The present invention demonstrates that the diagnosis and prediction of clinical behavior in patients with hematologic malignancies, such as leukemia, can be accomplished by analysis of proteins present in a plasma sample. Thus, in particular embodiments the present invention uses plasma to create a diagnostic or prognostic protein profile of a hematologic malignancy comprising collecting plasma samples from a population of patients with hematologic malignancies; generating protein spectra from the plasma samples with or without fractionation; comparing the protein spectra with clinical data; and identifying protein markers in the plasma samples that correlate with the clinical data. Protein markers identified by this approach can then be used to create a protein profile that can be used to diagnose the hematologic malignancy or determine the prognosis of the hematologic malignancy. Potentially these specific proteins can be identified and targeted in the therapy of these malignancies.

Description of the Invention

BACKGROUND OF THE INVENTION

1. Field of the Invention

The present invention relates generally to the fields of proteomics. More particularly, it concerns the use of proteomics for diagnosis and the prognosis of hematologic malignancies. Also, the invention relates to predicting the response to therapy and stratifying patients for therapy.

2. Description of Related Art

Hematologic malignancies are cancers of the blood and bone marrow, including leukemia and lymphoma. Leukemia is a malignant neoplasm characterized by abnormal proliferation of leukocytes and is one of the four major types of cancer. Leukemia is diagnosed in about 29,000 adults and 2,000 children each year in the United States. Leukemias are classified according to the type of leukocyte most prominently involved. Acute leukemias are predominantly undifferentiated cell populations and chronic leukemias have more mature cell forms.

The acute leukemias are divided into lymphoblastic (ALL) and non-lymphoblastic (ANLL) types and may be further subdivided by morphologic and cytochemical appearance according to the French-American-British classification or according to their type and degree of differentiation. Specific B- and T-cell, as well as myeloid cell surface markers/antigens are used in the classification too. ALL is predominantly a childhood disease while ANLL, also known as acute myeloid leukemia (AML), is a more common acute leukemia among adults.

Chronic leukemias are divided into lymphocytic (CLL) and myeloid (CML) types. CLL is characterized by the increased number of mature lymphocytes in blood, bone marrow, and lymphoid organs. Most CLL patients have clonal expansion of lymphocytes with B cell characteristics. CLL is a disease of older persons. In CML, the granulocytic cells predominate at all stages of differentiation in blood and bone marrow, but may also affect liver, spleen, and other organs.

Among patients with leukemia there can be a highly variable clinical course as reflected by varying survival times and resistance to therapy. Reliable individual prognostic tools are limited at present. Advances in proteomic technologies may provide new diagnostic and prognostic indicators for hematologic malignancies such as leukemia.

The term "proteome" refers to all the proteins expressed by a genome, and thus proteomics involves the identification of proteins in the body and the determination of their role in physiological and pathophysiological functions. The .about.30,000 genes defined by the Human Genome Project translate into 300,000 to 1 million proteins when alternate splicing and post-translational modifications are considered. While a genome remains unchanged to a large extent, the proteins in any particular cell change dramatically as genes are turned on and off in response to their environment.

As a reflection of the dynamic nature of the proteome, some researchers prefer to use the term "functional proteome" to describe all the proteins produced by a specific cell in a single time frame. Ultimately, it is believed that through proteomics, new disease markers and drug targets can be identified.

Proteomics has previously been used in the study of leukemia. For example, two-dimensional polyacrylamide gel electrophoresis (2-D PAGE) of proteins from the lymphoblasts of patients with ALL was used to identify polypeptides that could distinguish between the major subgroups of ALL (Hanash et al., 1986). In other studies of ALL using 2-D PAGE, distinct levels of a polypeptide were observed between infants and older children with otherwise similar cell surface markers (Hanash et al., 1989). Voss et al. demonstrated that B-CLL patient populations with shorter survival times exhibited changed levels of redox enzymes, Hsp27, and protein disulfide isomerase, as determined by 2-D PAGE of proteins prepared from mononuclear cells (Voss et al., 2001).

As these studies indicate, proteomics can be a useful tool in the study of hematologic malignancies. There is, however, a need for proteomics techniques that are more reliable and simple than those currently available in the art.

SUMMARY OF THE INVENTION

The present invention provides a novel approach that uses plasma proteomics to create a profile that can be used to diagnose hematologic malignancies and predict a patient's clinical behavior and response to therapy.

In one embodiment, the invention provides a method of creating a diagnostic or prognostic protein profile of a hematologic malignancy comprising: obtaining plasma samples from a population of patients with hematologic malignancies; generating protein spectra from the plasma samples; comparing the protein spectra with patients' clinical data relating to the hematologic malignancy; identifying a protein marker or group of protein markers in the plasma samples that correlate with the clinical data; and creating a protein profile based on the identified protein marker or group of protein markers, wherein the protein profile can be used to diagnose the hematologic malignancy or determine the prognosis of the hematologic malignancy.

In a preferred embodiment, the protein spectra is generated by mass spectrometry. The mass spectrometry may be, for example, SELDI (surface enhanced laser desorption/ionization), MALDI (matrix assisted desorption/ionization), or Tandem mass spectrometry (MS/MS). In other embodiments of the invention, the protein spectra is generated by two-dimensional gel electrophoresis. In certain aspects, the protein samples are fractionated before mass spectrometry analysis or two-dimensional gel electrophoresis. Fractionation can be according to a variety of properties, such as pH, size, structure, or binding affinity. In one aspect, plasma proteins are fractionated into 4 different fractions according to pH using strong anion exchange column (Fraction1.ident.pH9,pH7, Fraction2.ident.pH5, Fraction3.ident.pH4, Fraction4.ident.pH3, organic).

In certain aspects, the protein marker or group of protein markers that correlate with the clinical data are identified by univariate statistics, multivariate statistics, or hierarchical cluster analysis. In a preferred embodiment, the protein marker or group of protein markers that correlate with the clinical data are identified using correlation statistics with beta-uniform mixture analysis, genetic algorithms, univariate, and/or multivariate statistics. In other preferred embodiment, the protein marker or group of protein markers that correlate with the clinical data are identified using a decision tree algorithm. In some embodiments of the invention the clinical data comprises one or more of cytogenetics, age, performance status, response to therapy, type of therapy, progression, event-free survival, time from response to relapse, and survival time.

In preferred embodiments, the protein profile is used to diagnose the hematologic malignancy; classify the type of hematologic malignancy; predict a patient's response to drug therapy; predict a patient's survival time; or predict a patient's time from response to relapse. In certain embodiments, the hematologic malignancy is leukemia, non-Hodgkin lymphoma, Hodgkin lymphoma, myeloma, or myelodysplastic syndrome. The leukemia may be acute myeloid leukemia (AML), chronic myeloid leukemia (CML), acute lymphocytic leukemia (ALL), or chronic lymphocytic leukemia (CLL).

In another embodiment, the invention provides a method of predicting response to therapy in a patient with a hematologic malignancy comprising: obtaining a plasma sample from a patient; identifying a protein marker or group of protein markers in the plasma sample that is associated with response to therapy; and predicting the patient's response to therapy. In a preferred embodiment the hematologic malignancy is leukemia, non-Hodgkin lymphoma, Hodgkin lymphoma, myeloma, or myelodysplastic syndrome. The leukemia may be acute myeloid leukemia (AML), chronic myeloid leukemia (CML), acute lymphocytic leukemia (ALL), or chronic lymphocytic leukemia (CLL).

The method may be used to predict a patient's response to therapy before beginning therapy, during therapy, or after therapy is completed. For example, by predicting a patient's response to therapy before beginning therapy, the information may be used in determining the best therapy option for the patient.

In one aspect of the invention, the protein marker is a peak. The peak may be generated by mass spectrometry. The mass spectrometry may be, for example, SELDI, MALDI, or MS/MS. In another aspect of the invention, the protein marker is a spot. In a preferred embodiment the spot is generated by two-dimensional gel electrophoresis.

In certain embodiments of the invention the therapy is chemotherapy, immunotherapy, antibody-based therapy, radiation therapy, or supportive therapy (essentially any implemented for leukemia). In some embodiments, the chemotherapy is Gleevac or idarubicin and ara-C.

In some embodiments the protein marker or group of protein markers associated with response to a specific therapy in a patient with AML is one or more of Peak 1 to Peak 17 generated by SELDI mass spectrometry as defined in Table 1 (see Original Patent). In one embodiment, the group of protein markers associated with response to a specific therapy in a patient with AML comprises Peak 1 and Peak 2.

In one embodiment, the invention provides a method of predicting time to relapse in a patient with a hematologic malignancy comprising: obtaining a plasma sample from a patient; identifying a protein marker or group of protein markers in the plasma sample that is associated with time to relapse; and predicting the patient's time to relapse. In a preferred embodiment the hematologic malignancy is leukemia, non-Hodgkin lymphoma, Hodgkin lymphoma, myeloma, or myelodysplastic syndrome. The leukemia may be acute myeloid leukemia (AML), chronic myeloid leukemia (CML), acute lymphocytic leukemia (ALL), or chronic lymphocytic leukemia (CLL).

In one aspect of the invention, the protein marker is a peak. The peak may be generated by mass spectrometry. Preferably the peak is generated by SELDI mass spectrometry. In another aspect of the invention, the protein marker is a spot. In a preferred embodiment the spot is generated by two-dimensional gel electrophoresis.

In a preferred embodiment the protein marker or group of protein markers associated with time from response to idarubicin and ara-C to relapse in a patient with AML is one or more of the Peak 18 to Peak 29 generated by SELDI mass spectrometry as defined in Table 2 (see Original Patent).


In a preferred embodiment the protein marker or group of protein markers associated with relapse in a patient with ALL is one or more of the Peak 30 to Peak 49 generated by SELDI mass spectrometry as defined in Table 3 (see Original Patent).

In a preferred embodiment the protein marker or group of protein markers that differentiate between patients with L1/L2 ALL and patients with L3 ALL is one or more of the Peak 50 to Peak 69 generated by SELDI mass spectrometry as defined in Table 4 (see Original Patent).

Those skilled in the art will recognize that the specific identity of the proteins represented by the protein markers described herein, or of protein markers revealed by the methods described herein, is not necessary to create or utilize a diagnostic or prognostic protein profile. The presence or absence, or increased or decreased levels, of a protein marker or group of protein markers can be used to create or utilize a diagnostic or prognostic protein profile without knowledge of what the proteins are. For example, a diagnostic or prognostic protein profile could be created or utilized based on the pattern of a group of protein markers without needing to know the specific identity of the protein markers in the pattern.

In another embodiment, the invention provides a method of predicting response to therapy in a patient with a hematologic malignancy comprising: obtaining a bone marrow aspirate sample from a patient; identifying a protein marker or group of protein markers in the sample that is associated with response to therapy; and predicting the patient's response to therapy. In a preferred embodiment the hematologic malignancy is leukemia, non-Hodgkin lymphoma, Hodgkin lymphoma, myeloma, or myelodysplastic syndrome. The leukemia may be acute myeloid leukemia (AML), chronic myeloid leukemia (CML), acute lymphocytic leukemia (ALL), or chronic lymphocytic leukemia (CLL). In one aspect of the invention, the leukemia is CML.

In certain aspects, a protein marker of the present invention may be a P52rIPK homolog, follistatin-related protein 1 precursor, annexin A10, annexin 14, tumor necrosis factor receptor superfamily member XEDAR, a zinc finger protein, CD38 ADP-ribosyl cyclase 1, connective tissue growth factor, CD28, Bcl2-related ovarian killer, tumor necrosis factor receptor superfamily member 10D, X-linked ectodysplasin receptor, ectodysplain A2 isoform receptor, or chromosome 21 open reading frame 63.

It is contemplated that any method or composition described herein can be implemented with respect to any other method or composition described herein.

The use of the term "or" in the claims is used to mean "and/or" unless explicitly indicated to refer to alternatives only or the alternatives are mutually exclusive, although the disclosure supports a definition that refers to only alternatives and "and/or."

Throughout this application, the term "about" is used to indicate that a value includes the standard deviation of error for the device or method being employed to determine the value.

Following long-standing patent law, the words "a" and "an," when used in conjunction with the word "comprising" in the claims or specification, denotes one or more, unless specifically noted.

Other objects, features and advantages of the present invention will become apparent from the following detailed description. It should be understood, however, that the detailed description and the specific examples, while indicating specific embodiments of the invention, are given by way of illustration only, since various changes and modifications within the spirit and scope of the invention will become apparent to those skilled in the art from this detailed description.

DESCRIPTION OF ILLUSTRATIVE EMBODIMENTS

A. The Present Invention

Among patients with hematologic malignancies there can be a highly variable clinical course as reflected by varying survival times and resistance to therapy. Depending on the type of hematologic malignancy a patient has, therapy may include radiation, chemotherapy, bone marrow transplant, biological therapy, or some combination of these therapies. Thus, the accurate diagnosis of a patient's hematologic malignancy is important in determining which therapy option to pursue, as different malignancies respond differently to certain therapies. Even within a particular form of hematologic malignancy (e.g., AML, ALL, CML, CLL) there is significant variability in response to therapy among patients. For example, in acute myeloid leukemia (AML), response to standard chemotherapy (idarubicin+ara-C) varies significantly between patients, with approximately 50% of patients not responding to therapy. Although specific cytogenetic abnormalities in AML patients, such as -5, -7 and 11 q abnormalities, or poor performance status and advanced age are known to be associated with poor response to therapy, accurate prediction of response to therapy remains elusive. The ability to accurately diagnose and predict clinical behavior in patients with hematologic malignancies would allow stratification of patients for therapy options.

Current methods for determining diagnosis or clinical behavior in patients with hematologic malignancies are not reliable and typically depend on one molecule. The present invention enables the evaluation of thousands of proteins at the same time from which a protein profile can be generated that can be used to diagnose or predict clinical behavior in patients with hematologic malignancies. In addition, the invention uses proteomics in combination with blood plasma. Blood plasma is easy to collect and provides the most complex human-derived proteome, making it superior to cells and serum for proteomic studies of hematologic malignancies.

The present invention demonstrates that the diagnosis and prediction of clinical behavior in patients with hematologic malignancies can be accomplished by analysis of proteins present in a plasma sample. Thus, in particular embodiments the present invention uses plasma to create a diagnostic or prognostic protein profile of a hematologic malignancy comprising collecting plasma samples from a population of patients with hematologic malignancies; generating protein spectra from the plasma samples; comparing the protein spectra with clinical data; and identifying protein markers in the plasma samples that correlate with the clinical data. Protein markers identified by this approach can then be used to create a protein profile that can be used to diagnose the hematologic malignancy or determine the prognosis of the hematologic malignancy. In some embodiments, protein markers may be identified by comparing the protein profile from patients with hematologic malignancies with protein profiles from unaffected individuals.

Using the methods of the invention, those skilled in the art will be able to identify protein markers that can accurately diagnose hematologic malignancies, predict a patient's response to therapy, predict a patient's time to relapse, and predict a patient's survival time. Furthermore, the invention provides several protein markers shown to accurately predict response to therapy in patients with AML, as well as several protein markers shown to accurately predict the time to relapse in patients with AML.

B. The Plasma Proteome

Blood plasma is easy to collect and provides the most complex human-derived proteome, containing other tissue proteomes as subsets. The protein content of plasma can be classified into the following groups: proteins secreted by solid tissues and that act in the plasma; immunoglobulins; "long distance" receptor ligands; "local" receptor ligands, temporary passengers; tissue leakage products; aberrant secretions from cancer cells and other diseased cells; and foreign proteins (Anderson and Anderson, 2002).

Other body fluids including cerebrospinal fluid, synovial fluid, and urine share some of the protein content with plasma. These samples, however, are more difficult to obtain in a useful state than plasma. For example, collection of cerebrospinal fluid and synovial fluid are invasive procedures that can be painful and involve some risk, while processing urine to a useful sample for protein analysis can be difficult in a clinical setting. Blood plasma, however, may be easily collected by venipuncture. For example, venous blood samples can be drawn and collected in sterile ethylene diamine tetra acetate (EDTA) tubes. The plasma can then be separated by centrifugation. If desired, the plasma may be stored at -70.degree. C. for later analysis.

Characterizing the proteins in plasma can be challenging due to the large amount of albumin present and the wide range in abundance of other proteins. The present invention, however, shows that proteomics in combination with plasma can provide a reliable approach to diagnosing hematologic malignancies and predicting clinical behavior in patient's with hematologic malignancies.

C. Protein Analysis

The present invention employs methods of separating proteins from plasma. Methods of separating proteins are well known to those of skill in the art and include, but are not limited to, various kinds of chromatography (e.g., anion exchange chromatography, affinity chromatography, sequential extraction, and high performance liquid chromatography) and mass spectrometry. The separation and detection of the proteins in a plasma sample generates a protein spectra for that sample.

1. Mass Spectrometry

In preferred embodiments the present invention employs mass spectrometry. Mass spectrometry provides a means of "weighing" individual molecules by ionizing the molecules in vacuo and making them "fly" by volatilization. Under the influence of combinations of electric and magnetic fields, the ions follow trajectories depending on their individual mass (m) and charge (z). Mass spectrometry (MS), because of its extreme selectivity and sensitivity, has become a powerful tool for the quantification of a broad range of bioanalytes including pharmaceuticals, metabolites, peptides and proteins.

Of particular interest in the present invention is surface-enhanced laser desorption ionization-time of flight mass spectrometry (SELDI-TOF MS). Whole proteins can be analyzed by SELDI-TOF MS, which is a variant of MALDI-TOF (matrix-assisted desorption ionization-time of flight) mass spectrometry. In SELDI-TOF MS, fractionation based on protein affinity properties is used to reduce sample complexity. For example, hydrophobic, hydrophilic, anion exchange, cation exchange, and immobilized-metal affinity surfaces can be used to fractionate a sample. The proteins that selectively bind to a surface are then irradiated with a laser. The laser desorbs the adherent proteins, causing them to be launched as ions. The "time of flight" of the ion before detection by an electrode is a measure of the mass-to-charge ration (m/z) of the ion. The SELDI-TOF MS approach to protein analysis has been implemented commercially (e.g., Ciphergen).

2. Two-Dimensional Electrophoresis

In certain embodiments the present invention employs high-resolution electrophoresis to separate proteins from a biological sample such as plasma. Preferably, two-dimensional gel electrophoresis is used to generate a two-dimensional array of spots of proteins from a sample.

Two-dimensional electrophoresis is a useful technique for separating complex mixtures of molecules, often providing a much higher resolving power than that obtainable in one-dimension separations. Two-dimensional gel electrophoresis can be performed using methods known in the art (See, e.g., U.S. Pat. Nos. 5,534,121 and 6,398,933). Typically, proteins in a sample are separated by, e.g., isoelectric focusing, during which proteins in a sample are separated in a pH gradient until they reach a spot where their net charge is zero (i.e., isoelectric point). This first separation step results in one-dimensional array of proteins. The proteins in one dimensional array is further separated using a technique generally distinct from that used in the first separation step. For example, in the second dimension, proteins separated by isoelectric focusing are further separated using a polyacrylamide gel, such as polyacrylamide gel electrophoresis in the presence of sodium dodecyl sulfate (SDS-PAGE). SDS-PAGE gel allows further separation based on molecular mass of the protein.

Proteins in the two-dimensional array can be detected using any suitable methods known in the art. Staining of proteins can be accomplished with colorimetric dyes (coomassie), silver staining and fluorescent staining (Ruby Red). As is known to one of ordinary skill in the art, spots/or protein profiling patterns generated can be further analyzed for example, by gas phase ion spectrometry. Proteins can be excised from the gel and analyzed by gas phase ion spectrometry. Alternatively, the gel containing proteins can be transferred to an inert membrane by applying an electric field and the spot on the membrane that approximately corresponds to the molecular weight of a marker can be analyzed by gas phase ion spectrometry.

3. Other Methods of Protein Analysis

In addition to the methods described above, other methods of protein separation known to those of skill in the art may be useful in the practice of the present invention. The methods of protein analysis may be used alone or in combination. a. Chromatography

Chromatography is used to separate organic compounds on the basis of their charge, size, shape, and solubilities. A chromatography consists of a mobile phase (solvent and the molecules to be separated) and a stationary phase either of paper (in paper chromatography) or glass beads, called resin, (in column chromatography) through which the mobile phase travels. Molecules travel through the stationary phase at different rates because of their chemistry. Types of chromatography that may be employed in the present invention include, but are not limited to, high performance liquid chromatography (HPLC), ion exchange chromatography (IEC), and reverse phase chromatography (RP). Other kinds of chromatography include: adsorption, partition, affinity, gel filtration and molecular sieve, and many specialized techniques for using them including column, paper, thin-layer and gas chromatography (Freifelder, 1982). i. High Performance Liquid Chromatography

High performance liquid chromatography (HPLC) is similar to reverse phase, only in this method, the process is conducted at a high velocity and pressure drop. The column is shorter and has a small diameter, but it is equivalent to possessing a large number of equilibrium stages.

Although there are other types of chromatography (e.g., paper and thin layer), most applications of chromatography employ a column. The column is where the actual separation takes place. It is usually a glass or metal tube of sufficient strength to withstand the pressures that may be applied across it. The column contains the stationary phase. The mobile phase runs through the column and is adsorbed onto the stationary phase. The column can either be a packed bed or open tubular column. A packed bed column is comprised of a stationary phase which is in granular form and packed into the column as a homogeneous bed. The stationary phase completely fills the column. An open tubular column's stationary phase is a thin film or layer on the column wall. There is a passageway through the center of the column.

The mobile phase is comprised of a solvent into which the sample is injected. The solvent and sample flow through the column together; thus the mobile phase is often referred to as the "carrier fluid." The stationary phase is the material in the column for which the components to be separated have varying affinities. The materials which comprise the mobile and stationary phases vary depending on the general type of chromatographic process being performed. The mobile phase in liquid chromatography is a liquid of low viscosity which flows through the stationary phase bed. This bed may be comprised of an immiscible liquid coated onto a porous support, a thin film of liquid phase bonded to the surface of a sorbent, or a sorbent of controlled pore size.

High-performance chromatofocusing (HPCF) produces liquid pI fractions as the first-dimension of protein separation followed by high-resolution reversed-phase (RP) HPLC of each of the pI fractions as the second dimension. Proteins are now mapped (like gels), but the liquid fractions make for easy interface with mass spectrometry (MS) for detailed intact protein characterization and identification (unlike gels) on more selective basis without resorting to protein digestion. ii. Reversed-Phase Chromatography

Reversed phase chromatography (RPC) utilizes solubility properties of the sample by partitioning it between a hydrophilic and a lipophilic solvent. The partition of the sample components between the two phases depends on their respective solubility characteristics. Less hydrophobic components end up primarily in the hydrophilic phase while more hydrophobic ones are found in the lipophilic phase. In RPC, silica particles covered with chemically-bonded hydrocarbon chains (2-18 carbons) represent the lipophilic phase, while an aqueous mixture of an organic solvent surrounding the particle represents the hydrophilic phase.

When a sample component passes through an RPC column the partitioning mechanism operates continuously. Depending on the extractive power of the eluent, a greater or lesser part of the sample component will be retained reversibly by the lipid layer of the particles, in this case called the stationary phase. The larger the fraction retained in the lipid layer, the slower the sample component will move down the column. Hydrophilic compounds will move faster than hydrophobic ones, since the mobile phase is more hydrophilic than the stationary phase.

Compounds stick to reverse phase HPLC columns in high aqueous mobile phase and are eluted from RP HPLC columns with high organic mobile phase. In RP HPLC compounds are separated based on their hydrophobic character. Peptides can be separated by running a linear gradient of the organic solvent.

Along with the partitioning mechanism, adsorption operates at the interface between the mobile and the stationary phases. The adsorption mechanism is more pronounced for hydrophilic sample components while for hydrophobic ones the liquid-liquid partitioning mechanism is prevailing. Thus the retention of hydrophobic components is greatly influenced by the thickness of the lipid layer. An 18 carbon layer is able to accommodate more hydrophobic material than an 8 carbon or a 2 carbon layer.

The mobile phase can be considered as an aqueous solution of an organic solvent, the type and concentration of which determines the extractive power. Some commonly used organic solvents, in order of increasing hydrophobicity are: methanol, propanol, acetonitrile, and tetrahydrofuran.

Due to the very small sizes of the particles employed as the stationary phase, very narrow peaks are obtained. In some embodiments, reverse phase HPLC peaks are represented by bands of different intensity in the two-dimensional image, according to the intensity of the peaks eluting from the HPLC. In some instances, peaks are collected as the eluent of the HPLC separation in the liquid phase. To improve the chromatographic peak shape and to provide a source of protons in reverse phase chromatography acids are commonly used. Such acids are formic acid, trifluoroacetic acid, and acetic acid. iii. Ion Exchange Chromatography

Ion exchange chromatography (IEC) is applicable to the separation of almost any type of charged molecule, from large proteins to small nucleotides and amino acids. It is very frequently used for proteins and peptides, under widely varying conditions. In protein structural work the consecutive use of gel permeation chromatography (GPC) and IEC is quite common.

In ion exchange chromatography, a charged particle (matrix) binds reversibly to sample molecules (proteins, etc.). Desorption is then brought about by increasing the salt concentration or by altering the pH of the mobile phase. Ion exchange containing diethyl aminoethyl (DEAE) or carboxymethyl (CM) groups are most frequently used in biochemistry. The ionic properties of both DEAE and CM are dependent on pH, but both are sufficiently charged to work well as ion exchangers within the pH range 4 to 8 where most protein separations take place.

The property of a protein which govern its adsorption to an ion exchanger is the net surface charge. Since surface charge is the result of weak acidic and basic groups of protein; separation is highly pH dependent. Going from low to high pH values the surface charge of proteins shifts from a positive to a negative charge surface charge. The pH versus net surface curve is a individual property of a protein, and constitutes the basis for selectivity in IEC.

As in all forms of liquid chromatography, conditions are employed that permit the sample components to move through the column with different speeds. At low ionic strengths, all components with affinity for the ion exchanger will be tightly adsorbed at the top of the ion exchanger and nothing will remain in the mobile phase. When the ionic strength of the mobile phase is increased by adding a neutral salt, the salt ions will compete with the protein and more of the sample components will be partially desorbed and start moving down the column. Increasing the ionic strength even more causes a larger number of the sample components to be desorbed, and the speed of the movement down the column will increase. The higher the net charge of the protein, the higher the ionic strength needed to bring about desorption. At a certain high level of ionic strength, all the sample components are fully desorbed and move down the column with the same speed as the mobile phase. Somewhere in between total adsorption and total desorption one will find the optimal selectivity for a given pH value of the mobile phase. Thus, to optimize selectivity in ion exchange chromatography, a pH value is chosen that creates sufficiently large net charge differences among the sample components. Then, an ionic strength is selected that fully utilizes these charge differences by partially desorbing the components. The respective speed of each component down the column will be proportional to that fraction of the component which is found in the mobile phase.

Very often the sample components vary so much in their adsorption to the ion exchanger that a single value of the ionic strength cannot make the slow ones pass through the column in a reasonable time. In such cases, a salt gradient is applied to bring about a continuous increase of ionic strength in the mobile phase.

D. Analysis of Protein Markers

1. Extraction of Protein Marker Locations

Following the generation of protein spectra by, for example, SELDI-TOF MS, protein markers are identified for further analysis. Protein marker detection can be made easier by reducing the background noise. The background noise can be reduced at different levels. One method of reducing background noise is to average the raw protein spectra data. First, peaks should be normalized to assure that equal amounts of samples are compared. There are several methods for normalization known to those skilled in the art. A common approach is normalizing according to intensity: Total Ion Current, height, area, or mass. A different method for normalization is using the following formula (I=intensity): Normalized I=CurrentI-MinimumI/MaximumI-minimumI

After normalization, reducing background can be achieved by eliminating peaks that are not seen in majority (50-70%) of samples.

Systems for mass spectra acquisition are commercially available. One example is the Ciphergen ProteinChip.RTM. Reader (Ciphergen Biosystems, Inc.). The chip reader may be used with peak detection software such as CiphergenExpress 3.0. This software calculates clusters by determining peaks that are above a given signal-to-noise ratio, and that are present in multiple spectra. Various settings for noise subtraction, peak detection, and cluster completion may be evaluated to optimize the analysis. For example, a first pass peak detection of 5.0 signal-to-noise on both peaks and valleys, and a cluster completion window of 1.0 times peak width, with a second pass signal-to-noise setting of 2.0 for both peaks and valleys may be used.

The use of total ion current as a normalization factor is a common practice in SELDI data analysis; however, other methods of normalization may be used. For example, normalization could be done using the peak ratio approach in which the ratios of peaks near each other (e.g., within 5 peaks upstream and downstream) are used for normalizing. The peak ratio approach has an additional advantage of possibly detecting post-translational modifications more effectively.

Peaks may also be detected manually. The results of manual peak detection may then be analyzed using software, such as Matlab (MathWorks, Natick, Mass.), followed by decision tree analysis. A non-limiting example of decision tree analysis software is CART from Salford Systems, which is implemented in Biomarker Patterns Software 4.0 from Ciphergen Biosystems, Inc.

Replicate samples can be analyzed to confirm the reproducibility of the protein spectra generated according to the methods of the invention. Those of skill in the art are familiar with statistical methods that can be used to determine the reproducibility of the analysis. For example, an agglomerative clustering algorithm may be used to show that replicate samples cluster as nearest neighbors, thus confirming reproduciblitiy. Agglomerative clustering analysis is the searching for groups in the data in such a way that objects belonging to the same cluster resemble each other. The computer analysis proceeds by combining or dividing existing groups, producing a hierarchical structure displaying the order in which groups are merged or divided. Agglomerative methods start with each observation in a separate group and proceed until all observations are in a single group.

2. Determining the Relevance of Protein Markers

To test the relevance of the protein markers identified in the protein spectra, various methods of statistical analysis known to those of skill in the art may be employed. For example, a univariate model, multivariate model, or hierarchical cluster analysis may be used. a. Multivariate Modeling

A multivariate model is a model that aims to predict or explain the behavior of a dependent variable on the basis of a set of known independent variables. The purpose of using multivariate analysis is to demonstrate that the proteomic analysis as a variable in predicting response, survival, and duration of response is independent from the currently known variables that can predict the same thing. If the proteomic data adds to the model that includes the conventional markers, the p-value will be significant, but if the proteomic data does not add to the model and similar prediction can be achieved using other conventional markers, the p-value will not be significant even if it was significant in univariate analysis.

For predicting a the response to therapy of a patient with a hematologic malignancy, a multivariate model is preferred. An example of a multivariate model for predicting response to therapy in a patient with AML is (Response.about.Cytogenetics+Performance.Status+Age).

Cytogenetic findings represent the chromosomal abnormalities that were found in the tumor cells. Dependent on these abnormalities, the leukemia/tumor can be classified as good, intermediate, or bad. For example, in a patient with AML and cytogenetic abnormalities including deletion of chromosome 5 or 7 or abnormalities on chromosome 11, this patient has a "bad" disease (>90% die within one year and will not respond to therapy). Patients with AML and t(8;21), t(15;17), or Inv 16 are classified as "good" disease and the rest are with "intermediate" disease.

With regard to age, the older the patient the worse the disease (continuous variable). Patients >65 years old are classified with "bad" disease.

Performance status is a scoring system to evaluate the patient's overall health as described below in Table 5 (see Original Patent). Obviously, the higher the grade (ECOG), the less likely the patient will survive.

To test the relevance of a specific protein marker to the prediction of a behavior, the protein marker can be added to the multivariate model. For example, the value (i.e., height) of a protein peak identified by SELDI MS can be added to the base multivariate model for predicting response to therapy in a patient with AML to give the extended multivariate model of (Response.about.Cytogenetics+Performance.Status+Age+Peak Info) where Peak Info is information from a given peak. Preferably Peak Info is a transformed peak value, such as logPeak, logPeak+(logPeak).sup.2, Peak+Peak.sup.2, or Peak+logPeak.

After applying the peak value to the multivariate model, a p-value is produced. Those of skill in the art are familiar with methods of calculating p-values. For example, a p-value may be determined by applying ANOVA (analysis of variance between groups) on the base multivariate model and the extended multivariate model.

To adjust for multiple testing a beta-uniform mixture analysis may be used. The p-value is considered significant only if it is less than the cut-off as determined by the beta-uniform mixture analysis, in which the transformation is confirmed to be unique and not uniform. This adjusts for the multiple testing. b. Cox Model

Those of skill in the art are familiar with the Cox proportional hazards model, which is a commonly used regression model for analyzing data points with time, such as survival, time to progression, time to relapse, or time to therapy. The Cox model allows the estimation of nonparametric survival (or other event of interest) curves (such as Kaplan-Meier curves) in the presence of covariates. This can be performed with continuous or as dichotomized variables. The effect of the covariates upon survival is usually of primary interest. The Cox model can also be performed in the context of multivariate analysis by incorporating several variables. In the multivariate model, the analysis will first analyze the first variable, then analyze the second variable in the groups generated from the first variable and so on.

In one embodiment of the invention, protein peak values were fitted to the Cox model: h(t)=h.sub.0(t)exp(.beta.f(Peak)),

where h(t) is the hazard at time t, h.sub.0(t) is the baseline hazard, and f(Peak) is some transformation of the peak value. When the Cox model was applied to predict time to relapse, the "hazard" was relapse, and the "baseline hazard" was the risk of relapsing based on variables other than peak value. Resulting p-values may be analyzed by means of a beta-uniform mixture analysis. A positive value of the coefficient .beta. means that an increased peak height corresponds to increased risk of relapse. The p-value was considered significant only if it is less than the cut-off as determined by the beta-uniform mixture analysis, in which the transformation is confirmed to be unique and not uniform. This adjusts for the multiple testing.

In addition to the analyses described herein, many additional questions can be asked using the Cox model. For example, the data can be used to predict patients who will have fungal infection, or patients who would die in the first two weeks. Similar statistical analysis can be used to determine response to second therapy after relapsing c. Decision Tree Algorithm

In one embodiment of the present invention, a decision tree algorithm was used to identify protein spectra useful for predicting clinical outcome (e.g., responders versus non-responders). CART software from Salford Systems is one example of a commercially available decision tree tool. CART automatically sifts large, complex databases, searching for and isolating significant patterns and relationships. This information can then be used to generate predictive models. Variables that may be included in the analysis along with peak values and peak ratios include clinical outcome, patient demographics, and cellular analysis. When using decision trees, caution must be exercised to prevent overfitting (Wiemer and Prokudin 2004). When approach to limiting overfitting is to limit the number of levels allowed. For example, the number of levels may be limited to two, meaning that the model could only be comprised of at most two variables from the set of all peak values and all observational variables (e.g., clinical outcomes, patient demographics, cellular analysis).

Claim 1 of 4 Claims

1. A method of predicting an increased risk of relapse following therapy or distinguishing between L1/L2 and L3 in a patient with acute lymphoblastic leukemia (ALL) comprising: (a) performing mass spectrometry on a plasma sample from said patient to generate a protein spectra comprising protein peaks; (b) identifying a protein peak or group of protein peaks in the protein spectra corresponding to one or more of Peak 30 (7727.972 Daltons), Peak 31 (61940.76 Daltons), Peak 32 (124797.7 Daltons), Peak 33 (53623.64 Daltons), Peak 34 (10216.72 Daltons), Peak 35 (145023.4 Daltons), Peak 36 (6808.864 Daltons), Peak 37 (7249.661 Daltons), Peak 38 (6588.005 Daltons), Peak 39 (78971.03 Daltons), Peak 40 (4924.562 Daltons), Peak 41 (55864.83 Daltons), Peak 42 (6801.569 Daltons), Peak 43 (13298.19 Daltons), Peak 44 (83531.42 Daltons), Peak 45 (39542.43 Daltons), Peak 46 (159276.8 Daltons), Peak 47 (106256.1 Daltons), Peak 48 (88687.58 Daltons), Peak 49 (135305.2 Daltons), Peak 50 (7727.865343 Daltons), Peak 51 (10214.09619 Daltons), Peak 52 (9263.336516 Daltons), Peak 53 (10217.12293 Daltons), Peak 54 (7722.657526 Daltons), Peak 55 (7728.041349 Daltons), Peak 56 (9268.979905 Daltons), Peak 57 (7741.020002 Daltons), Peak 58 (9248.709422 Daltons), Peak 59 (7720.190664 Daltons), Peak 60 (13870.3916 Daltons), Peak 61 (7725.474001 Daltons), Peak 62 (9275.311795 Daltons), Peak 63 (41782.2775 Daltons), Peak 64 (8896.712054 Daltons), Peak 65 (4911.78345 Daltons), Peak 66 (83363.03733 Daltons), Peak 67 (45087.95748 Daltons), Peak 68 (121673.475 Daltons), or Peak 69 (7727.155842 Daltons), and (c) predicting risk of relapse following therapy or distinguishing between L1/L2 and L3 based on the identification of one or more of Peaks 30-69, wherein Peaks 30-49 are predictive of an increased risk of relapse following therapy, and Peaks 50-69 distinguish between L1/L2 and L3 ALL.

____________________________________________
If you want to learn more about this patent, please go directly to the U.S. Patent and Trademark Office Web site to access the full patent.