The overall experimental plan involves creating a modest ...



a) SPECIFIC AIMS

Attempts over the last fifteen years to apply methods of protein characterization to the study of disease states and treatment have challenged the central genomic dogma of “one gene-one protein” [Gooley (1997)]. It is now well established that disease can alter not only the transcription of proteins, but it can also alter their co- and post-translational modifications. Consequently, methods that provide amino acid sequence information have been increasingly supplemented by techniques, like time-of-flight mass spectrometry (TOF-MS), that yield ancillary information regarding off-normal signaling or abnormal metabolic processes at the cellular level. TOF-MS methods now include laser desorption and ionization methods (LDI). These methods depend upon prior sample processing that mixes the biological material of interest with photon-absorbing “matrix” compounds (matrix-assisted LDI, or MALDI). A more recent variant of MALDI, Surface Enhanced Laser Desorption Ion Mass Spectrometry (SELDI-MS © Ciphergen), uses a proprietary surface preparation to specifically bind bio-molecules that are thought to be of particular interest. Both MALDI and SELDI are designed to give the exquisite sensitivity needed to provide markers for the early detection of diseases like cancer. A major long-term goal of this research area is to reliably identify the presence of abnormal components in body fluids at concentrations ranging from 10-21 moles/liter (zeptomolar) to 10-18 moles/liter (attomolar). Another long-term goal is to develop analytical methods that yield information about subtle conformational changes (secondary through quaternary structures) in proteins that might be associated with the presence of disease. Such conformational changes may show up as small changes in the relative concentrations of normal end products of cellular post-translational modification (e.g. different phosphorylations of the same protein [Machida (1996)).

Achievement of these two long-term goals will require significant improvement of LDI instruments. One must provide maximum sensitivity while at the same time providing maximum dynamic range. In addition, one must improve the mass resolution of LDI over a wide mass range. Finally, for the resulting technology to have any clinical impact, the instruments developed must have reasonable cost and be easy to use. The most exquisite instrument imaginable will have very little impact on the treatment of human disease if is not fully integrated into a clinical diagnostic system. Our proposed improvements are based upon our experience with the use of SELDI instruments to identify cancer biomarkers. This ongoing project, funded under separate NIH grants, is a collaboration among William and Mary, the bioinformatics company INCOGEN, and the proteomics group at Eastern Virginia Medical School. Previous research by this group has revealed that TOF-MS data can potentially provide disease biomarkers, but it has also revealed the limitations in existing LDI-MS technology. This proposal will show that a fast, imaging, counting detector can lead to simultaneous major advances in sensitivity and dynamic range. Our discussions will concentrate on SELDI methods since that is the area in which we have an already established a partnership, but we emphasize that the work will be equally applicable to MALDI and other forms of TOF-MS methods for proteomic work.

Our specific aims are

1) We will implement the use of an imaging detector to improve mass discrimination.

2) We will use the imaging detector to optimize the surface characteristics of the SELDI-TOF source processes.

3) We will develop a more quantitative model of SELDI plume processes by including surface effects and electrostatic interactions that affect the plume shape.

4) We will create an inherently digital detection scheme to improve the signal to noise and to provide a “stationary” noise background to permit better application of signal processing methods we are developing under a separate program.

5) We will begin the preliminary design of a fully multiplexed imaging detector that can be retrofitted to existing instruments for next-generation SELDI experiments.

b) Background and Significance

Mass spectrometry (MS) techniques hold great promise for disease diagnosis because these methods can extract large amounts of quantitative information from small samples of analyte. But, there are significant challenges before MS methods can routinely construct “protein profiles” or “metabolic profiles” of individuals to diagnose disease or monitor treatment therapies. The large bio-molecules thought to be of importance in disease processes are much larger, and much more fragile, than the materials previously analyzed via MS with great success. In order to measure the mass of a molecule, that molecule must be separated from the surrounding material environment, ionized, and then manipulated using electromagnetic fields in order to arrive at the particle detector at a time and a position that encodes information about its charge to mass. For application to disease diagnosis, the analytical process must be carried out without breaking these large and floppy bio-molecules into fragments. Even then, the complex biochemistry of tissue or blood samples will provide the MS analyst with a bewildering variety of masses. It is possible that the presence of disease will be revealed most reliably by the trace presence of a protein or metabolite that should not be there, rather than variations in the large number of common proteins that are normally present. Any significant fragmentation of the analyte bio-molecules will turn what is already a difficult challenge into a combinatoric nightmare.

The usefulness of MS profiling for diagnosis will ultimately be measured by the ability to identify features in a variety of data sets that allow reliable diagnostic discrimination and prognosis. To become a clinically viable (inexpensive, non-invasive, fast, and reliable) cancer diagnostic tool, MS applications have to face the challenge of providing quantitative measures of the relative amounts of many proteins that depend on the medical condition. Recent research confirms that approaches that focus on a single protein biomarker may not be effective in improving detection, diagnosis, and prognosis (Wright 1999, Vlahou 2001). Instead, approaches that evaluate multiple biomarkers simultaneously maximize the amount of information that can be extracted from the data source (Ibid.). MS survey instruments, such as MALDI and SELDI, are ideal for such applications because of their wide mass range and high sensitivity.

All Laser Desorption Ionization Mass Spectrometers (LDI-MS) use a laser pulse to desorb or ablate sample material from a carefully prepared surface. The matrix in MALDI consists of low-mass molecules of ~100’s Da (e.g. sinapinic acid, M~225 Da) designed to provide both efficient binding of bio-molecules to the surface and to generate the initial “plume” of material formed by the laser pulse. The analyte consists of the bio-molecules embedded within, or on the surface of, the matrix crystals. Viscous effects within the laser-initiated matrix plume drag the heavier proteins along with it. Ions are formed in the plume via a complex combination of photochemical processes and ion-molecule reactions, including proton transfer and other charge-exchange reactions between the matrix and analyte molecules. Thus it is believed that plume collisions are required for the dominant ionization process (Vertes, 2002). These ions are then extracted from the formation region, accelerated by an applied electric field and, in the simplest designs, the charge-to-mass ratio is inferred from the time it takes for the ion to arrive at a detector a known distance from the sample. Such time of flight (TOF) methods are well suited for reasonable cost clinical instruments because they have a wide mass range and can be physically compact (the ions are separated in time, not space). For the remainder of the proposal we will refer to SELDI and MALDI techniques generically as “LDI-TOF”. The discussion of specific surface preparation problems below will apply only to SELDI, although similar issues are encountered in MALDI.

LDI-TOF techniques are of great interest for disease profiling because this method of ionization produces less fragmentation of the high-mass proteins than other methods, such as secondary ion mass spectroscopy (SIMS). While SIMS presently can be used for MS analysis of proteins up to perhaps 10 kDa in mass, LDI-TOF techniques routinely can measure protein masses up to hundreds of kDa’s. LDI-TOF methods are capable of high sensitivity (fmol or attmol) over a wide mass range, but present instruments have limited resolution. For example, M/ΔM ~ 1000 is common for high-throughput survey instruments with a mass range up to several 100 kDa. MALDI instruments that are tuned to look for a much narrower mass range can achieve significantly higher resolution, but current generation SELDI instruments have resolutions of less than a thousand, and yield little or no information below M = 2 kDa.

An Eastern Virginia Medical School (EVMS) proteomics team has employed SELDI-MS to develop a non-invasive blood serum assaying technique for cancer screening (Xiao 2001, Adam 2001, Petricoin 2002, Adam 2002, Ball 2002). However, low mass resolution and incorrect assignment of peak masses due to hardware limitations are among the fundamental weaknesses of SELDI-MS profiling (Adam 2002). This prevents SELDI from identifying biomarkers in complex mixtures that could be directly linked to known proteins associated with the various disease states. The few successful attempts to identify disease-related changes in MS profiles (Xiao 2001, Vlahou 2001, Ball 2002, Adam 2002, Petricoin 2002, Lilien 2003) have relied either on labor-intensive repetitive manual searches of the large data sets by expert Ph.D.-level researchers, or ad hoc applications of computationally intensive, limited scope, statistical techniques which are often done without proper cross-validation, which is necessary to provide a reliable prediction of clinical performance. The lack of consistent and objective cross-validation, coupled with inadequate noise characterization, limits the pace at which this important disease profiling method can achieve broader use in clinical diagnostics.

To address some of these issues, the EVMS team formed a collaboration with INCOGEN, a bioinformatics company, and with a team of mathematicians and physicists from the College of William and Mary (CWM) to examine their SELDI data more closely and to develop a systematic approach to finding biomarkers in clinical samples. This collaboration meets weekly, and over the last eighteen months has analyzed SELDI samples related to prostate cancer, and leukemia. The team has developed a modular approach to biomarker discovery, as illustrated in Figure 1.

The logic of the approach is simple: the raw spectrum data must first be filtered so as to remove any instrument-related variability and artifacts before passing the processed data on to a set of classification or pattern recognition algorithms (here indicated by the two modules ‘Profile Construction’ and ‘Discrimination / Classification’). Such noise filtering requires a careful characterization of all noise processes in the instrument itself. Instrumentation noise can originate in a wide variety of places, as will be discussed more fully in the section ‘Preliminary Studies’. Only by attacking the problem within this unified framework, can one be assured that the end goal of effective clinical predictions is optimized by changes to any of the steps intervening between the ‘raw data’ and the ‘clinical prediction’ itself. The software platform needed for this integrated research strategy is being developed by INCOGEN under a separate grant proposal. Because the approach is modular, and because our collaboration is already in place, this framework will easily incorporate any improvements to the hardware and provide a ‘fast track’ for moving any hardware improvements directly to bear upon clinical studies.

Here, we propose to replace much of the signal analysis box of Figure 1 with a radical redesign of the detector that will produce a 2D digital image of the TOF separated ions at a very high frame rate (>100 MHz). Because of its digital nature, this detector will eliminate most of the current noise sources, and because it is an imaging detector, it will speed advances in the preparation and characterization of LDI-TOF processes. In particular, this new detector will lead to immediate advances in the understanding of the LDI-TOF plume dynamics and in the role of the surface preparation in LDI-TOF.

There have already been attempts to use ion-counting technology to improve the statistics of LDI-TOF (Westmacott 2002); however, those attempts either severely reduced the signal throughput to achieve ion-counting conditions, or used a quadrupole ion filter/trap to preprocess the TOF signal. Since the ions are discrete, a detection system that counts events can reduce the noise to the fundamental Poisson limit. However, this often limits the dynamic range of a detector, because if multiple ions arrive at the same time, most detectors will only register a single count. Consequently, counting detectors are most often used when counts arrive well spaced in time. For TOF measuring devices, this typically means simultaneously increasing the measurement repetition rate while decreasing the average signal size. This is not possible for LDI-TOF since a reasonable heavy-ion count rate requires a minimum threshold laser fluence and a minimum laser spot size (Dreisewerd, 1995; Westmacot 2002), which together always produce a large matrix signal. The matrix must be present in order to drag the bio-molecules off the surface, yet it is of no biological interest. This is the basic hurdle that every LDI-TOF instrument must face. To make matters worse, for reasons that are not understood, the heavy ion signals only persist for a few tens of laser shots in any one spatial location (Furnival 1997).

Yet, it is not clear on physical grounds why there appears to be a minimum useful laser fluence and spot size. Most computational models predict a threshold laser fluence for ion production (Zenobi 1998; Zhigilei 2003), but careful measurements show a heavy ion signal that increases as the fluence raised to a high power (~7-10) when the fluence changes by a factor of five (Westmacott 2002). It is possible that in addition to increases in the ionization mechanisms, kinetic or hydrodynamic effects due to the ablated plume are increasing the signal size. Such an increase would be expected to follow dynamic shaping of the plume that focuses more of the embedded ion beam onto the detector. Other one dimensional models, developed by Vertes (Vertes, 1993) have found similar power law dependences without invoking such hydrodynamic focusing, however. Measurements have shown that the plume dynamics produces initial ion velocities in the 300-1000 m/s range (Gluckmann 1999). A recent attempt to image the neutral molecules in a MALDI plume using laser-induced fluorescence (LIF) (Puretzky 1999) showed that the plume does appear to produce focusing of the heavy components. This work used much higher fluences and larger spot sizes than typical for a LDI-TOF measurement in order to observe the LIF, but the work suggests that plume dynamics plays an important role in the anomalous signal increase with laser fluence. This work also strongly suggests that the divergence of the plume is mass dependent, and can be used to image the high mass components to detect them selectively. An earlier measurement by Zhang and Chait (Zhang, 1998) imaged the ion plume at very low laser fluence (just at threshold). They used a high resolution CCD camera to image a phosphor screen, but their time resolution was limited by a long phosphor fluorescence time and the camera’s integration time. They used a pulsed high voltage acceleration on the fluorescence plate to gate their TOF signal for a single time window per laser shot and averaged several laser shots, since the frame readout rate of the camera was slow. As Vertes has recently pointed out (Vertes, 2002) the plume dynamics and the ionization processes are both still poorly understood, and they interact in complex ways. The imaging detector we propose here will help fill in an important piece of the puzzle because we will be able to image the ion plumes in a system whose ion optics have been carefully mapped out, allowing us to trace the ion trajectories back close to the surface of last scattering where they entered into free flight.

There are a variety of physical mechanisms through which laser pulses can generate plumes. For example (see Zhigilei, 2003): plumes can be generated by single particle desorption (which occurs at very low fluence and leads to a very tenous plume), phase explosion (where the absorbed energy superheats the surface past a critical temperature for transition to a phase which ‘explodes’, producing a dense plume with many embedded clusters of the surface material), hydrodynamic sputtering (where large parts of the surface ‘flow’ into the surrounding vacuum, producing spikes and droplets), and ablation (where multiple monolayers of the material are lofted, or spalled, off the surface without undergoing a prior phase transition). Which of these mechanisms occurs in any given laser shot depends strongly upon the laser fluence (energy/area), the pulse time, the orientation of the laser with respect to the local surface normal, and the chemical nature of the surface. Furthermore, the ionization mechanism in LDI, which is typically performed at fluences of ~ 10-100 mJ/cm2 , ,is believed to be dominated bv ion-molecules reactions due to collisions that occur in the immediate vicinity of the surface, rather than by photoionization due to the laser pulse itself. If this is correct, ionization must occur very quickly before the plume density drops so low that collisionality becomes negligible. This is because the photon energies in most LDI experiments are too low to directly ionize the molecules (Vertes, 2002). We note, however, that at higher powers it is well known that processes such as inverse bremsstrahlung can initiate ionization by promptly freeing electrons, thereby creating a plasma which can interact strongly with the laser pulse. Under such high-power irradiation, measurement of the backscattered light reveal how much of the laser energy is absorbed by the surface and provide a way to better calibrate the initial plume conditions. We note in passing that Braren et al. (Braren, 1991) have observed plume emission from polymer surfaces that continues ~500 times longer than the laser pulse (up to ~6 μs for a 20 ns pulse) for much higher fluences (~ 1 J/cm2). If MALDI or SELDI plumes outgassed for such long times, and ionization took place throughout the plume generation time, it would destroy the mass resolution, hence this work does not appear to be relevant to MALDI or SELDI-MS, though the gas-dynamical models that these authors proposed might be useful starting points for comparison.

In the following section on Preliminary Studies, we will illustrate the difficulties resulting from an analog detection system, and discuss the improvements possible using a digital detector. This work was peformed using SELDI samples prepared in a clinical environment by our EVMS partners and so represents the current state-of-the-art for cancer marker development. We will also illustrate how a fast multiplexed detector can yield the benefits of counting while maintaining a large dynamic range in the Research Design section. Also in the Research Design section, we will show how an imaging detector can answer some of these vexing problems of LDI-TOF. The research proposed here will provide the scientific basis needed before proceeding to an upgrade of clinical instruments, either by retrofitting existing LDI-TOF instruments with new sources, detectors and software tools, or by building a new instrument from the ground up. This would require a much larger effort, of course, and would be carried out in a follow-on project.

c) PRELIMINARY STUDIES

In our preliminary work in collaboration with EVMS and INCOGEN, we have measured the detector characteristics of EVMS’s SELDI-MS system; we have made initial calculations of the mass-dependent effects of Coulomb repulsion in an ion plume; and we have used the extensive surface analytical capabilities at the Applied Research Center of the College of William and Mary (CWM) to examine SELDI chips.

Noise effects due to detector saturation:

After viewing typical clinical data samples, we asked our colleagues at EVMS to perform a number of simple calibration runs on their equipment so that we could better interpret the noise sources. The most important calibration runs recorded the results from a set of single laser shot exposures (repeated in the same location) on IMAC-Cu affinity chips with sera samples. These were compared to sets of single shots on the backsides of blank aluminum chips, as received from the spectrometer manufacturer. The blank chip runs, which provided us with raw data without any background subtraction, showed a number of important features. (The Ciphergen SELDI instrument comes equipped with a software package that performs background subtraction and ‘peak enhancement’. This software was found to be ineffective for dealing with the issues described here and therefore was not used.) For example, in Figure 2, we show that compounds present even on the received blank aluminum chips produce ion signals that saturate the detector and obscure any near-by signals. Even though the saturated peak value was constrained to be no more than 110% of the maximum measurable level (100 arbitrary units) and the saturated level was applied for only a short period of time, several problems arose. First, the amplitude of the signal no longer followed the input during the interval between 710-830 points, as the system gain entered an unknown, non-linear state. Second, even after the system apparently recovered, the baseline suddenly shifted at 830 points, so that to use the data beyond 830 points requires a “jump” correction to the baseline. Third, the apparent random noise around the baseline is higher for a significant period of time following the saturation event, making small peaks much more difficult to see. Finally, the recovery of the system to the initial gain can effectively shift the timing relative to the system master clock to produce a signal timing jitter that depends on location in the spectrum making accurate mass assignments for parent peaks and for adducts very difficult.

The saturation effects become even worse in a biological sample, where copious matrix ions are generated along with the biologically interesting ions (Figure 3). Usually, all data for masses less than 2 kDa are ignored as being part of the “matrix”. This in itself is unfortunate since there are many metabolites of interest in that region. Even worse, however, nonlinear effects in the detector chain produce a detectable ringing in the time series, shown in Figure 3, that is present for much of the remaining time record.

Even without the overload, the detector’s response to either a blank chip or a processed chip shows non-stationary inherent noise. Figure 4 shows the logarithm of the local variance of the signal (calculated using a 128 point moving window) for a 192 shot average TOF spectrum. The large structures on the top curve (from a biological sample) are due to real occasional mass peaks, but note the strong correlation between the shapes of the two curves at very high masses. This correlation is noteworthy because the blue baseline represents the inherent noise in the system, since it was taken from a blank aluminum sample. The noise decreases monotonically, with occasional bursts, over several orders of magnitude and clearly has both random and non-random components. This non-stationary change in system response makes it very hard to use time filtering and background subtraction methods to detect small signals. This variation also complicates any attempt to normalize the signal based on the integral of the total ion current.

The time-dependent variation in the system response also complicates any attempts to use time gating to eliminate the massive matrix signal. If, for example, one uses a pulsed electrostatic gate to divert the matrix plume out of the path of the later heavy ions, the radio-frequency pick-up is likely produce even more ringing in the detector. Thus far, such attempts have been less than successful. Clearly, it would be highly desirable to eliminate the matrix effects by using the natural physical features of the plume to provide initial mass discrimination to provide a possible answer to part of this problem. An imaging detector with variable gain in the spatial dimension might separate the heavy and light ions without a pulsed gate. Moreover, by combining this imaging detector with single-ion pulse counting techniques, all ringing problems will be eliminated as well.

Noise effects due to surface preparation:

Figure 5 shows a series of single shot spectra of pooled-serum on IMAC-Cu affinity chips at selected mass peaks in the SELDI spectrum associated with known peptides present in the sample. They all decrease in a highly irregular fashion until their supply is exhausted at approximately 30 shots. Although several models describing the production of LDI-TOF signals have been proposed in the literature [Vertes 2002, Gooley 1997), none properly accounts for even the general features of the above combination of observations, and certainly no model can describe the detailed shot-to-shot variation in Figure 5. It is evident that there is no strong correlation of peaks and valleys for even the three (peptide) masses displayed. This is presumably due to a combination of effects, with the surface characteristics of the chip likely playing a very important role.

With this in mind, we have examined the surface characteristics of SELDI chips, and have found a wide variation of crystal faces, sizes, and orientations presented to the incident laser pulse, as shown in the TOF-SIMS image in Figure 6. This has been noted by other authors as well (see, e.g. Puretzky and Geohegan, 1998). We found no visible evidence of surface melting of the SELDI surfaces no matter how many laser shots the samples were given. Melting to liquid, which would show up as a rounding of corners or edges, is completely absent for these crystals of sinapinic acid. It is known that sinapinic acid sublimates upon reaching its melting point of 203-205 C, so it is not surprising that no liquid phase appears. It is surprising, however, that the crystals show no evidence of ablative or evaporative recession of any of their exposed surface. Indeed, there is little or no visible difference samples with or without irradiation.

The observed surface variation presumably leads to a wide spatial variation of laser absorption and reflection, and a variation in the expanding plume physics (Dreisewerd 1995; Fournier 2002) since the laser spot will typically contain only a handful of crystals. All of these effects are further exacerbated by large variations in the spatial distribution of surface composition, where our TOF-SIMS images have shown large variations in the locations of sodium, copper, and silicon.

Plume dynamics

The plume dynamics at early times and near the surface can have a strong impact upon the resulting ion signals since the ionization is most likely due to collisions very near the surface (Vertes, 2002). In addition to its effects upon the total ion signal, the plume dynamics can also effect the spatial distribution of ions through a combination of hydrodynamic and kinetic effects that are poorly understood. For example, Puretzky et al. (Puretzky, 1999) have directly imaged the neutral plumes in MALDI experiments and shown that the low-mass plume has a very different spatial distribution from the high-mass plume. The low-mass plume expands self-similarly as with an elliptical shape while the high-mass plume is more tightly focused in the direction normal to the surface. Kelly et al. (Kelly, 1998) have noted that the plume aspect ratio depends strongly upon the original spot shape as well. They interpret this as due to gas dynamic effects (i.e. anisotropic pressure gradients in the initial dense plume), though we have found that it could also arise from a purely kinetic effect. This is because high velocity particles will tend to escape preferentially in the direction where the plume is thinnest (typically normal to the surface). These two distinct physical explanations for the observed density anisotropy can be distinguished by a detailed measurement of the velocity distributions. If the anisotropy is pressure driven, Puretzky’s model predicts the fluid velocity field should also be anisotropic, while if it is due to a temperature anisotropy, emission of free streaming particles from a small source predicts that far from the source the fluid velocity field should be isotropic.

Ionization must occur close to the surface where the initial density of the plume will be close to solid densities (Moskovets, 2002). Even if the ionization fraction is low, the mean separation between ions can still be quite small on macroscopic scales. For example, an ionization fraction of 10-5 with an initial neutral density of 1021 cm-3 gives an ion density of 1016 cm-3 and a typical ion-ion separation distance of ~5 x 10-6 cm = .05 μm. In SELDI, an applied electrostatic field of about 20 kV/cm is applied to the ions over a distance of about 1 cm, pulling the positive ions away from the surface and accelerating them toward the detector. Ignoring shielding effects due to electrons and negative ions, if the initial ion separation is of the order of 0.1 μm then a simple estimate shows that the Coulomb force between a single pair of ions is initially comparable to that of the extraction field. In addition, the electrostatic potential energy between a pair of ions at this separation is about 10-3 eV. If that potential energy is converted into perpendicular kinetic energy (1/2 mvperp2),and the extraction potential is converted into longitudinal kinetic energy (1/2 mvz2), then this naïve calculation suggests that vperp/vz ~ 10-4. Therefore, independent of their mass, over a flight distance of ~1m Coulomb effects will cause the ions to spread laterally by about .1 mm. We must recognize, however, that the initial ion temperatures are probably of the order of a few eV, hence it might be argued the Coulomb effects will be swamped by the thermal spreading due to ordinary gasdynamic effects. But, these Coulomb effects increase rapidly with the total number of ions, and depend sensitively upon the initial separation, (while the thermal spreading does not) hence much stronger ion plume spreading can be expected in the proper conditions. The initial ion density and total count, in turn, depend critically upon the very early stages of the plume dynamics and the surface chemistry. Pursuing this line of reasoning: recall that the simple estimate based upon a pair of ions above led to the prediction that there is no mass dependence to the spreading. However, the lighter ions from the matrix will be most dense and, hence, will produce the largest Coulomb effects. They will also be extracted by the applied field first. Their high initial density potentially leads to a ‘Coulomb explosion’ for the light matrix ions transverse to the ion beam direction. The slower moving, and much less numerous, high-mass ions by contrast will see the local charge density in the early plume rapidly drop as the light ions are pulled away. They will, as a consequence, experience a much weaker Coulomb-driven expansion. Thus there will be a mass-dependent effect and it tends to ‘focus’ the heavier ions relative to the light ions. This ‘mass focusing’ effect is distinct from the gas dynamic effect observed by Puretzky et al. (Puretzky, 1998) for the neutral plume. Mass-dependent radial spreading of MALDI ion beams has been reported by Zhang and Chait (Zhang, 1998) and they report that the higher masses tend to ‘lag’ but do not propose an explanation for it; we note that mass-dependent “slip” in a hydrodynamic expansion is not unexpected. Zhang and Chait’s measurements were made in a laser fluence regime where only a few hundred ions were generated per laser shot, but the effect was so large that they suggest it will require the ion beam be refocused to avoid losing signal. If this effect is only due to thermal effects, then it would not have a strong dependence upon the laser fluence, but if it is partly due to Coulomb effects, then this spreading effect will grow rapidly as the laser fluence is increased to enhance the ion yield. Our proposal is not to simply refocus the ion beam, but to exploit the mass dependence of this effect. We have performed direct numerical simulations of the Coulomb effects, integrating the classical equations of motion for N=100-1000 ions. We find results that are consistent with the estimates quoted above, and in agreement with Zhang and Chait’s experimental results. Even without thermal spreading, by the time the light ion, which start out separated by a fraction of a micron, reach the detector plane, the radius of the beam has increased by several orders of magnitude. The much less numerous heavy ions arrive much later with a much tighter radial focus, even though both light and heavy ions were fully mixed spatially at t=0. We expect that as the fluence is increased above threshold, and the ion production increases substantially (it typically increases as Y~I7-8 [Westmacott, 2002]), the Coulomb effects will become more pronounced and it will be possible to ‘fine tune’ them to optimize the separation of the matrix ions from the heavier bio-molecules.

As noted in Kelly (Kelly, 1998) and Puretzky (Puretzky, 1999), the neutral plume anisotropy – both longitudinal vs. radial and in the two radial dimensions transverse to the beam - depends strongly upon the spot size and shape. The model of Puretzky for the neutral plume shape suggests that the anisotropy of pressure gradients, due to the different length scales of the initial plume, will drive an anisotropy in the density. It also apparently, leads to anisotropy in the velocity fields, though this not discussed explicitly by them. In the measurements of Zhang and Chait (Zhang, 1998) the radial velocities are ~150-200 m/s while the initial longitudinal velocities are 3-4 times larger for heavy ions and even higher for the matrix (Beavis, 1991). A simple kinetic calculation for the emission of free streaming particles from a small source can also recover an anisotropic density, provided the initial velocity distribution is anisotropic. But, in this case, the ‘hydrodynamic’ velocity field, defined as the average kinetic particle velocity at a given spatial point, will be completely isotropic far from the source. Hence, a measurement of the velocity field (inferred from the space-time imaging of the ion plume) can be used to distinguish these two different possibilities. Because the neutral plume was imaged using laser induced fluorescence, Puretzky et al. needed to use fluences that were well above threshold for ion production, even though these ions were not imaged. Zhang and Chait (Zhang, 1999) and Spengler and Bokelman (Spengler, 1993) have imaged the ion plumes and found mass dependence as well as focusing effects.

The plume dynamics and ion production should also depend upon the pulse length since that determines the penetration depth of the hot spot. If the pulse length is short enough, and most photons are absorbed near the surface, then phonons created by absorption events will not have time to traverse a distance that is characteristic of the matrix crystal thickness. (Using a sound speed of ~ 1 km/s and a crystal thickness of a few microns, this suggests that laser pulses shorter than 1 ns should not have time to reflect off the back of a matrix crystal. However, we note that the extinction depth of the matrix material is not well known and may well be measured in microns, which is the value used by Vertes (Vertes, 1993). Hence the assumption that most photons are absorbed at or near the surface and energy transported away from the surface either by phonons or some other mechanism in the solid needs to be treated with caution. The depth of the hot spot, along with the thermal transport properties of the matrix, determines the time it takes for the plume emission to shut down and for the plume to detach. Once the plume detaches from the surface, it expands and cools rapidly, shutting off the ion production.

Small spot sizes should produce matrix plumes that expand more in the radial direction (Kelly, 1998), potentially providing better spatial separation between the low- and high-mass components. Vertes (Vertes, 1993) has found that with a simple one-dimensional hydrodynamical model of the plume it is possible to recover the experimentally observed strong power-law dependence of total plume mass yield as a function of laser fluence. We anticipate, however, following the work of Kelly (Kelly, 1998), Puretzky (Puretzky, 1999) and Zhang (Zhang, 1997), that there will be a strong spatial dependence in the radial direction of the plume with focusing effects that depend upon the mass, and that this will be different for the neutral and the ion components of the plume. This will be particularly true for small spot sizes (which expand more quickly in the radial direction than large spot sizes) and for higher fluences (due to Coulomb effects).

Variation of the extraction voltage will modify the time it takes for the ions of different masses to segregate themselves in the plume. We expect that for lower extraction voltages, the heavier mass ions will remain intermingled with the far more numerous matrix ions and, hence, experience a greater radial spread at the detector due to Coulomb effects. A systematic study of this dependence will give insight into the regime where ions begin to segregate by mass. Radial spreading of the ions will be due to thermalization with respect to the background neutrals, and Coulomb interactions with other ions. Hence, varying the extraction voltage allows us to probe these two effects and begin to distinguish their impact on the ion plume.

Initial test of an imaging detector for ions

We have tested a new fast imaging detector that consists of two microchannel plate particle multipliers followed by fast phosphor plate and observed the single count signals using a 1P28 phototube. These preliminary tests were performed in an existing chamber, under far from optimal conditions in order to assess the viability of the method. Even with the phototube mounted 25 cm away from the fluorescent screen (solid angle of less than 0.02), with low voltage on the microchannel plates (approximately 850 V each) and only 1.3 kV bias on the phosphor screen, we easily observe signals from single ions that are in excess of 10 mV, as shown in figure 8. The signal size decreases rapidly at high count rates, since the chevron microchannel plate detector has a very small capacitance (approximately 100 picofarad per plate) which leads to charge depletion on the second plate for a short time after an event. We have observed that two pulses must be separated by at least 2 microseconds to have equivalent amplitudes. For times shorter than this, the second pulse has an amplitude that is approximately proportional to the time between the pulses. We intend to improve this either by replacing the chevron assembly with a single plate, or by using a bias network to increase the capacitance per plate. In any event, it is still easy to distinguish between two arrival events that are separated by less than 50 ns. With proper imaging optics, and a faster response photo-detector, we expect to distinguish between events separated by less than 10 ns, yielding a large improvement in mass resolution for SELDI.

d) RESEARCH DESIGN AND METHODS

The overall experimental program will have three major teams: a detector team, a surface preparation team and a modeling team. The detector team will create a modest resolution time-of-flight (TOF) apparatus that accepts a standard SELDI chip at the front end and replaces the normal particle detector at the back end with a fast, imaging, particle-counting detector. The surface preparation team will develop methods to shape SELDI chip surfaces to maximize the plume dynamics ability to separate high and low mass components without sacrificing the biologically interesting molecules. The modeling team will create models of the plume dynamics and propose imaging measurements to the detector team to check the models’ validity. The goal of this modeling and code development effort will be to produce a suite of tools capable of taking measurements from the imaging detector, tracking the ions back to where they emerged from the early plume, and then comparing this ‘retrodiction’ with the output from simulations of the early plume in order to untangle the physics and optimize the instrument. Each of these teams has a variety of tasks that support these basic efforts, and they are outlined in Table 1.

Year 1:

Detector team

The detector team will begin with installation of an El-Mul dual microchannel plate (MCP) particle multiplier coupled to a fast phosphor plate that uses a proprietary E-37 phosphor. The chevron assembly of MCPs provide a net gain of over 106 electrons for each incident ion, and the fast phosphor can produce up to 30 photons for each incident electron, producing a net light pulse of nearly 3mW-peak power, lasting for 2 ns. Figure 8 shows two sequential ion signals produced by singly charged argon ions striking this detector in one of our test stations. The team will also measure how the gain of the system varies with position on the detector and with increasing count rate. We have already seen significant gain decreases for pulses separated by less than 2 microseconds, presumably due to charge depletion of the MCP or due to phosphor saturation. We will be able to minimize the first effect by introducing more capacitance in the bias network of the MCP, while the second effect should be limited to high-count rates at the same position on the detector. Eventually, this effect could be eliminated by a slight rastering of the detected beam, although we will not implement any such fix until we have determined the maximum count rate at which the phosphorescence signal deteriorates.

After characterizing the detector, we will install it in a second vacuum chamber that we have designed to study the plume characteristics of a SELDI signal. This chamber uses a minimal TOF region (20 cm long) that will be sufficient to let the plume expand so that collisionality ceases, and to separate the low and high-mass ions. This chamber allows us to apply an extraction voltage of 0-20 kV (as in a standard SELDI systems), so that we can separate the plume hydrodynamic effects from the Coulomb collisional effects during its expansion. The TOF region is intentionally short in this chamber to insure that our detector properly images the plume with a high efficiency. To test this arrangement, we will use an amplified Colliding Pulse Mode-locked (CPM) laser to generate point sources of Xe+ in the source region. The amplified CPM laser pulse has a wavelength of 620 nm, a pulse length of less than 300 fs, and a pulse energy in excess of 0.3 mJ. When focused to a spot size of less than 50 (m, this will produce copious Xe+ ions from a volume of less than 0.003 mm3, even when the background Xe gas pressure is less than 10-7 torr. By varying the position of the laser focus, we will be able to determine the imaging and collection efficiency of this chamber, prior to making detailed studies of the plume expansion into the detector and to map out the transfer function of source points to time and position points at the detector. This work will use a gated camera, rather than counting electronics. At high ion densities, we expect to see the Coulomb expansion effects as predicted by the modeling team.

After characterizing the transfer function, the detector team will create standard SELDI signals from SELDI plates in the low resolution TOF apparatus. The team will use a doubled dye laser pulse at 337 nm, since its optical beam quality can be made nearly TEM00 mode by spatial filtering, to mimic the nitrogen laser in a standard SELDI apparatus. With this initial apparatus, the team will observe the spatial extent of the ionic components of the plume and measure those parameters’ dependence on laser spot size and laser fluence over a wide range from threshold to the high fluence used in the LIF imaging experiments.

Surface Preparation team

The plume dynamic models discussed above share the common feature that the angular divergence of the emitted plume is assumed to be centered around the normal vector to the sample surface. There does not appear to be a strong dependence on the incident angle of the laser beam. As noted above, the current protocols for SELDI sample preparation lead to a surface consisting of jumbled field of inhomogeneous crystal sizes randomly oriented with respect to this vector. We will attempt to control this aspect of the plume geometry by using a chemical-mechanical polishing (CMP) technique to create optically smooth sample surfaces. Presently, such methods are routinely used in the production of nanoscale features in the semiconductor industry (Babu, 2001). The fundamental idea is to use rotate a polishing pad, under conditions of carefully controlled mechanical pressure, against a sample which has been wetted with slurry consisting of ultra-fine polishing grit (such as diamond or silicon carbide) suspended in appropriate chemical solvent. It is possible to create planar surfaces with local rms height variations on the order of 1nm - 10 nm over distances of several centimeters using this method (Ravi, 1999). This is comparable to the smoothest surfaces that can be created using any surface processing method and rivals the atomic smoothness of single crystal faces (Tang, 2000). This step can be performed after the usual EVMS SELDI sample preparation protocols, provided we can maintain (or increase) the matrix-bound protein or peptide concentrations that occur throughout the nascent surface selvedge layer of the rough crystalline field shown in Figure 6. The choice of solvent will therefore be important; we initially propose the use of liquid mixtures in which the crystals are only slightly soluble since this will give us the greatest control over composition and topography.

During the first year, this team will implement a planarization method and check to insure that it does not destroy the biologically interesting molecules by characterizing it using the EVMS SELDI. The team will do routine inspection using high magnification reflection (inverted) optical microscopy, and they will measure flatness and roughness using contact and non-contact profilometry. They will measure nano-scale surface topography with very high precision using atomic force microscopy and field-emission scanning electron microscopy. Finally, they will measure sample composition and its uniformity using imaging TOF-SIMS. The ultimate objective will be to find final processing protocols that maximize the SELDI signal and resolution and lead to the greatest angular separation of plume components.

Modeling team

During the first year, the modeling team will concentrate on determining how ion density effects the ion optics, and compare their calculations to the observations of the detector team using various densities of multiphoton ionized xenon atoms. The modeling team will also help design ion optics to emphasize the divergence differences that we expect to see between high and low masses.

Once the detector and the ion optics system has been characterized, this team will lead the study of how the high and low mass portions of a SELDI plume depend on the the laser fluence, the laser spot size, the pulse length, the extraction voltage (including time delay pulsing used for ‘mass focussing’), and the surface preparation. These effects will be studied both experimentally and via numerical simulation and theoretical modeling. The numerical code development the first year will primarily concern ion optics calculations in support of the characterization of the instrument and will be carried out using standard packages (e.g. SIMION) and custom software developed by the team. Initial theoretical studies and model development for the early dense plume will also be undertaken to improve our understanding of the potential relevance of the ‘explosive’ phase transformation (i.e. superheating of the matrix sublimation or melting transition) that is believed to be relevant (Zhigilei, 2003) and how it can drive hydrodynamic flow, the effects of electron and negative ion shielding (i.e. boundary layer plasma effects and whether a Saha equilibrium is appropriate in the early plume), and the possible effects of non-gaussian diffusion that occurs in the plume region between the Knudsen layer where (collisions dominate) and free flight. Often these two regimes are simply matched, but Braren et al. have pointed out (Braren, 1991, Kelly, 1998) that sometimes an intermediate layer (which they describe as unsteady adiabatic expansion) can play an important role. Unsteady jet behavior has also been seen in direct solutions of the Boltzmann equation (Aristor, 2002). Braren, et al. have proposed a hydrodynamic model for this effect, but it is known that even if only small population of particles can ‘runaway’ it can have large macroscopic effects and such phenomena are usually studied in plasmas using some type of hybrid model where the bulk is treated as a fluid and the sub-species kinetically, or by using ‘fractional’ diffusion models. For example, del Castillo Negrete, et al. (del Castillo Negrete, 2003) have recently shown that reaction diffusion models with non-gaussian diffusion can have reaction fronts that accelerate. The interaction of these effects with phase transformations appears to be unexplored (see, e.g. Zanette, 1997). The first year we will examine simple one-dimensional models to develop some feel for the relative importance of these different effects. Moskovets and Vertes have noted (Moskovets, 2002) that the initial plume dynamics and the ion production are closely coupled, hence by varying the plume dynamics, and measuring the space and time effects upon the ion plumes, we can begin to untangle this complex interplay. As part of this theoretical effort, we will also begin to examine the effects of surface shaping of the matrix (e.g. due to polishing) on the resulting plume hydrodynamics.

Year 2:

Detector team

Assuming that we will observe a difference in the high and low mass characteristics that depends on the laser factors, we will optimize the high mass SELDI signal and minimize the shot-to-shot signal variation by adjusting the conditions appropriately. In the final stage of this preliminary work we will modify our imaging detector so that it can operate in an ion counting mode. To do this, we will first introduce beam-shifting optics (if necessary) to separate the low mass and high mass signals spatially. This will allow us to introduce physical (rather than electronic) gain variation to decrease the overload of the low mass region. Following this, we will replace the gated camera with an array of detectors (segmented phototubes, avalanche photodiodes or fast photodiodes, depending on the net system gain). The array characteristics will be matched to the high mass beam spatial image. In this first detector, we will use no more than 8 multiplexed detectors, to allow us a maximum of 8 ion counts in any 10 ns bin of time. We will simultaneously develop a fast multiplexer/counter that will sum the counts from these detectors by feeding each signal into a fast comparator to generate standard 5 V amplitude signals that we will then sum together in an analog summing amplifier. We will then sample this output with a fast A/D converter.

During this year, we will replace the short TOF region with an extended (1m) drift tube. We will again characterize the point source transfer function from the source region to the detector, and we will image characterized plume geometries to verify the quality of source imaging that the extended system can maintain. With the extended system, we will accurately characterize the mass dependence of the plume expansion. At lowest order, we expect a breakpoint that separates high-mass form low-mass behavior, and we expect this break point to depend on the plume generation conditions.

Surface Preparation team

During the second year, this team will continue participating in studies of the plume characteristic’s dependence on the surface orientation in the short LDI-TOF apparatus. It will also begin using contour polishing to create convex or concave surfaces, while maintaining optical quality smoothness about the local surface normal. Under some circumstances, concave surfaces may permit us to take advantage of hydrodynamic focusing effects in the expanding plume in order to minimize unwanted effects in the ion optics. Thus we may be able to minimize the initial longitudinal size of the ion-creation zone and to simultaneously contract the radial dimension prior to extraction to reduce off-axis electrostatic lens distortion. In other cases, depending on the fundamental physics of the plume hydrodynamics and the mass difference between the matrix and the lowest peptide mass of interest, it may be desirable to use a convex surface to increase the differential divergence of low and high mass components.

Modeling team

The modeling team will begin implementing hybrid models of the plumes based upon the previous year’s theoretical studies. There are a variety of methods than could be employed here, each of which has merits and weaknesses that will be researched as part of the first year effort before committing significant resources to one particular numerical strategy. Potential methods to be examined include: Direct Simulation Monte Carlo methods (recently employed by Zhigilei, et al. to model MALDI plumes, [Zhigilei, 2003]), direct solution of the Boltzmann equation (Aristov, 2001), particle-in-cell (PIC) methods (Leboeuf, et al.), and hydrodynamic codes (possibly modified to include non-gaussian transport, if the first year’s theoretical efforts determine this to be important). In order to study the mass focussing effects that we expect, and the effects of surface shaping, these will have to be at least two-dimensional simulations, but direct comparison with experiment will eventually require three-dimensional simulations. Output from these simulations of the early plume dynamics and ion generation will then provide input to the ion optics codes developed in Year 1.

Year 3:

Detector team

The final step in this preliminary development will be to institute an ion counting system. The details of this system will depend upon the results of the plume characterization measurements, since one of the major obstacles is reducing the low-mass signal (matrix) relative to the high-mass signal (biomolecule). Our imaging detector will exploit the observed differences in the high and low-mass plume characteristics to implement a spatial gain variation that enhances the low mass signal relative to the high mass signal. This could be a decrease of the gain of one portion of the MCP or of the phosphor screen, or even an optical filter after the phosphor plate. Or, it could also be a time dependent variation introduced at any of these stages. We expect to avoid time dependent changes, since they are likely to introduce noise into the detection electronics, although a counting detector is much more insensitive to such noise sources. In this phase, we will implement up to an 8X8 array of detectors and we will still use the analog summing technique. However, we will also begin feasibility studies of using Field Programmable Gate Arrays to do fast pixel counting within an entirely digital system. It is crucial that we develop a scaleable model for this electronic summing, since we will increase this number by at least two orders of magnitude in a future improvement.

Surface Preparation team

The surface team will calibrate and correlate the contoured surfaces using both LDI-TOF and TOF-SIMS characterization.

Modeling team

The modeling team will continue refining its hybrid models of the plume in conjunction with the experimental observations.

Table 1: Timeline for tasks for each of the three teams.

|Task |Year 1 |Year 2 |Year 3 |

|Detector Development |Image detector transfer function |Multiplex 8 detectors |8X8 image prototype |

| |Image plume, low resolution |Implement 100 cm TOF |Design all-digital detection system using |

| |Plume dependence on spot size and fluence |Test diverging lenses |FPGA |

| | | |Integrated component design |

|Surface |Post LDI characterization |Surface effects on LDI |Calibrate and correlate LDI-TOF with |

|Characterization |Implement planarization scheme |Contour surfaces |SIMS-TOF |

| |Preliminary characterization of processed |Amino propyl silane surface | |

| |surfaces on EVMS SELDI | | |

|Simulation and |Ion optics characterization and inversion with |Hybrid models of plumes + ion optics |Refine models |

|modeling |and without Coulomb effects |Test diverging lenses | |

| |Surface shape effects on gas dynamics | | |

| |Develop ion divergence lenses | | |

| |Examine models of early plume dynamics | | |

g. Literature Cited

Adam B.-L., et al., “Proteomic approaches to biomarker discovery in prostate and bladder cancers”, Proteomics, v. 1, pp. 1264-1270 (2001)

Adam B.-L., et.al. “Serum Protein Fingerprinting Coupled with a Pattern-matching Algorithm Distinguishes Prostate Cancer from Benign Prostate Hyperplasia and Healthy Men”, v.62, pp.1609-3614 (July 2002)

Aristov, V. V., Direct Methods for Solving the Boltzmann Equation and Study of Nonequilibrium Flows (Kluwer Academic, Boston, 2001).

Babu S.V., Cadien K.C., and Yano H. (eds), Chemical-mechanical Polishing 2001, Advances and Future Challenges:2001, Materials Research Society Proceedings, Vol 671, MRS Press, Boston, (2001)

Ball G., et al, “An integrated approach utilizing artificial neural networks and SELDI mass spectrometry for the classification of human tumours and rapid identification of potential biomarkers”, Bioinformatics, v.18, pp.395-404, (2002)

Beavis R.C. and Chait B.T., “Velocity Distribution of Intact High Mass Polypeptide Molecule Ions Produced by Matrix Assisted Laser Desorption”, Chemical Physics Letters 181 (1991) 479.

Braren, B., Kelly, G. C. and Kelly, R., “On the Gas Dynamics of Laser-pulse Sputtering of Polymethylmethacrylate”, Nuclear Instrumentation and Methods in Physics Research B58 (1991) 463.

Dreisewerd K., Schürenberg M., M. Karas M., Hillenkamp F.,”Influence of the Laser Intensity and Spot Size on the Desorption of Molecules and Ions in Matrix-Assisted Laser Desorption/Ionization with a Uniform Beam Profile”, Int. J. Mass Spetrom. Ion Proc, v.14, pp. 127-148 (1995)

Fung E.T., Enderwick C.,“ProteinChip Clinical Proteomics: Computational Challenges and Solutions”, Comp. Prot. Sup., v.32, pp.34-41 (March 2002)

Furnival G.M. and Wilson R.W., “Regression by leaps and bounds”, Technometrics, v.16, pp. 499-511 (1974)

Fournier I., Beavis RC, Blais, JC, Tabet J.C., and Bolbach G., “Hysteresis effects observed in MALDI using oriented, protein-doped matrix crystals” Int. J. Mass Spectrom and Ion Proc. 169 pp.19-29 (1997).

Fournier I., Tabet J.C., and Bolbach G., “Irradiation effects in MALDI and surface modifications” Int. J. Mass Spectrom. 219 pp.515–523 (2002).

Gluckmann M and Karas M, “The initial ion velocity and its dependence on matrix, analyte and preparation method in ultraviolet matrix assisted laser desorption/ionization,” J. Mass. Spectrom. 34, pp 467-477 (1999).

Part I: Sinapic acid monocrystals

∗

Gooley,A.A. and Packer,N.H., "The importance of Protein Co- and Post- Translational Modifications in Proteome Projects". Chapter 4 in Proteome Research: New Fronteirs in Functional Genomics, M.R. Wilkins, K.L. Williams, R.D. Appel, and D. F. Hochstrasser, eds, Springer-Verlag,Berlin-Heidelberg-New York, (1997)

 

Gnanadesikan R., “Methods for Statistical Data Analysis of Multivariate Observations”,

John Wiley & Sons, New York (1977)

Gusev A. I., et al. “Improvement of Signal Reproducibility and Matrix/Co-matrix Effects in MALDI Analysis, Analytical Chemistry, v. 67, pp. 1034-1041 (March 1995)

Hastie T., Tibshirani R., and Friedman J.H, “The Elements of Statistical Learning”, Springer Verlag, New York (2001)

Hawkins D.M., “The subset problem in multivariate analysis of variance”, J. Royal Stat. Soc., Series B, v. 38, pp. 132-139 (1976)

Hensel R. R., et al. “Electrospray Sample Preparation for Improved Quantitation in Matrix-Assisted Laser Desorption/Ionization Time-of-Flight Mass Spectrometry”, Rapid Communications in Mass Spectrometry, v. 11, pp. 1785-1793 (1997)

Kato H., Ishihara M., Nakata M., “Resolution Enhancement in Mass Spectrometry by Autoregressive Deconvolution”, J. Mass Spectrom. Soc. Jpn, v. 49, pp. 175-182 (2001)

Kelly, R., Miotello, A., Mele, A. and Guidoni, A. G., “Plume formation and Characterization in Laser-Surface Interactions”, in Laser Ablation and Desorption, J. C. Miller and R. F. Haglund, eds. (Academic Press, New York, 1998)

Leboeuf, J. N., Chen, K. R., Donato, J. M., Geohegan, D. B., Liu, C. L., Puretzky, A. A., “Modeling of Plume Dynamics in Laser Ablation Processes for Thin Film Deposition of Materials”, Phys. Plasmas 3 (1996) 2203.

Lilien R.H., Farid H. and Donald B.R., “Probabilistic Disease Classification of Expression-Dependent Proteomic Data from Mass Spectrometry of Human Serum”, J. Comp. Bio. (2003, in press)

Machida K., Nishimura N., Imaizumi M., Abe T., Ohnishi Y., Takagi K., Yoshii S., Hamaguchi M., "Tyrosine phosphorylation in lung cancer as a prognostic marker", Cancer Detection and Prevention, vol 20 (5), page xx, (1996)

 

Marshall A.G., “Fourier Transform in NMR, Optical and Mass Spectrometry” , Elsevier NY (1990) p.450.

MassTech, “AP/MALDI Source for LCQ Deca XP ion trap: Installation, Operation and Maintenance Manual”, (April, 2003)

McCabe G.P., “Computations for variable selection in discriminant analysis”, Technometrics, v. 17, pp. 103-109 (1975)

McHenry C.E., “Computation of a best subset in multivariate analysis”, App. Stat., v. 27, pp. 291-296 (1978)

Moskovets E. and Vertes A., “Fast Dynamics of Ionization in Ultraviolet Matrix-Assisted Laser Desorption Ionization of Biomolecules”, J. Phys. Chem. B 106 (2002) 3301.

Nicola A.J., et al. “Automation of Data Collection for Matrix-Assisted Laser Desorption /Ionization Mass Spectrometry Using a Correlative Analysis Algorithm”, Analytical Chemistry, v. 70, pp. 3213-3219 (1998)

Petricoin E.F., III, et al., “Use of proteomic patterns in serum to identify ovarian cancer”, The Lancet, v. 359, pp. 572-577 (2002)

Priebe C.E., Marchette D.J., DeVinney J.G., and Socolinsky D.A., “Classification using class cover catch digraphs”, J. Classific., v. 20, pp. 3-23 (2003)

Puretzky A.A., Geohegan D.B., “Gas-phase diagnostics and LIF-imaging of 3-hydroxypicolinic Acid MALDI-matrix Plumes”, Chemical Physics Letters 286 (1998) 425.

Puretzky A.A., Geohegan D.B., Hurst G.B., Buchanan M.V., and Luk’yanchuk B.S., “Imaging of vapor plumes produced by Matrix Assisted Laser Desorption: a plume sharpening effect,” Phys. Rev. Lett 83, pp 444-447 (1999)

Ravi K.V., Future Fab Internat., vol 7, (1999), 207.

Read T.R.C. and N. A. C. Cressie N.A.C., “Goodness-of-Fit Statistics for Discrete Multivariate Data”, Springer-Verlag, New York (1988)

Robinson E.A. and Treitel, S., “Detection and Extraction of Signals in Noise” in “Statistical Communication and Detection“, Griffin, London, (1967) Ch.9, p.249

Senko M.W., Beu S.C., McLafferty F.W., J. Am. Soc. Mass Spectrom. 6 (1995) 229; ( ucsfhtml4.0/msiso.htm)

Tang, X.M., Wang, Qi, and Manos D.M., J. Vac. Sci. Technol B18,1262, (2000).

Trosset M.W., “Formulations of multidimensional scaling for cluster analysis and classification”, In 1999 Proc. Stat. Comp. Sect. Am. Stat. Assoc., pp. 195-200 (1999)

Vertes, A., “Hydrodynamic Model of Matrix-Assisted Laser Desorption Mass Spectrometry”, Anal. Chem. 65 (1993) 2389.

Veryovkin I.V., Constantinides I., Adriaens A., Adams F., (Brussels, Sept. 5-10, 1999) SIMS XII Conf. Proc., pp. 359-362 (2000)

Vlahou A., et.al., “Development of Novel Proteomics Approach for the Detection of Transitional Cell Carcinoma of the Bladder in Urine”, Am. J. Pathol., v. 158, pp.1491-1502 (2001)

Wagner M., Tyler B., Castner D., “Interpretation of Static TOF SIMS Mass Spectra of adsorbed protein films by multivariate pattern recognition”, Anal. Chem., v. 74, pp. 1824-1835 (2002)

Westmacott G, Ens W, Hillenkamp F, Dreisewerd K, Schurenberg M. “The influence of laser fluence on ion yield in matrix-assisted laser desorption ionization mass spectrometry.” Int. J. Mass Spectrom., v221, pp.67-81. (2002).

Wehofsky M., Hoffmann R., Hubert M., Spengler B., “Isotopic Deconvolution of Matrix assisted laser desorption/ionization mass spectra for substance-class specific analysis of complex samples”, Eur. J. Mass Spectrom., v.7, pp. 39-46 (2001)

Windig W., Phalp J.M., Payne A.W., “A Noise and Background Reduction Method for Component Detection in Liquid Chromatography/Mass Spectrometry”, Anal. Chem., v. 68, pp. 3602-3606 (1996)

Winograd N., Braun R.M., “TOF-SIMS Imaging Mass Spectrometry and Combinatorial Chemistry” PharmaGenomics, pp. 34-50 (May/June 2002)

Wright G.L.Jr., et.al., “ProteinChip SELDI Mass Spectrometry: a Novel Protein Biochip Technology for Detection of Prostate Cancer Biomarkers in Complex Protein Mixtures”, Prostate Cancer and Prostatic Diseases, pp.264-267(1999)

Xiao Zh., et al. “Quantitation of Serum Prostate-specific Membrane Antigen by a Novel Protein Biochip Immunoassay Discriminates from Malign Prostate Disease”, Cancer Research, v. 61, pp. 6029-6033 (August 2001)

Zenobi R and Knochenmuss R, “Ion formation in MALDI mass spectrometry,” Masss Spectrometry Review 17, 337-366 (1998).

Zhigilei L., Leveugle E., Garrison B., Yingling Y., and Zeifman M., “Computer simulations of laser ablation of molecular substrates,” Chem. Rev. 103, pp 321-347 (2003).

-----------------------

[pic]

Figure 2: The collaboration of EVMS, INCOGEN, and the College of William and Mary has developed this modular data flow/analysis structure to aid in the discovery of biomarkers from survey MS data. The project proposed here will significantly improve the quality of the ‘raw spectrum data’ and replace much of the signal processing module by an advanced, imaging, digital detector.

[pic]

Figure 2: TOF record for a single laser shot on a blank SELDI chip, showing contaminants on the aluminum chip. With sufficient gain to see a mild saturation at 730 points, the detector response is erratic for more than 100 points.

[pic]

Figure 3: The low mass, or matrix, region shows severe saturation and ringing when conditions are set to optimize the high mass region.

200

100

Mass (kDa)

[pic]

Figure 5: The SELDI peak intensity decreases to almost zero after 30 laser shots in this sample of polled blood serum. Note that not all mass peaks show the same structure.

[pic]

Figure 4: Local variance of detector signal for blank aluminum chip (lower blue curve) and pooled serum SELDI signal (top green curve).

|[pic] |[pic] |

Figure 6: Total ion Ga+ SIMS images of the central areas of two spots on an IMAC-Cu chip (scale bar is 100 mðm). The left spot received no shots; the right spot received 204 shots of 33 mðJ (3 ns pu Total ion Ga+ SIMS images of the central areas of two spots on an IMAC-Cu chip (scale bar is 100 μm). The left spot received no shots; the right spot received 204 shots of 33 μJ (3 ns pulse length) from a nitrogen laser.

[pic]

Figure 8: Two single ion signals detected using the fast fluorescence detector. The second pulse is attenuated unless it occurs at least 2 microseconds after the first.

Figure 8: Schematic showing pulsed SELDI source with observational microscopic camera with magnification sufficient to observe crystal structure, and interferometric determination of source position, extraction region with Einzel lens, TOF tube, final focusing Einzel lens, El-Mul detector, imaging optics, gated camera/photo-detectors

Imaging lens

focusing lens

SELDI chip

Ion lens

TOF Tube

Imaging Particle Detector

Gated camera

Collecting Optics

................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download