METHOD, ARRAY AND USE THEREOF

Field of Invention

The present invention relates to methods for diagnosis of pancreatic adrenocarcinoma, and biomarkers and arrays for use in the same.

Background

One of the remaining challenges in oncology is the ability to stratify patients, relating to their probability to experience tumor relapse or drug treatment resistance, or to their survival expectancy.

Pancreatic ductal adenocarcinoma is the most lethal malignancy by anatomic site, with >30,000 new cases and deaths annually in the United States alone, and with a 5-year survival of 3-5%. This extreme mortality is due to the lack of effective early diagnostic methods (5) and to poor efficacy of existing therapies for advanced disease. Even the patients (10-20%) diagnosed with a surgically resectable tumor, ultimately die of recurrent and metastatic disease. An increased ability to detect and predict cancer is therefore crucial for individual patient management.

Antibody microarray technology (3) has the potential to provide a highly multiplexed analysis (6,7) and has been suggested as the technology platform that eventually will deliver a defined protein signature, i.e. a combination of serum proteins that distinguish cancer from normal patients. Microarray technology has now matured to the point were the initial hurdles have been overcome and minute amounts of proteins in complex proteomes can be analyzed (8-12). However, gene expression profiling of cancer has only demonstrated the ability to predict time of survival in few cases (1 ,2) and no combination of serum proteins has so far been associated with any of the above clinical parameters.

A serum sample analysis that can predict survival time would allow a more individualized cancer therapy. This has been emphasized for e.g. pancreatic adenocarcinomas, where no tumor-specific markers exist - although most patients will have an elevated CA 19-9 at time of diagnosis, individual prognostic markers have shown to be inconclusive (4). Furthermore, non-invasive approaches, such as computed tomography, is not sufficiently sensitivity to detect small cancers, whereas e.g. endoscopic ultrasonography can be used to survey at-risk individuals for pancreatic lesions (5).

Against this background, the inventors have now developed a proteomic approach to prognostic diagnosis of cancer and identified the first sets of serum biomarkers for detection of pancreatic cancer and for predicting survival.

Summary of the Invention

Accordingly, in a first aspect, the invention provides a method for determining the presence of pancreatic adrenocarcinoma in an individual comprising the steps of:

a) providing a protein sample to be tested (e.g. serum or plasma); b) determining a protein signature of the test sample by measuring the presence and/or amount in the test sample of one or more proteins selected from the group defined in Table 1 ;

wherein the presence and/or amount in the test sample of one or more proteins selected from the group defined in Table 1 is indicative of the presence of pancreatic adrenocarcinoma.

By "protein signature" we include the meaning of a combination of the presence and/or amount of serum proteins present in an individual having cancer and which can be distinguished from a combination of the presence and/or amount of serum proteins present in an individual not afflicted with cancer (such as pancreatic adenocarcinoma) - i.e. a normal, or healthy, individual.

As exemplified in the accompanying Examples, the presence and/or amount of certain serum proteins present in a test sample may be indicative of the presence

of cancer, such as pancreatic adenocarcinoma, in an individual. For example, the relative presence and/or amount of certain serum proteins in a single test sample may be indicative of the presence of cancer, such as pancreatic adenocarcinoma, in an individual.

Preferably, the individual is a human, but may be any mammal such as a domesticated mammal (preferably of agricultural or commercial significance including a horse, pig, cow, sheep, dog and cat).

Preferably, the method of the first aspect of the invention further comprises the steps of:

c) providing a control serum or plasma sample from an individual not afflicted with pancreatic adrenocarcinoma; d) determining a protein signature of the control sample by measuring the presence and/or amount in the control sample of the one or more proteins measured in step (b);

wherein the presence of pancreatic adrenocarcinoma is identified in the event that the presence and/or amount in the test sample of the one or more proteins measured in step (b) is different from the presence and/or amount in the control sample of the one or more proteins measured in step (b).

Preferably, the presence and/or amount in the test sample of the one or more proteins measured in step (b) is significantly different (i.e. statistically different) from the presence and/or amount in the control sample of the one or .more proteins measured in step (b). For example, as discussed in the accompanying

Examples, significant difference between the presence and/or amount of a particular protein in the test and control samples may be classified as those where p<0.05. ^{;' "} .

Typically, the method of the first aspect comprises measuring the presence and/or amount in the test sample of all of the proteins defined in Table 1 - i.e. all 19 of the proteins in Table 1.

Alternatively, the method of the first aspect may comprise measuring the presence and/or amount in the test sample of 1 or 2 or 3 or 4 or 5 or 6 or 7 or 8 or

9 or 10 or 11 or 12 or 13 or 14 or 15 or 16 or 17 or 18 or 19 of the proteins defined in Table 1.

In a preferred embodiment, the method of the first aspect comprises measuring the presence and/or amount in the test sample of Rantes and/or Eotaxin and/or El and/or TNF-b(1 ) and/or TNF-b(2) and/or GLP-1 and/or VEGF and/or IL-13 and/or CD40.

In a second aspect, the invention provides a method for determining the survival time of an individual afflicted with pancreatic adrenocarcinoma comprising the steps of:

i) providing a serum or plasma sample to be tested; ϋ) determining a protein signature of the test sample by measuring the presence and/or amount in the test sample of one or more proteins selected from the group defined in Table 2;

wherein the survival time of an individual is identified in the event that the presence and/or amount in the test sample of one or more proteins selected from the group defined in Table 2 is indicative of a survival time of less than 12 months or longer than 12 months or longer than 24 months.

Preferably, the method according to the second aspect further comprises the steps of: iii) providing a first control serum or plasma sample from an individual having a survival time of less than 12 months and/or a second control serum or plasma sample from an individual having a survival time longer- than 12 months and/or longer than 24 months; iv) determining a protein signature of the first and/or the second control sample by measuring the presence and/or amount of the one or more proteins measured in step (ii);

wherein the survival time of an individual is identified by comparing the presence and/or amount of the one or more proteins in the test sample measured in step (ii) with the presence and/or amount of the one or more proteins in the first and/or second control sample measured in step (iv).

By comparing the presence and/or amount of the selected one or more proteins in the test sample and the control sample, it is possible to determine the survival time of the individual afflicted with pancreatic adrenocarcinoma. For example, if the test sample has the same {i.e. identical) or substantially similar or significantly similar presence and/or amount of the selected one or more proteins as a control sample from a patient known to have a survival time of more than 24 months, the test sample will be determined as a sample from a patient having a survival time of more than 24 months. Other such comparisons will be understood by a person skilled in the art of diagnostics.

Typically, the method of the second aspect comprises measuring the presence and/or amount in the test sample of all of the proteins defined in Table 2 - i.e. all 22 of the proteins in Table 2.

Alternatively, the method of the first aspect may comprise measuring the presence and/or amount in the test sample of 1 or 2 or 3 or 4 or 5 or 6 or 7 or 8 or 9 or 10 or 11 or 12 or 13 or 14 or 15 or 16 or 17 or 18 or 19 or 20 or 21 or 22 of the proteins defined in Table 2.

In a preferred embodiment, the method of the first aspect comprises measuring the presence and/or amount in the test sample of CD40 ligand and/or mucine and/or IL-16 and/or Rantes and/or Eotaxin and/or MCP-4 and/or IL-11 and/or TNF-b and/or IL-1ra and/or MCP-3 and/or IL-Ia and/or IL-3 and/or C3 and/or LDL (1 ) and/or LDL (2) and/or Lewis Y.

Preferably, the first aspect of the invention provides a method wherein step (b) and/or step (d) is performed using a first binding agent capable of binding to the one or more proteins. Preferably, the second aspect of the invention provides a method wherein step (ii) and/or step (iv) is performed using a first binding agent capable of binding to the one or more proteins.

Binding agents (also referred to as binding molecules) can be selected from a library, based on their ability to bind a given motif, as discussed below.

At least one type, more typically all of the types, of the binding molecules may be an antibody or fragments or variants thereof.

Thus, a fragment may contain one or more of the variable heavy (VH) or variable light (V _L ) domains.- For example, the term antibody fragment includes Fab-like molecules (Better et al (1988) Science 240, 1041 ); Fv molecules (Skerra et al ^■ (1988) Science 240, 1038); single-chain Fv (ScFv) molecules where the VH and VL partner domains are linked via a flexible oligopeptide (Bird et a/ (1988) Science 242, 423; Huston et al (1988) Proc. Natl. Acad. ScL USA 85, 5879) and single domain antibodies (dAbs) comprising isolated V domains (Ward et al (1989) Nature 341, 544).

The term "antibody variant" includes any synthetic antibodies, recombinant antibodies or antibody hybrids, such as but not limited to, a single-chain antibody molecule produced by phage-display of immunoglobulin light and/or heavy chain variable and/or constant regions, or other immunointeractive molecule capable of binding to an antigen in an immunoassay format that is known to those skilled in the art.

A general review of the techniques involved in the synthesis of antibody fragments which retain their specific binding sites is to be found in Winter & Milstein (1991 ) Nature 349, 293-299.

Additionally or alternatively at least one type, more typically all of the types, of the binding molecules is an aptamer.

Molecular libraries such as antibody libraries (Clackson et al, 1991 , Nature 352, 624-628; Marks et al, 1991 , J MoI Biol 222(3): 581-97), peptide libraries (Smith,

1985, Science 228(4705): 1315-7), expressed cDNA libraries (Santi et a! (2000) J

MoI Biol 296(2): 497-508), libraries on other scaffolds than the antibody framework such as affibodies (Gunneriusson et al, 1999, Appl Environ Microbiol

65(9): 4134-40) or libraries based on aptamers (Kenan et al, 1999, Methods MoI Biol 118, 217-31) may be used as a source from which binding molecules that are specific for a given motif are selected for use in the methods of the invention.

The molecular libraries may be expressed in vivo in prokaryotic (Clackson et al,

1991 , op. cit; Marks et al, 1991 , op. cit.) or eukaryotic cells (Kieke et al, 1999, Proc Natl Acad Sc/ USA, 96(10):5651-6) or may be expressed in vitro without involvement of cells (Hanes & Pluckthun, 1997, Proc Natl Acad Sci USA

94(10):4937-42; He & Taussig, 1997, Nucleic Acids Res 25(24):5132-4; Nemoto et al, 1997, FEBS Lett, 414(2):405-8).

In cases when protein based libraries are used often the genes encoding the libraries of potential binding molecules are packaged in viruses and the potential binding molecule is displayed at the surface of the virus (Clackson et al, 1991 , op. cit; Marks et al, 1991, op. cit, Smith, 1985, op. cit).

The most commonly used such system today is filamentous bacteriophage displaying antibody fragments at their surfaces, the antibody fragments being expressed as a fusion to the minor coat protein of the bacteriophage (Clackson et al, 1991, op. cit] Marks et al, 1991 , op. cit). However, also other systems for display using other viruses (EP 39578), bacteria (Gunneriusson et al, 1999, op. cit; Daughterly et al, 1998, Protein Eng 11(9):825-32; Daugherty et a/, 1999, Protein Eng 12(7):613-21 ), and yeast (Shusta et al, 1999, J MoI Biol 292(5):949-

56) have been used.

In addition, recently, display systems utilising linkage of the polypeptide product to its encoding mRNA in so called ribosome display systems (Hanes & Pluckthun, 1997, op. cit; He & Taussig, 1997, op. cit; Nemoto et al, 1997, op. cit), or alternatively linkage of the polypeptide product to the encoding DNA (see US Patent No. 5,856,090 and WO 98/37186) have been presented.

When potential binding molecules are selected from libraries one or a few selector peptides having defined motifs are usually employed. Amino acid residues that provide structure, decreasing flexibility in the peptide or charged, polar or hydrophobic side chains allowing interaction with the binding molecule may be used in the design of motifs for selector peptides. For example - (i) Proline may stabilise a peptide structure as its side chain is bound both to the alpha carbon as well as the nitrogen;

(ii) Phenylalanine, tyrosine and tryptophan have aromatic side chains and are highly hydrophobic, whereas leucine and isoleucine have aliphatic side chains and are also hydrophobic;

(iii) Lysine, arginine and histidine have basic side chains and will.be positively charged at neutral pH, whereas aspartate and glutamate have acidic side chains and will be negatively charged at neutral pH;

(iv) Asparagine and glutamiπe are neutral at neutral pH but contain a amide group which may participate in hydrogen bonds; (v) Serine, threonine and tyrosine side chains contain hydroxyl groups, which may participate in hydrogen bonds. •

Typically selection of binding molecules may involve the use of array technologies and systems to analyse binding to spots corresponding to types of binding molecules.

Preferably, the first binding agent is an antibody or a fragment thereof; more preferably, a recombinant antibody or fragment thereof. Conveniently, the antibody or fragment thereof is selected from the group consisting of: scFv; Fab; a binding domain of an immunoglobulin molecule.

The variable heavy (V _H ) and variable light (V _L ) domains of the antibody are involved in antigen recognition, a fact first recognised by early protease digestion experiments. Further confirmation was found by "humanisation" of rodent antibodies. Variable domains of rodent origin may be fused to constant domains of human origin such that the resultant antibody retains the antigenic specificity of the rodent parented antibody (Morrison et al (1984) Proc. Natl. Acad. Sci. USA 81, 6851-6855).

That antigenic specificity is conferred by variable domains and is independent of the constant domains is known from experiments involving the bacterial expression of antibody fragments, all containing one or more variable domains. These molecules include Fab-like molecules (Better et al (1988) Science 240, 1041); Fv molecules (Skerra et al (1988) Science 240, 1038); single-chain Fv (ScFv) molecules where the VH and V _L partner domains are linked via a flexible oligopeptide (Bird et al (1988) Science 242, 423; Huston et al (1988) Proc. Natl. Acad. Sci. USA 85, 5879) and single domain antibodies (dAbs) comprising isolated V domains (Ward et al (1989) Nature 341, 544). A general review of the techniques involved in the synthesis of antibody fragments which retain their specific binding sites is to be found in Winter & Milstein (1991) Nature 349, 293-299.

By "ScFv molecules" we mean molecules wherein the V _H and VL partner domains are linked via a flexible oligopeptide.

The advantages of using antibody fragments, rather than whole antibodies, are several-fold. The smaller size of the fragments may lead to improved pharmacological properties, such as better penetration of solid tissue. Effector functions of whole antibodies, such as complement binding, are removed. Fab, Fv, ScFv and dAb antibody fragments can ali be expressed in and secreted from E. coli, thus allowing the facile production of large amounts of the said fragments.

Whole antibodies, and F(ab') ₂ fragments are "bivalent". By "bivalent" we mean that the said antibodies and F(ab') ₂ fragments have two antigen combining sites. In contrast, Fab, Fv, ScFv and dAb fragments are monovalent, having only one antigen combining sites.

The antibodies may be monoclonal or polyclonal. Suitable monoclonal antibodies may be prepared by known techniques, for example those disclosed in "Monoclonal Antibodies: A manual of techniques", H Zola (CRC Press, 1988) and in "Monoclonal Hybridoma Antibodies: Techniques and applications", J G R Hurrell (CRC Press, 1982), both of which are incorporated herein by reference.

In a preferred embodiment, the invention provides methods wherein the one or more proteins in the test sample is labelled with a detectable moiety. Preferably, the first aspect provides a method wherein the one or more proteins in the control sample is labelled with a detectable moiety. Alternatively, in the second aspect the one or more proteins in the first and/or second control sample is labelled with a detectable moiety.

By a "detectable moiety" we include the meaning that the moiety is one which may be detected and the relative amount and/or location of the moiety (for example, the location on an array) determined.

Detectable moieties are well known in the art.

A detectable moiety may be a fluorescent and/or luminescent and/or chemiluminescent moiety which, when exposed to specific conditions, may be detected. For example, a fluorescent moiety may need to be exposed to radiation (i.e. light) at a specific wavelength and intensity to cause excitation of the fluorescent moiety, thereby enabling it to emit detectable fluorescence at a specific wavelength that may be detected.

Alternatively, the detectable moiety may be an enzyme which is capable of converting a (preferably undetectable) substrate into a detectable product that can be visualised and/or detected. Examples of suitable enzymes are discussed in more detail below in relation to, for example, ELISA assays.

Alternatively, the detectable moiety may be a radioactive atom which is useful in imaging. Suitable radioactive atoms include ^99m Tc and ¹²³ I for scintigraphic studies. Other readily detectable moieties include, for example, spin labels for magnetic resonance imaging (MRI) such as ¹²³ I again, ¹³¹ 1, ¹¹¹ In, ¹⁹ F, ¹³ C, ¹⁵ N, ¹⁷ O, gadolinium, manganese or iron. Clearly, the agent to be detected (such as, for example, the one or more proteins in the test sample and/or control sample described herein and/or an antibody molecule for use in detecting a selected protein) must have sufficient of the appropriate atomic isotopes in order for the detectable moiety to be readily detectable.

The radio- or other labels may be incorporated into the agents of the invention (i.e. the proteins present in the samples of the methods of the invention and/or the binding agents of the invention) in known ways. For example, if the binding moiety is a polypeptide it may be biosynthesised or may be synthesised by chemical amino acid synthesis using suitable amino acid precursors involving, for example, fluorine-19 in place of hydrogen. Labels such as ^99m Tc, ¹²³ 1, ¹⁸⁶ Rh, ¹⁸⁸ Rh and ¹¹¹ In can, for example, be attached via cysteine residues in the binding moiety. Yttrium-90 can be attached via a lysine residue. The IODOGEN method (Fraker et al (1978) Biochem. Biophys. Res. Comm. 80, 49-57) can be used to incorporate ¹²³ I. Reference ("Monoclonal Antibodies in Immunoscintigraphy", J-F Chatal, CRC Press, 1989) describes other methods in detail. Methods for conjugating other detectable moieties (such as enzymatic, fluorescent, luminescent, chemiiuminescent or radioactive moieties) to proteins are well known in the art.

The accompanying Examples provide an example of methods and detectable moieties for labelling agents of the invention (such as, for example, proteins in the sample of the methods of the invention and/or binding molecules) for use in the methods of the invention.

Preferably, the detectable moiety is selected from the group consisting of: a fluorescent moiety; a luminescent moiety; a chemiluminescent moiety; a radioactive moiety; an enzymatic moiety.

Preferably, the first aspect provides a method wherein step (b) and/or step (d) is performed using an array. Conveniently, in the second aspect, step (ii) and/or step (iv) is performed using an array.

Arrays per se are well known in the art. Typically they are formed of a linear or two-dimensional structure having spaced apart (i.e. discrete) regions ("spots"), each having a finite area, formed on the surface of a solid support. An array can also be a bead structure where each bead can be identified by a molecular code or colour code or identified in a continuous flow. Analysis can also be performed sequentially where the sample is passed over a series of spots each adsorbing the class of molecules from the solution. The solid support is typically glass or a polymer, the most commonly used polymers being cellulose, polyacrylamide, nylon, polystyrene, polyvinyl chloride or polypropylene. The solid supports may be in the form of tubes, beads, discs, silicon chips, microplates, polyvinylidene difluoride (PVDF) membrane, nitrocellulose membrane, nylon membrane, other porous membrane, non-porous membrane (e.g. plastic, polymer, perspex, silicon, amongst others), a plurality of polymeric pins, or a plurality of microtitre wells, or any other surface suitable for immobilising proteins, polynucleotides and other suitable molecules and/or conducting an immunoassay. The binding processes are well known in the art and generally consist of cross-linking covalently binding or physically adsorbing a protein molecule, polynucleotide or the like to the solid support. By using well-known techniques, such as contact or non-contact printing, masking or photolithography, the location of each spot can be defined. For reviews see Jenkins, R.E., Pennington, S. R. (2001 , Proteomics, 2,13-29) and LaI et ai (2002, Drug Discov Today 15;7(18 Suppl):S143-9).

Typically the array is a microarray. By "microarray" we include the meaning of an array of regions having a density of discrete regions of at least about 100/cm ² , and preferably at least about 1000/cm ² . The regions in a microarray have typical dimensions, e.g., diameters, in the range of between about 10-250 μm, and are separated from other regions in the array by about the same distance. The array may also be a macroarray or a nanoarray.

Once suitable binding molecules (discussed above) have been identified and isolated, the skilled person can manufacture an array using methods well known in the art of molecular biology.

Alternatively, the first aspect of the invention provides a method wherein step (b) and/or step (d) is performed using an assay comprising a second binding agent capable of binding to the one or more proteins, the second binding agent having a detectable moiety.

Typically, in the second aspect, step (ii) and/or step (iv) is performed using an assay comprising a second binding agent capable of binding to the one or more proteins, the second binding agent having a detectable moiety.

Binding agents are described ih detail above.

In a preferred embodiment, the second binding agent is an antibody or a fragment thereof; typically a recombinant antibody or fragment thereof. Conveniently, the antibody or fragment thereof is selected from the group consisting of: scFv; Fab; a binding domain of an immunoglobulin molecule. Antibodies are described in detail above.

It is preferred that, where an assay is used, the invention provides a method wherein the detectable moiety is selected from the group consisting of: a fluorescent moiety; a luminescent moiety; a chemiluminescent moiety; a radioactive moiety; an enzymatic moiety. Examples of suitable detectable

^■ moieties for use in the methods of the invention are described above. ,

Preferred assays for detecting serum or plasma proteins include enzyme linked immunosorbent assays (ELISA), radioimmunoassay (RIA), immunoradiometric ^■ assays (IRMA) and immunoenzymatic assays (IEMA), including sandwich assays , using monoclonal and/or polyclonal antibodies. Exemplary sandwich assays are described by David et al in US Patent Nos. 4,376,110 and 4,486,530, hereby incorporated by reference. Antibody staining of cells on slides may be used in methods well known in cytology laboratory diagnostic tests, as well known to those skilled in the art.

Typically, the assay is an ELISA (Enzyme Linked Immunosorbent Assay) which typically involve the use of enzymes which give a coloured reaction product, usually in solid phase assays. Enzymes such as horseradish peroxidase and phosphatase have been widely employed. A way of amplifying the phosphatase reaction is to use NADP as a substrate to generate NAD which now acts as a coenzyme for a second enzyme system. Pyrophosphatase from Escherichia coii provides a good conjugate because the enzyme is not present in tissues, is . stable and gives a good reaction colour. Chemi-luminescent systems based on enzymes such as luciferase can also be used.

Conjugation with the vitamin biotin is frequently used since this can readily be detected by its reaction with enzyme-linked avidin or streptavidin to which it binds with great specificity and affinity.

In a third embodiment, the invention provides an array for determining the presence of pancreatic adrenocarcinoma in an individual comprising one or more binding agent according to the invention. Preferably, the one or more binding agent is capable of binding to all of the proteins defined in Table 1.

In a fourth embodiment, the invention provides an array for determining the survival time of an individual afflicted with pancreatic adrenocarcinoma comprising one or more binding agent according to the invention. Preferably, the one or more binding agent is capable of binding to all of the proteins defined in _. Table 2.

Arrays suitable for use in the methods of the invention are discussed above. In a further embodiment, the invention provides the use of an array in the methods of the first and/or second aspects of the invention.

In a fifth embodiment, the invention provides the use of one or more proteins selected from the group defined in Table 1 as a diagnostic marker for determining the presence of pancreatic adrenocarcinoma in an individual. Conveniently, all of the proteins defined in Table 1 are used as a diagnostic marker for determining the presence of pancreatic adrenocarcinoma in an individual.

In a sixth embodiment, the invention provides the use of one or more proteins selected from the group defined in Table 2 as a diagnostic marker for determining

the survival time of an individual afflicted with pancreatic adrenocarcinoma. It is preferred that all of the proteins defined in Table 2 are used as a diagnostic marker for determining the survival time of an individual afflicted with pancreatic adrenocarcinoma.

In a seventh embodiment, there is provided a kit for determining the presence of pancreatic adrenocarcinoma comprising:

A) one or more first binding agent or an array according to the invention;

B) instructions for performing the method according to the first aspect of the invention.

In an eighth aspect, the invention provides a kit for determining the presence of pancreatic adrenocarcinoma comprising:

A) one or more second binding agent as defined herein

B) instructions for performing the method as defined in the first aspect of the invention.

In a ninth aspect, there is provided a kit for determining the survival time of an individual afflicted with pancreatic adrenocarcinoma comprising:

1) one or more first binding agent as defined herein; 2) instructions for performing the method according to the second aspect of the invention. .

In a tenth aspect the invention provides a kit for determining the survival time of an individual afflicted with pancreatic adrenocarcinoma comprising: • ^{' • '} .

1) one or more second binding agent as defined herein;

2) instructions for performing the method as defined in the second aspect of the invention.

The listing or discussion of a prior-published document in this specification should not necessarily be taken as an acknowledgement that the document is part of the ^■ state of the art or is common general knowledge.

Preferred, non-limiting examples which embody certain aspects of the invention wilt now be described, with reference to the following tables and figures: .

Table 1: Serum protein profile for distinguishing normal vs. pancreatic cancer patients

Table 2: Serum protein signature for distinguishing short-term survivors (<12months) vs. long-term survivors (>24months) among the pancreatic cancer patients

Table 3: 129 recombinant antibody fragments, directed against 60 serum proteins, for use in microarray of the invention

Table 4: Patient demographics of patients from whom serum samples were obtained. PC= pancreatic adenocarcinoma.

Figure 1: Detection of pancreatic adenocarcinomas by serum protein expression analysis, using recombinant antibody microarrays. (a) A scanned antibody microarray image containing 1280 data points; (b) A multidimensional analysis represented as an unsupervised Sammon plot based on all 129 antibody fragments, where cancer patients (red) are shown to be completely separated from healthy individuals (blue); (c) A dendrogram, where cancer patients (PA) are completely separated from normal individuals (N); The two-way hierarchical clustering was based on 19 serum biomarkers that were significantly (p<0.05) differentially expressed in cancer vs. normal individuals, using a training set composed of 28 serum samples. Subsequently, a test set of 16 serum samples (marked with *) were 100% correctly classified. Columns represents donors, where blue is normal individuals (N) and red is cancer patients (PA). Each row represents a serum biomarker, as denoted on the right hand side, where each pixel demonstrates the expression level of that particular biomarker in each donor (overexpression (red), underexpression (green) or no change (black) in pancreatic cancer sera vs. normal sera.); (d) Several of the serum biomarkers, such as IL-4, IL-5, IL-13, and MCP-3, were also analyzed by ELlSA to confirm the microarray results. A representative data set is shown for IL-13, demonstrating that conventional ELISA and antibody microarray analysis generated similar

results. The sensitivity of the microarray analysis was equal to or better compared to what was obtained by ELISA (data not shown).

Figure 2: Identification of a predictive serum protein biomarker signature, discriminating between two patient cohorts of short (<12 months) vs. long (>24 months) survivors, (a) A Receiver Operator Curve (ROC) area as a function of the number of analytes included in a predictive signature, which clearly demonstrates that the two cohorts of survivors could be well discriminated using a signature >29 analytes; (b) The ROC area of a predictive serum biomarker signature, based on 29 antibody identified analytes; (c) A SVM was trained with the biomarker signature chosen by the training set. A test set consisting of 10 randomly chosen patients (samples marked with *) was then classified, using the SVM Prediction Value; (d) A heat map based on the 22 non-redundant serum proteins in the predictive signature. The columns represents cancer patients, where blue is long (>24 months) survivors and red is short (<12 months) survivors. See legend to Figure 1c for color coding.

Figure 3: Principle of the recombinant antibody microarray technology.

EXAMPLE

Overview

The driving force behind oncoproteomics is to identify protein signatures that are associated with a particular malignancy. Based on recombinant antibody microarray analysis of unfractionated human whole serum proteomes, derived from pancreatic carcinomas and normal healthy donors, we have identified a protein signature, based on 22 non-redundant analytes, discriminating between cancer and healthy patients. The specificity and sensitivity were predicted to be 99.7 and 99.9 %, respectively. Furthermore, a protein signature consisting of 19 protein analytes was defined that had the potential to predicted survival amongst cancer patients. This novel predictor signature distinguished between patients having <12 months or >24 months survival time and suggests new possibilities in individualized medicine.

The present study describes an affinity proteomic approach to prognostic diagnosis of cancer based on a recombinant antibody microarray, utilizing array adapted recombinant scFv fragments (12,13). The results demonstrated that an array of antibody fragments, specific for immunoregulatory proteins, can discriminate between human serum proteomes derived from either cancer patients or healthy individuals. We present the first sets of serum biomarkers for detection of pancreatic cancer as well as for predicting patient survival.

Materials and Methods

Production and purification of scFv - 129 human recombinant scFv antibody fragments against 60 different proteins mainly involved in immunregulation, were selected from the n-CoDeR library (13) and' kindly provided by Biolnvent Internationa! AB (Lund, Sweden). Thus, each antigen was recognized by up to four different scFv fragments. All scFv antibodies were produced in 100 ml E. coli cultures and purified from expression supernatants, using affinity chromatography on Ni-NTA agarose (Qiagen, Hilden, Germany). Bound molecules were eluted with 250 mM imidazole, extensively dialyzed against PBS, and stored at 4°C, until further use. The protein concentration was determined by measuring the

absorbance at 280 nm (average concentration 210 μg/ml, range 60-1090 μg/ml). The purity was evaluated by 10 % SDS-PAGE (Invitrogen, Carlsbad, CA, USA).

Serum Samples - in total, 44 serum samples supplied by Stockholm South General Hospital (Sweden) and Lund University Hospital (Lund, Sweden) were included in this study. 24 serum samples (PA1-PA30) were collected from patients with pancreatic cancer at the time of diagnosis. 20 serum samples (N 1-

N20) (no clinical symptoms) were collected from healthy donors. Patient demographics are shown in Table 4. All samples were aliquoted and stored at - 80 °C, following standard operating procedures.

Labeling of serum samples - The serum samples were labeled using previously optimized labeling protocols for serum proteomes (9,14,15). All serum samples were biotinylated using the EZ-Link Sulfo-NHS-LC-Biotin (Pierce, Rockford, IL, USA). 50 μl serum aliquots were centrifuged at 16.000 x g for 20 minutes at 4 DC and diluted 1 :45 in PBS, resulting in a concentration of about 2 mg/ml. The samples were then biotinylated by adding Sulfo-NHS-biotin to a final concentration of 0.6 mM for 2 h on ice, with careful Vortexing every 20 minute. Unreacted biotin was removed by dialysis against PBS for 72 hours, using a 3.5 kDa MW dialysis membrane (Spectrum Laboratories, Rancho Dominguez, CA, USA). The samples were aliquoted and stored at -20 ° ⁿ o.

Enzyme linked immunosorbent assay - The serum concentration of 4 protein analytes (MCP-3, lL-4, IL-5 and 1L-13) were measured in all samples, using commercial ELlSA kits (Quantikine, R&D Systems, Minneapolis, MN, USA). The measurements were performed according to the recommendations provided by the supplier.

Fabrication and processing of antibody microarrays - For production of the antibody microarrays, we used a set-up previously optimized and validated

(9,12,14,15). Briefly, the scFv microarrays were fabricated, using a non-contact printer (Biochip Arrayer, Perkin Elmer Life & Analytical Sciences), which deposits approximately 330 pL/drop, using piezo technology. The scFv antibodies were arrayed by spotting 2 drops at each position and the first drop was allowed to dry out before the second drop was dispensed. The antibodies were spotted onto black polymer MaxiSorb microarray slides (NUNC AJS, Roskilde, Denmark), resulting in an average of 5 fmol scFv per spot (range 1.5 - 25 fmol). Eight

replicates of each scFv clone were arrayed to ensure adequate statistics. In total, 160 antibodies and controls were printed per slide orientated in two columns with . 8 x 80 antibodies per column. To assist the alignment of the grid during the quantification a row containing Cy5 conjugated streptavidin (2 μg/ml) was spotted for every tenth row. A hydrophobic pen (DakoCytomation Pen, DakoCytomation, Glostrup, Denmark) was used to draw a hydrophobic barrier around the arrays. The arrays were blocked with 500 μl 5% (w/v) fat-free milk powder (Semper AB, Sundbyberg, Sweden) in PBS overnight. All incubations were conducted in a humidity chamber at room temperature. The arrays were then washed four times with 400 μl 0.05% Tween-20 in PBS (PBS-T), and incubated with 350 μl biotinylated serum, diluted 1 :10 (resulting in a total serum dilution of 1 :450) in 1% (w/v) fat-free milk powder and 1% Tween in PBS (PBS-MT) for 1h. Next, the arrays were washed four times with 400 μl PBS-T and incubated with 350 μl 1 μg/ml Alexa- 647 conjugated streptavidin, diluted in PBS-MT for 1 h. Finally, the arrays were washed four times with 400 μl PBS-T, dried immediately under a stream of nitrogen gas and scanned with a confocal microarray scanner (ScanArray Express, Perkin Elmer Life & Analytical Sciences) at 5 μm resolution, using six different scanner settings. The ScanArray Express software V2.0 (Perkin Elmer Life & Analytical Sciences) was used to quantify the intensity of each spot, using the fixed circle method. The local background was subtracted and to compensate for possible local defects, the two highest and the two lowest replicates were automatically excluded and each data point represents the mean value of the remaining four replicates. The coefficient of correlation for intra- assays was >0.99 and for inter-assays >0.96, respectively.

Data normalization - Only non-saturated spots were used for further analysis of the data. Chip-to-chip normalization of the data sets was performed, using a semi-global normalization approach, conceptually similar to the normalization developed for DNA microarrays. Thus, the coefficient of variation (CV) was first calculated for each analyte and ranked. Fifteen % of the analytes that displayed the lowest CV-values over all samples were identified, corresponding to 21 analytes, and used to calculate a chip-to-chip normalization factor. The normalization factor N ₁ was calculated by the formula U ₁ = S,/μ, where S ₁ - is the sum of the signal intensities for the 21 analytes for each sample and μ is the sum of the signal intensities for the 21 analytes averaged over all samples. Each data-set generated from one sample was divided with the normalization factor Nj. For the intensities, Iog2 values were, used in the analysis.

Data analysis - The Sammon map was performed using Euclidean distance in the space of all 129 analytes. Supervised classification was done with a Support Vector Machine (SVM) using a linear kernel (16-18). The cost of constraints violation (the parameter C in the SVM) was fixed to 1 , which is the default value in the R function svm, and no attempt was done to tune it. This absence of parameter tuning was chosen to avoid overfitting and to make the classification procedure easier to understand. The output of the SVM on a test sample is a SVM decision value, which is the signed distance to the hyperplane. In Figure 1C and 2C, the split into training and test set was done randomly once and kept fixed

-from thereon. In Figure 2A, a leave-one-out cross validation procedure is used.

For every number K between 1 and 129 the following procedure was carried out.

For a training set, i.e., all samples except one, the K highest ranked analytes with a Wilcoxon test were chosen, and a SVM was trained with those K analytes. A SVM decision value was then calculated for the left out sample with this classifier. As is common practice, this was done for all samples in the leave-one-out cross validation.

A Receiver Operating Characteristics (ROC) curve was constructed using the SVM decision values and the area under the curve was found. Figure 2A shows the ROC area as a function of K. Figure 2B shows the ROC curve for the value K = 29. All statistics were done in R (19).

Results

Pancreatic ductal adenocarcinoma is a cancer with poor prognosis and improved diagnostic tool facilitating the clinical decision making would significantly benefit the patients. One approach to improved ¹ diagnosis is to identify a set of biomarkers that can detect cancer and that also is predict clinical outcome. Consequently, to be able to identify a protein signature linked to pancreatic cancer with high sensitivity, we have designed the first large-scale microarray (Figure 1A) based on 129 recombinant antibody fragments (12,14,15), directed against 60 serum proteins, mainly of immunoregulatory nature (Table 2). In this study, labeled sera from 24 pancreatic cancer patients and 20 healthy patients were incubated on the antibody microarrays, which subsequently were quantified, using a confocal scanner. First, to test our ability to detect cancer, the microarray

data was displayed in an unsupervised Sammon plot based on all antibodies and two distinct populations could clearly be distinguished (Figure 1B). This indicated the existence of a clear difference between the cancer and the normal proteomes, in relation to the serum analytes analyzed by the microarray. We subsequently ran a . leave-one-out cross-validation, with a Support Vector Machine (SVM), and collected the decision values for each sample. The decision value is the output of the predictor, and samples with a prediction value above (below) a threshold are predicted to be pancreatic carcinomas (healthy). The threshold parameterizes the trade-off between sensitivity and specificity and is often, but not always, set to zero. The 24 pancreatic carcinoma samples obtained decision values in the interval from 0.30 to 1.93, and the healthy samples in the interval from -1.84 to -0.30. Thus, with a threshold value of zero, or any other value between -0.30 and 0.30, the sensitivity and specificity is 100% in our data set. However, to extrapolate the sensitivity and specificity to a larger population, we first verified that the decision values were approximately normally distributed, within the normal vs. cancer groups, respectively and calculated the means and variances. By setting the classification threshold halfway between the two means and using normal distributions, we found a 99.9% sensitivity and 99.3% specificity, which indicated excellent classification power even in a larger population. To illustrate the clear separation between the normal and cancer group, we randomly selected a training set, consisting of 18 cancer and 10 normal samples. This training set of cancer and normal serum proteomes defined a smaller set of biomarkers, consisting of 19 non-redundant serum proteins that differed significantly (p<0.05) between the two samples. These differentially expressed proteins were subsequently used to construct a dendrogram of the 28 training samples and the 16 remaining samples, which were used as a test set. As can be seen in Figure 1 C, the cancer samples are completely separated from the normal samples for both the training and test set

^• . . (100% sensitivity and specificity).

An interesting observation was the fact that we had blindly obtained three serum samples from one patient (PA14), drawn at different occasions a few weeks apart but 11-12 months before that patient was diagnosed with pancreatic cancer. Still, all samples were correctly classified as cancer, when used in the test set (data not shown). Importantly, the protein signature, defined by the training and used for classification of the test samples, is specific for pancreatic adenocarcinomas and differs from serum signatures found by our microarray set-up in other

cancers, such as gastric (9) and breast adenocarcinomas (manuscript in preparation). Some of the microarray data was also confirmed by analyzing several of. the serum proteins by conventional enzyme . linked immunosorbent assay (Figure 1 D). However, analysis based on microarray measurements is often more sensitive, compared to traditional enzyme immunoassays. Consequently, analytes could only be validated when the ELISA sensitive was enough, but then our microarray data was confirmed.

While an early detection of cancer has its merits, especially in pancreatic cancer, serum protein profiling has also been suggested as the approach to define signatures that, apart from classify cancer vs. normal, also could be associated with clinical parameters (16). To be able to predict expected survival time would be of high relevance, since this could influence the therapeutic regimes assigned to each patient. Consequently, to further interrogate our recombinant antibody microarray platform, we compared two cohorts of cancer patients divided into short survivors (<12 months) vs. long survivors (>24 months). First, calculated the area under the receiver operator characteristic (ROC) curves, as a function of the total number of antibody-defined analytes in a predictive signature, using a Wilcoxon test to filter analytes, followed by a Support Vector Machine (Figure 2A). These calculations included all 129 antibodies and since we had 1 to 4 antibody/serum analyte, a certain redundancy was present in the biomarker size of the predictive signature. From these calculations, it was evident that the two cohorts could be discriminated, with a ROC area (AUC) of >0.80. Of note, this curve also demonstrated that a protein signature consisting of <26 analytes provided a more variable and less robust predictor. Consequently, we chose a predictor signature consisting of 29 analytes, for further analysis. The ROC curve for 29 analytes has an area under the curve of 0.86 (Figure 2B).

Again, to illustrate the predictive ability of this biomarker signature the pancreatic cancer patients (n = 23), comprising the short and long term survivors, were . randomly split into a training set of 13 patients and a test set of 10 patients.

Since the mean survival for patients with unresectable disease remains 5-6 months there was an inevitable bias in cohort size, and the long-term survivor cohort consisted of only 5 patients. A SVM was trained with the biomarker signature chosen by the training set and the test set could then be classified, as shown in Figure 2C. All patients surviving <12 months were correctly classified, using a SVM prediction value of <0, which was considered the most important

classification. One long-term survivor was miss-classified. The 29 most significant analytes separating long and short term survivors among all 23 patients in a Wilcoxon test corresponds to 22 non redundant serum proteins (7 of the 29 analytes were duplicates but defined by different antibody clones). This novel predictor signature, represented by 22 non-redundant proteins, and the differential analyte response displayed by short and long survivors, respectively, are shown as a heat map in Figure 2D. When analyzing the individual proteins there was no strict consensus patter among the serum proteins, although it was evident that cytokines, such as IL-Ia, IL-3, IL-8, and IL-11 were upregulated in short term survivors, while Rantes, IL-16, IL-4 and eotaxin were mostly upregulated in long term survivors (Figure 2D). The significance of this remains to be validated but it could possibly indicating a more active T-cell compartment in the latter population.

Discussion

Antibody microarrays, as a tool in affinity proteomics, have evolved over the last several years from a promising tool to an approach that is starting to deliver promising results in oncoproteomics (3, 12, 20, 21 ). The main focus in these endeavors is to detect cancer at an early stage, to predict tumor relapse and treatment resistance, or to select patients for a particular treatment regime (3). This is in particular important for cancers with poor prognosis, which is also intrinsic to pancreatic cancer since it rapidly metastasize to e.g. lymph nodes, lungs, peritoneum (4, 23) and is difficult to diagnose at an early stage. However, the ability of a biomarker signature to distinguish between different carcinomas or between cancer and inflammation has so far been difficult to achieve (for review see ref. 3)(20). The reason for the observed distinction between cancer and normal serum proteomes in this study is most likely dependent on the range of antibody specificities on the microarray, which is also recently supported by the rationally designed array, reported by Sanchez-Carbayo et al. (21 ). These investigators could stratify patients with bladder tumors on the basis of their overall survival, using antibodies generated against differentially expressed gene products. During the last years, we have developed a high-performing, recombinant antibody microarray platform for complex proteome analysis (6, 9, 11 , 12, 14, 15), by evaluating and optimizing key technological parameters (24), such as probe and substrate design (13, 25), array/assay design (9, 15) and

sample format (9, 14, 15). This has allowed us to perform the first differential protein expression profiling of the human plasma proteome, using the optimized scFv microarrays, targeting mainly immunoregulatory proteins. In agreement with previous result, this antibody microarrays displayed sensitivities in the pM to fM range, readily detecting low-abundant cytokines. Furthermore, we maintained an assay reproducibility with a coefficient of correlation in the range of 0.96 - 0.99, which is a key feature of multiplexed analysis and which compares well with previous reports (12, 26). Moreover, the antibody microarray data was compared to ELISA and, when sensitive enough, this conventional assay _. corroborated our results.

Patients with pancreatic cancer are often diagnosed late, resulting in a poor prognosis. Due to low incidence it is difficult to gather large sample numbers, especially for long-term survivors, i.e. >24 months. We had access to 25 patients for this study, which using a rigorous statistical evaluation still allowed us to classify cancer and normal proteomes. This is a supervised classification and we have employed a Support Vector Machine as the classifier, although we obtained very similar results for this data set with a naive Bayesian classifier (data not shown). The SVM separated the two groups by finding a hyperplane in space of all analytes, and assigned samples on one side of the hyperplane to one of the groups and those on the other side to the other group. The distance to the hyperplane is called the prediction or decision value (Figure 2C). The hyperplane and, thus, the classification of groups, were found by using our training set. The performance of the classifier was then estimated by subsequently utilizing a test set, where no overlap between the training and test set was allowed. However, a data set can randomly be split into different training and a test set, which are then used to train and test the classifier, respectively. The drawback of this is that the final result depends on the split into training and test set. Consequently, we used cross validation as the procedure of making several splits of our data set and used the average performance of the test sets as a. measure of the accuracy of data classification. Thus, in the leave-one-out cross validation that was performed, the test set contains one sample and the training set contains the rest. The performance of the SVM can be measured by the ROC curve and, in particular, the area under the ROC curve. The normal and pancreatic carcinoma samples were remarkably well separated, since the SVM classified all samples correctly with a gap between the two groups. Extrapolation of the decision values

gave very high sensitivity (99.9%) and specificity (99.3%), showing that it would take hundreds of samples to get one misclassification.

In this study, we could not compare the pancreatic cancers with a cohort of patients with pancreatitis, which would have been a desirable comparison, but we instead used normal serum samples. Of note, the present pancreatic cancer associated biomarker signature had, however, only eotaxin, IL-5 and IL-13 in common with fourteen biomarkers found as a result of a bacterial infection, associated with another gastrointestinal cancer (12), which indicated that the pancreatic signature was not related to general inflammation. Furthermore, this signature was not similar to biomarkers found in systemic lupus erythematosus, an autoimmune disorder with a significant inflammatory component (Wingren et a!., manuscript in preparation). The signature was also completely different from what Orchekowski et aL reported (26), when profiling pancreatic cancer serum samples, using a microarray based on monoclonal and polyclonal antibodies. They targeted high-abundant serum proteins, such as albumin, transferrin and hemoglobin, as well as more common inflammation markers, such as C-reactive protein (CRP), serum amyloid A and immunoglobulins, whereas only eight cytokines were analyzed. On the other hand, our present cancer signature contained a number of over-expressed TH2 cytokines (IL-4, -5, -10 and -13), whereas classical TH1 cytokines (IL-12 and TNF-b) were down-regulated, which also was in agreement with the study of Belone et al., who showed that TGF-b and IL-10 were up-regulated in pancreatic cancer sera (27). These authors also showed that blood-derived monocytes from pancreatic cancer patients were primed to develop a TH2-like response rather than a TH 1 -like response, with increased expression of IL-4 and decreased expression of IL-12. .

Finally, we investigated the possibility to identify a signature, that apart from being able to. classify cancer vs. normal samples, also could be used to predict patient survival. Initially, the SVM could classify the short and long survivors with a ROC area of 0.81, using all analytes (data not shown), which was very promising. Then a classifier was made for every number of biomarkers, by selecting the most significant analytes, which subsequently was used to distinguishing the two sample groups in the training set. As seen in Figure 2A, the performance of the classifier was stable above 26 analytes, and we could demonstrate that a 29 biomarker (22 non-redundant analytes) signature gave a ROC of 0.86. A study with more than 18 short survivors and 5 long survivors is,

however, needed to firmly establish a survival classifying protein profile, but this study certainly establishes the possibility for such a profile.

In conclusion, using a recombinant antibody microarray against , immunoreguiatory proteins, we have been able to specifically detect pancreatic adenocarcinomas and completely discriminate between cancer vs. normal serum proteomes. More importantly, the first attempt to define a signature capable of predicting survival of cancer patients is presented, indicating the power of affinity oncoproteomics for clinical decision making.

References

I . Rosenwald A. et a/. Cancer Cell 3, 195-197 (2003). 2. van de Vijver, et al. N. Eng. J. Med 347, 1999-2009 (2002).

3. Borrebaeck, C.A.K. Expert Opin. Biol. Ther β, 833-838 (2006).

4. Garcea, G., Neal, C.P., Pattenden, CJ. , Steward, WP. & Berry, D.P. Eur. J. Cancer 41, 2213-2236 (2005).

5. Yeo, T. P. etal. Curr. Probl. Cancer26, 176-275 (2002). 6. Wingren, C. & Borrebaeck, C.A.K.. Exp. Rev. Proteomics 1, 355-364 (2004).

7. Pavlickova, P., Schneider, M.E., & Hug, H., CHn. Chim. Acta 342, 17-35 (2004).

8. Haab, BB, et a/. Genome Biol. 2, 1 -13 (2001 ). 9. Wingren, C, et al., Microarrays based on affinity-tagged single-chain Fv antibodies: sensitive detection of analyte in complex proteomes. Proteomic 5, 1281-1291 (2005). 10. Pawlak, M. et al. Proteomics 2, 383-393 (2002).

I 1. Wingren, C. & Borrebaeck, C.A.K. OMICS 3, 411 -427 (2006). 12. Ellmark, P., et al., Identification of protein expression signatures associated with H. pylori infection and gastric adenocarcinoma using recombinant antibody microarrays. MoI Cell Proteomics 5, 1638-1646 (2006).

13. Soderlind, E., et al., Recombining germline-derived CDR sequences for creating diverse single-framework antibody libraries. Nat. Biotechnol, 18, 852-856 (2000)

14. Ingvarsson, J.; Larsson ¹ , A.; Sjόholm, L.; Truedsson, L.; Jansson, B.; Borrebaeck, C. A.K. and Wingren, C. Design of recombinant antibody

microarrays for serum protein profiling: Targeting of complement proteins. J. Proteome Res.in press

15. Wingren, C ₁ Ingvarsson, J., Dexlin, L, Szul, D. and Borrebaeck, CAK. Design of recombinant antibody microarrays for complex proteome analysis: choice of sample labelling-tag and solid support. Proteomics in press

16. Eisen, M. B., et al., Cluster analysis and display of genome-wide expression patterns. Proc Natl Acad Sci U S A. 95, 14863-14868 (1998). ^•

17. N. Cristianini and J. Shawe-Taylor, An introduction to support vector machines (and other kernel-based learning methods), Cambridge University Press (2000).

18. Chih-Chung Chang and Chih-Jen Lin, LIBSVM: a library for support vector machines, available at http://www.csie.ntu.edu.tw/~cjlin/libsvm

19. R. lhaka and R. Gentleman, R: A language for data analysis and graphics, J. Comp. Graph Stat. 5, 299-314, (1996) 20. Sanchez-Carbayo, M., Socci, N. D., Lozano, JJ. , Haab, B.B. &Cordon-

Cardo, C. Am. J, Pathol. 168, 93-103 (2006). 21. Schafer, M.W., Mangold, L., Partin, A. W. and Haab, B.B. (2007) Antibody array profiling reveals serum TSP-I as a marker to distinguish benign from malignant prostatic disease, The Prostate 67, 255-267. 22. Rustgi, A.K. Gastroenterology 129, 1344-1347 (2005).

23. More, G., et al. Proc. Natl. Acad. Sci (USA) 102, 7677-7682 (2005).

24. Wingren, C. and Borrebaeck, CA. K. (2007) nya reviewn

25. Steinhauer, C ₁ et al., Biocompatibility of surfaces for antibody microarrays: design of macroporous silicon substrates. Anal Biochem 341 , 204-13 (2005) 26. Orchekowski, R., et al., Antibody microarray profiling reveals individual and combined serum proteins associated with pancreatic cancer. Cancer Res. 65, 11193-202 (2005)

27. Bellone, G., et al., Tumor-associated transforming growth factor-beta and interleukin-10 contribute to a systemic Th2 immune phenotype in pancreatic carcinoma patients. Am J Pathol. 155, 537-47 (1999)