anal. Chem., vol. 75, pp. 5207-5212, 2003. http://dx.doi.org/10.1021/ac0300892

S. -R. Park and H. Swerdlow, "Concentration of DNA in a Flowing Stream for High- Sensitivity Capillary Electrophoresis," Anal. Chem., vol. 75, pp. 4467-4474, 2003. http://dx.doi.org/10.1021/ac034209h

J. Astorga- Wells, T. Bergman, and H. Jornvall, "Multistep Microreactions with Proteins Using Electrocapture Technology," Anal. Chem., vol. 76, pp. 2425-2429, 2004. http://dx.doi.org/10.1021/ac0354342

Q. Wang, B. Yue, and M. L. Lee, "Mobility-based selective on-line preconcentration of proteins in capillary electrophoresis by controlling electroosmotic flow," J. Chromatogr. A, vol. 1025, pp. 139-146, 2004 12. J. P. Quirion and S. Terabe, "Exceeding 5000-Fold Concentration of Dilute Analytes in Micellar Electrokinetic Chromatography," Science, vol. 282, pp. 465-468, 1998

J. P. Quirion and S. Terabe, "Approaching a Million-Fold Sensitivity Increase in Capillary Electrophoresis with Direct Ultraviolet Detection: Cation-Selective Exhaustive Injection and Sweeping," Anal. Chem., vol. 2000, pp. 1023-1030, 2000

M. Molina and M. Silva, "Micellar electrokinetic chromatography: Current developments and future," Electrophoresis, vol. 23, pp. 3907-3921, 2002

13. R. D. Oleschuk, L. L. Shultz-Lockyear, Y. Ning, and D. J. Harrison, "Trapping of Bead- Based Reagents within Micro fluidic Systems: On-Chip Solid-Phase Extraction and Electrochromatography," Anal. Chem., vol. 72, pp. 585-590, 2000. http://dx.doi.org/ 10.1021 /ac990751 n

C. Yu, M. H. Davey, F. Svec, and J. M. J. Frechet, "Monolithic Porous Polymer for On-Chip Solid-Phase Extraction and Preconcentration Prepared by Photoinitiated in Situ Polymerization within a Microfluidic Device," Anal. Chem., vol. 73, pp. 5088-5096, 2001. http://dx.doi.org/] 0.1021/acOl 06288

B. S. Broyles, S. C. Jacobson, and J. M. Ramsey, "Sample Filtration, Concentration, and Separation Integrated on Microfluidic Devices," Anal. Chem., vol. 75, pp. 2761-2767, 2003

D. L. Huber, R. P. Manginell, M. A. Samara, B.-I. Kim, and B. C. Bunker, "Programmed Adsorption and Release of Proteins in a Microfluidic Device," Science, vol. 301, pp. 352-354, 2003. http://www.scicncemag.org,^/cgi/contcnt/abstractβO 1/5631/352

K. W. Ro, W.-J. Chang, H. Kim, Y.-M. Koo, and J. H. Hahn, "Capillary electrochromatography and preconcentration of neutral compounds on poly(dimethylsiloxane) microchips," Electrophoresis, vol. 24, pp. 3253-3259, 2003

14. J. Khandurina, S. C. Jacobson, L. C. Waters, R. S. Foote, and J. M. Ramsey, "Microfabricated Porous Membrane Structure for Sample Concentration and Electrophoretic Analysis," Anal. Chem., vol. 71, pp. 1815-1819, 1999

S. Song, A. K. Singh, and B. J. Kirby, "Electrophoretic Concentration of Proteins at Laser- Patterned Nanoporous Membranes in Microchips," Anal. Chem., vol. 76, pp. 4589-4592, 2004. http://dx.d0i.0rg/l 0.1021 /acO497151

15. G. Wu, R.H. Datar, K.M. Hansen, T. Thundat, R.J. Cote, and A. Majumdar, "Bioassay of prostate-specific antigen (PSA) using microcantilevers," Nature Biotech. 19 856 (2001).

16. Y. Cui, Q. Wei, H. Park, and CM. Lieber, "Nanowire nanosensors for highly sensitive and selective detection of biological and chemical species," Science 293, p. 1289 (2001). 17. J. Fritz, E. B. Cooper, S. Gaudet, P. K. Sorger, and S.R. Manalis, "Electronic detection of DNA by its intrinsic molecular charge," Proceedings of the National Academy of Sciences, 99 14142 (2002).

18. P. Bergveld, "The future of biosensors," Sensors and Actuators A, 56 p. 65 (1996).

19. Bao L, et al. "Thymosin βl5: A novel regulator of tumor cell motility upregulated in metastatic prostate cancer." Nature Medicine;2: 1322-1328 (1996).

20. Bao L, Loda M, Zetter BR., "Thymosin βl5 expression in tumor cell lines with varying metastatic potential." Clin Exptl Metastasis;16:227-233 (1998).

21. Hutchinson L, Chang EL, Becker CM, Shih MC, Brice M, DeWoIfWC, Gaston SM, Zetter BR. Thymosin bl5: a prostate cancer urinary biomarker. Prostate. 2005 Online.

22. Landers KA, Burger MJ, Tebay MA, Purdie DM, Scells B, Samaratunga H, Lavin MF, Gardiner RA. "Use of multiple biomarkers for a molecular diagnosis of prostate cancer." Int J Cancer, May 10;l 14(6):950-6 (2005)

23. Hutchinson LM, Chang EL, Becker CM, Ushiyama N, Behonick D, Shih MC, DeWoIf WC, Gaston SM, Zetter BR. "Development of a sensitive and specific enzyme-linked immunosorbent assay for thymosin b 15, a urinary biomarker of human prostate cancer," J Clin Chem. Submitted. (Tb-15) paper.

24. Y. -C. Wang and J. Han, "Million-Fold Biomolecule Pre-concentration by Nanofluidic Electrokinetic Trapping," presented at 18th International Symposium on MicroScale Bioseparations (MSB), New Orleans, LA, 2005.

Y. -C. Wang, A. L. Stevens, and J. Han, "Million-fold Preconcentration of Proteins and Peptides by Nanofluidic Filter," Anal. Chem., vol. submitted, 2005

25. P. Mao and J. Han, to be published.

26. N. A. Mishchuk and P. V. Takhistov, "Electroosmosis of the second kind," Colloids and Surfaces A: Physicochemical and Engineering Aspects, vol. 95, pp. 119-131, 1995.

27. Y. Li, T. Pfohl, J. H. Kim, M. Yasa, Z. Wen, M. W. Kim, and C. R. Safmya, "Selective Surface Modification in Silicon Microfluidic Channels for Micromanipulation of Biological Macromolecules," Biomedical Microdevices, vol. 3, pp. 239-244, 2001

28. S. Hjerten, "High-Performance Electrophoresis Elimination of Electroosmosis and Solute Adsorption," J. Chromatogr., vol. 347, pp. 191-198, 1985

29. A. C. Henry, T. J. Tutt, M. Galloway, Y. Davidson, C. S. McWorter, S. A. Soper, and R. L. McCarley, "Surface Modification of Poly(methyl methacrylate) Used in the Fabrication of Microanalytical Devices," Anal. Chem., vol. 72, pp. 5331-5337, 2000 30. D. C. Duffy, J. C. McDonald, O. J. A. Schueller, and G. M. Whitesides, "Rapid Prototyping of Micro fluidic Systems in Poly(dimethylsiloxane)," Anal. Chem., vol. 70, pp. 4974-4984, 1998

31. B. A. Grzybowski, R. Haag, N. Bowden, and G. M. Whitesides, "Generation of Micrometer-Sized Patterns for Microanalytical Applications Using a Laser Direct- Write Method and Microcontact Printing," Anal. Chem., vol. 70, pp. 4645-4652, 1998

32. S. Hu, X. Ren, M. Bachman, C. E. Sims, G. P. Li, and N. Allbritton, "Surface Modification of Poly(dimethylsiloxane) Micro fluidic Devices by Ultraviolet Polymer Grafting," Anal. Chem., vol. 74, pp. 4117-4123, 2002

33. S. L. R. Barker, M. J. Tarlov, H. Canavan, J. J. Hickman, and L. E. Locascio, "Plastic Micro fluidic Devices Modified with Polyelectrolyte Multilayers," Anal. Chem., vol. 72, pp. 4899-4903, 2000.

34. G. Ocvirk, M. Munroe, T. Tang, R. Oleschuk, K. Westra, and D. J. Harrison, "Electrokinetic control of fluid flow in native poly(dimethylsiloxane) capillary electrophoresis devices," Electrophoresis, vol. 21, pp. 107-115, 2000

35. T. Yang, S.-y. Jung, H. Mao, and P. S. Cremer, "Fabrication of Phospholipid Bilayer- Coated Microchannels for On-Chip Immunoassays," Anal. Chem., vol. 73, pp. 165-169, 2001

36. A. Papra, A. Bernard, D. Juncker, N. B. Larsen, B. Michel, and E. Delamarche, "Micro fluidic Networks Made of Poly(dimethylsiloxane), Si, and Au coated with Polyethylene Glycol for Patterning Proteins onto Surfaces," Langmuir, vol. 17, pp. 4090- 4095, 2001

37. T.P. Burg and S.R. Manalis, "Suspended microchannel resonators for biomolecular detection," Applied Physics Letters, 83 2698 (2003).

38. W. Kauzmann, K. Moore, D. Schultz, "Protein densities from X-ray crystallographic coordinates," Nature 248, p.447 (1974).

39. J. Berenschot, N. Tas, T. Jammerink, M. Elwenspoek, and A. van den Berg, "Advanced sacrificial poly-Si technology for fluidic systems," Journal of Micromechanics and Microengineering 12, p. 621-624 (2002).

40. M. Stern, M. Geis, J. Curtin, "Nanochannel fabrication for chemical sensors, " J. Vac. Sci. Technol. B., 15 p. 2887-2891 (1997).

41. R. Raiteri, B Margesin, M. Grattarola, "Atomic force microscope estimation of the point of zero charge of silicon insulators," Sensors and Actuators B: Chemical, B46 (2): 126-32 (1998). 42. G. MacBeath, S. L. Schreiber, "Printing Proteins as Microarrays for High- Throughput Function Determination," Science 289 1760 (2000).

43. J. Fritz, E. B. Cooper, S. Gaudet, P. K. Sorger, and S.R. Manalis, "Electronic detection of DNA by its intrinsic molecular charge," Proceedings of the National Academy of Sciences, 99 14142 (2002).

44. CA. Savran, T. P. Burg, J. Fritz, and S.R. Manalis, "Microfabricated mechanical biosensor with inherently differential readout," Applied Physics Letters, 83 1659 (2003).

45. CA. Savran. S.M. Knudson, A.D. Ellington, and S.R. Manalis, "Micromechanical detection of proteins using aptamer-based receptor molecules," Analytical Chemistry, 76 3194 (2004).

46. Irraeli RS et al, Cancer Res. 1994:54:1807-1811.

47. Kadmon D, Thompson TC, Lynch GR and Scardino PT. Elevated plama chromogranin-A concentrations in prostatic carcinoma. J Urol 1991 ' 146:358-361)

48. Tricoli JV, Schoenfeldt M and Conley BA. Clin Cancer Res.;10:3943-3953 (2004).

49. References for biomarkers:

PSA: Johnson ED, Kotowski TM, "Detection of prostate specific antigen by ELISA". J Forensic Science, 38:250-258 (1993).

Tβl5: Hutchinson et al. "Development of a sensitive and specific enzyme-linked immunosorbent assay for thymosin βl5, a urinary biomarker of human prostate cancer" J Clin Chem. Submitted. (Tb- 15) paper.

PSMA: Huang. S, Bennett M, Thorpe PE. "Anti-tumor effects and lack of side effects in mice of an immunotoxin directed against human and mouse prosate-specifϊc membrane antigen." Prostate 2004;61 : 1-11

Chromogranin A: Tsao KC and Wu JT. "Development of an ELISA for the detection of serum chromogranin A (CgA) in prostate and non-neuroendocrine carcinomas." Clin Chim Acta 2001; 313:21-29

Claims

What is claimed is:

1. Micro fluidic device comprising: a nanofluidic concentrator for amplifying a sample containing a biomarker until the biomarker concentration approaches the disassociation constant of a biomarker/antibody complex; and a suspended microchannel resonator for receiving the amplified sample and generating signal related to the number of biomarkers contained in the sample.