Biomarkers for Typing Allograft Recipients

Abstract

The invention relates to biomarkers for typing or classifying allograft recipients as belonging to a transplant rejection group associated with antibody-mediated rejection (ABMR). The invention also provides for the treatment of typed allograft recipients suffering from antibody-mediated rejection by administration of an appropriate therapeutic agent.

Claims

1. A method for typing an allograft recipient for the presence or absence of an antibody mediated rejection (ABMR), comprising the steps of measuring in a sample comprising proteins from an allograft recipient a protein level for at least two genes selected from the group consisting of TF, SERPINA1, APOA4, AFM, AZGP1, ORM1, ORM2, C3, A1BG, SERPINC1, LRG1, IGHA1, IGHG4, TFAP2C, HPX, A2M, CARD6, SERPINA7, CCDC73, CYSTM1 and APOA1; comparing said measured protein level to a reference protein level for said at least two genes; and typing said allograft recipient for the presence or absence of an ABMR on the basis of the comparison of the measured protein level and the reference protein level.

2. The method according to claim 1, wherein the method is for typing a sample of said allograft recipient for the presence or absence of an ABMR.

3. The method according to claim 1, wherein the allograft recipient is a renal allograft recipient.

4. The method according to claim 1, wherein the sample is a body fluid sample.

5. The method according to claim 1, wherein a protein level is measured for at least 6 genes selected from the group consisting of TF, SERPINA1, APOA4, AFM, AZGP1, ORM1, ORM2, C3, A1BG, SERPINC1, LRG1, IGHA1, IGHG4, TFAP2C, HPX, A2M, CARD6, SERPINA7, CCDC73, CYSTM1 and APOA1.

6. The method according to claim 1, wherein a protein level is measured for at least two genes selected from the group consisting of A1BG, AFM, APOA1, APOA4, IGHA1, IGHG4, LRG1, SERPINA1, SERPINC1 and TF.

7. The method according to claim 6, wherein a protein level is measured for at least 6 genes selected from the group consisting of A1BG, AFM, APOA1, APOA4, IGHA1, IGHG4, LRG1, SERPINA1, SERPINC1 and TF.

8. The method according to claim 1, wherein said allograft recipient, or sample thereof, is typed as having an ABMR when said protein levels are increased as compared to a reference protein level for said at least two genes in a reference sample of an allograft recipient not having an ABMR.

9. The method according to claim 1, further comprising the steps of: digesting proteins in said sample with trypsin so as to provide a mixture of peptides; subjecting said mixture of peptides to a step of liquid chromatography so as to provide an eluate comprising peptides; and performing a step of mass spectrometry on said eluate to measure a peptide level for at least two peptides, said peptide level for said at least two peptides representing the protein level for said at least two genes.

10. The method according to claim 9, wherein said at least two peptides are selected from SEQ ID NOs: 1-22.

11. The method according to claim 1, wherein measurement of said protein level is performed by an enzyme-linked immunosorbent assay (ELISA).

12. A method for assigning an allograft recipient to an ABMR group or a non-ABMR group comprising the steps of: measuring in a sample comprising proteins from an allograft recipient suffering, or at risk of suffering, from transplant rejection a protein level for at least two genes selected from the group consisting of TF, SERPINA1, APOA4, AFM, AZGP1, ORM1, ORM2, C3, A1BG, SERPINC1, LRG1, IGHA1, IGHG4, TFAP2C, HPX, A2M, CARD6, SERPINA7, CCDC73, CYSTM1 and APOA1; comparing said measured protein level to a reference protein level for said at least two genes; and assigning said allograft recipient to said ABMR group or to said non-ABMR group on the basis of the comparison of the measured protein level and the reference protein level.

13. The method of claim 1 wherein the sample is a urine sample and at least two proteins from the at least two genes serve as a marker for the presence or absence of an ABMR in an renal allograft recipient.

14. A standard-of-care therapeutic agent for use in the treatment of an allograft recipient suffering from transplant rejection associated with an ABMR, wherein said recipient, or a sample thereof, is typed as having an ABMR according to the method of claim 1.

15. The standard-of-care therapeutic agent for use according to claim 14, wherein said agent is selected from the group consisting of corticosteroids, rituximab, intravenous immunoglobulin (IVIG) products, bortezomib, eculizumab and combinations thereof; and wherein said agent is for administration according to a therapeutically effective dosing regimen.

16. A method for treating an allograft recipient suffering from transplant rejection associated with an ABMR, comprising the steps of: administering a therapeutically effective amount of a standard-of-care therapeutic agent to an allograft recipient suffering from transplant rejection associated with an ABMR, wherein said recipient, or a sample thereof, is typed as having an ABMR according to the method of claim 1.

17. A standard-of-care therapeutic agent for use in the treatment of an allograft recipient suffering from transplant rejection associated with an ABMR, wherein said recipient, or a sample thereof, is assigned to the ABMR group according to a method of claim 12.

18. The standard-of-care therapeutic agent for use according to claim 17, wherein said agent is selected from the group consisting of corticosteroids, rituximab, intravenous immunoglobulin (IVIG) products, bortezomib, eculizumab and combinations thereof; and wherein said agent is for administration according to a therapeutically effective dosing regimen.

19. A method for treating an allograft recipient suffering from transplant rejection associated with an ABMR, comprising the steps of: administering a therapeutically effective amount of a standard-of-care therapeutic agent to an allograft recipient suffering from transplant rejection associated with an ABMR, wherein said recipient, or a sample thereof, is assigned to the ABMR group according to a method of claim 12.

20. The method of claim 19 wherein the standard-of-care therapeutic agent is selected from the group consisting of corticosteroids, rituximab, intravenous immunoglobulin (IVIG) products, bortezomib, eculizumab and combinations thereof; and wherein said agent is for administration according to a therapeutically effective dosing regimen.

Description

FIGURE LEGENDS

[0096] FIG. 1. List of biomarkers segregating ABMR from non-ABMR in renal allograft recipients.

[0097] Top 21 of selected upregulated proteins that segregate ABMR from non-ABMR phenotypes in step 1 and 2 (training cohort) in a case-control setup. The six proteins selected in green (grey) are the proteins for which two unique peptides (vide Table 1) were used to train and validate a statistical SVM model (Example 1). All 21 proteins are upregulated in AMBR cases (minimal fold change in log 2 is 0.8 in step 1 or step 2).

[0098] FIG. 2. ROC curve of validation dataset.

[0099] Receiver operating characteristic (ROC) curve obtained by employing the six proteins (Example 1)detected in the form of the twelve peptides listed in Table 1as biomarkers in a validation cohort (N=240) comprising patients having received a kidney allograft.

[0100] FIG. 3. Study outline

[0101] Study outline, showing the training (step 1 and 2) and validation cohorts (step 3).

[0102] FIGS. 4 and 5. ROC curves on training data set (FIG. 4) and on validation data set (FIG. 5).

[0103] Diagnostic accuracy of the protein biomarkers in the training and the validation set (Example 2). Receiver Operating Characteristic (ROC) curves are shown for the full model with 10 proteins (Table 4) for the training set (N=249; FIG. 4)) and for the validation set (N=391; FIG. 5). In the training dataset, the full model with 10 proteins has 98% AUC. In the validation dataset, this model has 88% AUC.

[0104] FIG. 6. Classification with two random proteins of Table 5.

[0105] Number of random proteins of Table 5 needed to achieve ABMR classification.

EXAMPLES

Example 1. Training and Validation of Biomarkers Segregating Antibody-Mediated Kidney Allograft Rejection from Other Kidney Allograft Rejection Phenotypes

[0106] Materials and Methods

[0107] Study Population

[0108] We performed a multicentre retrospective study. Patient who received a kidney allograft in four European clinical centres (University Hospital Leuven, Paris Necker, CHU Limoges and Hannover Medical School), were included in the study with written informed consent. Protocol or indication biopsies were performed and urine samples were collected. In this proteomics study, only urine samples were used for analysis. Biopsies were read by local and central pathologists, who classified all samples in four different phenotypes: Normal (NL), antibody mediated rejection (ABMR), T-cell mediated rejection (TCMR) and interstitial fibrosis and tubular atrophy (IFTA). Combinations of phenotypes were also possible.

[0109] General Study Design

[0110] The present study can be generally divided in three steps. In step 1, 130 urine samples of kidney allograft recipients were analysed, and in step 2 urine samples of 133 kidney allograft recipients were analysed. Steps 1 and 2 relate to the identification, and training, of the differentially expressed proteins for use as diagnostic ABMR biomarkers. Step 3 relates to independent validation of the biomarkers, in which step the diagnostic performance of the biomarkers was tested on 240 samples of kidney allograft recipients.

[0111] Urine Collection

[0112] Urine samples were collected in the four different clinical centres. Fresh urine samples, preferably the second voiding, were collected in the morning before the biopsy was taken. Urine creatinine, haemoglobin, leucocytes, glucose and protein content was measured locally using dipstick tests. Urine samples were centrifuged at 2,000 g at 4 C. for 20 minutes to remove cell debris and casts within 2 h after collection. The supernatant was stored below 20 C. until shipment to the analytical centres. Upon arrival, the samples were stored at 80 C.

[0113] Per step, samples were randomized in batches of 24 samples, taking into account that all batches contained samples from every clinical centre and all different phenotypes.

[0114] Sample Preparation

[0115] After a first concentration determination using the Pierce BCA Protein Assay Kit (Thermo Scientific), 2 mg of protein was processed on an Amicon Ultra-0.5 Centrifugal Filter Unit with a 10 kDa molecular weight cut-off membrane (Merck Millipore). The protein concentration of the concentrated samples was again determined using the same BCA Protein assay. Subsequently, 100 g of protein was loaded on the Pierce Albumin Depletion Kit (Thermo Scientific) spin columns to deplete the samples from human albumin. After albumin depletion, the protein concentration was determined a last time. 20 g of protein was denatured in 0.1% Rapigest (RapiGest SF, Waters). After denaturation, proteins were reduced by adding 2 l of 200 mM TCEP (Tris(2-carboxyethyl)phosphine; Thermo Scientific) and incubating the sample for 1 h at 55 C. Afterwards, samples were alkylated by adding 2 l of 375 mM IAA (Iodoacetamide; Thermo Scientific) for 30 minutes at room temperature protected from light. To precipitate the proteins, 1 ml of pre-chilled acetone was added and incubated overnight at 20 C. After a centrifugation step (10,000 g, 15 min., 4 C.), the protein pellet was resuspended in 20 l 200 mM TEAB (Triethylammonium bicarbonate; Sigma-Aldrich). 1 g of trypsin (Trypsin Gold, Mass Spectrometry Grade; Promega) was added to digest the proteins while incubating overnight at 37 C. The digestion was stopped and Rapigest was hydrolyzed by adding HCl to a final concentration of 200 mM (30 min. at room temperature). After a centrifugation step (10,000 g, 15 min., 4 C.), the pellet was removed and the samples were diluted in 2% acetonitrile, 0.1% formic acid to a final concentration of 0.2 g/l. All samples were spiked with 4 fmol/l GFP ([Glu1]-Fibrinopeptide B human; Sigma-Aldrich).

[0116] Nano Reversed Phase Liquid Chromatography and Mass Spectrometry

[0117] In total, 1 g of the peptide mixture, spiked with 20 fmol GFP, was loaded on the LC column. The tryptic peptide mixture was analysed on a Nano Acquity Ultra Performance LC system (Waters) using a nanoACQUITY UPLC Symmetry C18 Trap Column (180 m20 mm; Waters) coupled to a ACQUITY UPLC Peptide BEH C18 nanoACQUITY column (100 m100 mm; Waters). A linear gradient of mobile phase B (98% acetonitril, 0.1% formic acid, pH=2) from 5 to 45% in 68 min. was followed by a steep increase to 90% mobile phase B in 3 minutes. The flow was set at 400 nl/min. The nano-LC was coupled online to the LTQ Velos Orbitrap mass spectrometer (Thermo Scientific) via the nanospray ion source (Thermo Scientific).

[0118] The LTQ Velos orbitrap was set up in MS/MS shotgun mode, where a full MS1 precursor scan (300-2000 m/z, resolution 60,000) was followed by a maximum of 10 collision induced dissociation (CID) MS2 spectra of the 10 most intense precursor peaks. CID spectra were obtained in the linear ion trap of the mass spectrometer. The normalized collision energy used in CID was set at 35%. We applied a dynamic exclusion of 30 s for data dependent acquisition.

[0119] Quality Control Analysis

[0120] The MS/MS results (raw data) together with the Proteome Discoverer results were inspected in a quality control (QC) analysis. QC analysis is done systematically as it guarantees the quality of the sample and the MS instruments at each moment for each sample. If for several reasons, samples do not meet the requested QC parameters, these samples are excluded for further data analysis steps.

[0121] Data Enrichment Process

[0122] To get quantitative data on all peptides for each sample, we developed in house software to look up peak intensities in the raw MS1 data. In short, this software tool looks up m/z values in raw MS1 data with a delta ppm of 5 in a retention time window of 10 minutes. That way, a data matrix is obtained containing quantitative information from almost all identified peptides. The algorithm also cleans the resulting data by using a decoy search and also peak shape is checked.

[0123] Model Building

[0124] Data of step 1 and step 2 were used as training data. Step 1 data was used for selecting the significantly differentially expressed proteins. Step 2 data was used as a first verification dataset. The hypothesis we tested was ABMR vs. no-ABMR. ANOVA was used to select proteins that were significantly upregulated or downregulated in ABMR cases. ANOVA was applied on generated data of each protein of step 1 and step 2 samples independently. In the final list, the top 21 proteins were selected that are upregulated in ABMR1 compared to ABMR0 (i.e. ABMR vs. no-ABMR) with at least a fold change of 0.8 (log 2 fold change) in one of the two steps (on the protein level) (FIG. 1).

[0125] Once the proteins were selected, we selected 2 unique peptides per protein (no miscleavages) (Table 1). The selection was based on results in peak scoring and their suitability for targeted analysis. If a protein in the final list does not have two peptides fulfilling those criteria, the protein is left out of the model. This way, the model can be validated in an extra validation step following this study, using targeted proteomics. Only unique peptides were used in this ANOVA analysis. Results of step 1 and step 2 analyses are listed by log 2 fold change and p-value in Table 1 below.

[0126] Model Validation

[0127] Finally, the support vector machine (SVM) that was modelled, is applied on the normalized, enriched data obtained from the step 3 samples, which serve as an independent validation dataset. An ROC curve is generated and the results are inspected and plot per individual patient.

[0128] Results

[0129] Peptide Identification

[0130] All data was searched against the human Uniprot database using Proteome Discoverer software (version 2.1; Thermo Scientific). Both search engines Mascot and Sequest were used. The following search parameters were used: precursor mass tolerance 10 ppm, fragment mass tolerance 0.5 Da. Trypsin was chosen as the cleavage enzyme and 2 missed cleavages were allowed. Carbamidomethylation was set as a fixed modification on cysteine and methionine oxidation and serine, tyrosine and threonine phosphorylation were set as variable modifications. The resulting peptide identification results were filtered using the following settings: only the high confident with a False Discovery Rate (FDR)<5% based on the target-decoy approach and the first ranked peptides were included, to yield the proteins indicated in FIG. 1.

[0131] In an alternative experimental set-up, it was established that the expression products of the genes listed in FIG. 1 do not provide for differentiation between ABMR and non-ABMR phenotypes when said expression products are mRNA (data not shown).

[0132] Statistical AnalysisModel Building

[0133] ANOVA was applied on the data obtained in step 1 and step 2, and results were listed by p-value. The top 21 selected proteins are shown in FIG. 1. We selected two unique peptides with high confident identification per protein that can also be used in targeted analysis. If such unique peptides were not available for a listed protein, the protein was not used in the model. This way, our final model contained 12 peptides from 6 proteins. The selected peptides are shown in Table 1 below. Biopsy results are considered as outcome variable in our analysis. We trained a support vector machine on the data of step 1 and step 2 using the 12 selected peptides as parameters. After fixing the cutoff point, we obtained a sensitivity of 84.7% and a specificity of 78.3% (vide Table 2).

TABLE-US-00001 TABLE1 SelectedlistofuniquepeptidesusedfortrainingtheSVMmodel. Protein Accession Step1_ Step2_ Step1_ Step2_ SEQ GeneID No. Sequence foldchange foldchange pvalue pvalue ID SERPINC1 P01008 EQLQDMGLVDLFSPEK 1.23 1.58 0.00257 0.00007 1 SERPINC1 P01008 VAEGTQVLELPFK 1.40 1.64 0.17546 0.00013 2 SERPINA1 P01009 LSITGTYDLK 1.30 1.63 0.00027 0.00039 3 SERPINA1 P01009 SVLGQLGITK 1.26 1.36 0.00017 0.00012 4 IGHA1 P01876 DASGVTFTWTPSSGK 1.14 1.71 0.00081 0.00006 5 IGHA1 P0187G TFTcTAAYPESK 0.86 2.01 0.00145 0.00002 6 TF P02787 cSTSSLLEAcTFR 1.94 3.00 0.00001 0.00000 7 TF P02787 DSGFQMNQLR 1.54 2.11 0.00018 0.00002 8 A1BG P04217 ATWSGAVLAGR 1.97 1.30 0.00002 0.02962 9 A1BG P04217 cEGPIPDVTFELLR 1.91 1.96 0.00003 0.01734 10 AFM P43652 AESPEVcFNEESPK 1.54 1.66 0.00153 0.00001 11 AFM P43G52 FTDSENVcQER 1.34 1.71 0.00076 0.00001 12

TABLE-US-00002 TABLE 2 Contingency table for the training dataset. biopsy diagnosis no ABMR ABMR model no ABMR 160 13 173 classification ABMR 29 47 76 on 189 60 249 sensitivity = 84.7% specificity = 78.3%

[0134] Table 2 shows an overview of the samples classification using the model compared to the biopsy results. We see that for diagnosing ABMR, the model is more conservative than the pathologist's decision, because in almost 30% of the cases for which the biopsy results in the diagnosis of ABMR, the model disagrees, while for ABMR0 (i.e. no ABMR), the model only disagrees in 15% of the cases. However, these 15% are very important cases, because here the model could possibly pick up signs of ABMR before the histology reveals any signs of rejection. In the case of ABMR diagnosis, the model is envisaged to help pathologists make a more accurate diagnosis.

[0135] Statistical AnalysisModel Validation

[0136] The SVM model is fixed using data from step 1 and 2 as a training dataset. Thus the step 3 samples are used as a completely independent validation dataset. Validation took place in a large cohort (240) of independent patients that previously received a kidney allograft.

[0137] 77% of the samples is correctly classified. After fixing the cutoff point, a sensitivity of 79.1% and a specificity of 70.3% was obtained (vide Table 3).

TABLE-US-00003 TABLE 3 Contingency table for the validation dataset Validation data cutoff = 0.75 biopsy diagnosis no ABMR ABMR model no 117 11 128 classification ABMR ABMR 31 26 57 148 37 185 sensitivity = 79.1% specificity = 70.3%

Example 2

[0138] This Example builds on Example 1.

[0139] Additional renal allograft recipients were included in the study. Sample preparation and protein expression level measurement are as indicated in Example 1.

[0140] The training data set again represented 249 kidney allograft recipients, and the validation data set now represented 391 kidney allograft recipients. All biopsies included in this study were reviewed and graded in a blinded fashion by a central pathologist independent from the original center. Study outline is provided in FIG. 3.

[0141] Results

[0142] In the training set, 60/249 cases showed ABMR (24.1%), and 43/391 (11.0%) in the validation set.

[0143] Results of Example 1 were also achieved when expanding patient population.

[0144] Subsequently, in the same manner as in Example 1, a diagnostic model was build now on the basis of ten proteins (A1BG, AFM, APOA1, APOA4, IGHA1, IGHG4, LRG1, SERPINA1, SERPINC1 and TF) by selecting 2 unique peptides per protein. This set of 20 peptides is provided in Table 4 below.

TABLE-US-00004 TABLE4 Selectedlistofpeptidesthatwereusedfor traininginthetheSVMmodeltrainingset. GeneID peptide SEQID A1BG ATWSGAVLAGR 9 A1BG cEGPIPDVTFELLR 10 AFM AESPEVcFNEESPK 11 AFM FTDSENVcQER 12 APOA1 DLATVYVDVLK 13 APOA1 DYVSQFEGSALGK 14 APOA4 ISASAEELR 15 APOA4 SLAELGGHLDQQVEEFR 16 IGHA1 DASGVTFTWTPSSGK 5 IGHA1 TFTcTAAYPESK 6 IGHG4 TTPPVLDSDGSFFLYSR 17 IGHG4 YGPPcPScPAPEFLGGPSVFLFPPKPK 18 LRG1 ALGHLDLSGNR 19 LRG1 DLLLPQPDLR 20 SERPINA1 LSITGTYDLK 3 SERPINA1 SVLGQLGITK 4 SERPINC1 ADGEScSASMMYQEGK 21 SERPINC1 IEDGFSLK 22 TF cSTSSLLEAcTFR 7 TF DSGFQMNQLR 8

[0145] The diagnostic performance of each of said ten genes is shown in Table 5.

TABLE-US-00005 TABLE 5 List of 10 proteins that segregate ABMR from non ABMR phenotypes in the training data set. maximum total median FDR number of total number log2 fold corrected peptide Uniprot number of of unique change p-value spectrum maximum protein peptides peptides training training matches sequence GeneID Accession identified identified dataset dataset PSMs coverage A1BG P04217 12 4 1.13 0.01093 227 54.14 AFM P43652 14 14 1.00 0.00005 55 37.06 APOA1 P02647 16 3 0.61 0.04509 54 60.30 APOA4 P06727 21 21 0.60 0.00005 148 70.20 IGHA1 P01876 12 4 0.87 0.00030 79 68.84 IGHG4 P01861 2 2 0.78 0.00757 50 39.45 LRG1 P02750 9 9 0.68 0.00000 86 44.96 SERPINA1 P01009 24 18 1.29 0.00000 1057 71.29 SERPINC1 P01008 9 7 0.86 0.00022 35 44.61 TF P02787 53 31 1.37 0.00000 1420 82.38

[0146] This set of ten proteins, each protein represented by two peptides, is considered a good representation of ten random proteins selected from FIG. 1. The ROC curve of the ten protein model is shown in FIG. 4 (training data set) and FIG. 5 (validation data set). After fixing the cutoff point to 0.30, this model reached a sensitivity of 95% and a specificity of 96%. In addition, it was further shown that the diagnostic performance of this ten protein model is also achieved with six random proteins of said set of 10 proteins (Tables 6 and 7).

TABLE-US-00006 TABLE 6 Different sets of proteins included in the model. model model Proteins included in the model name description A1BG AFM APOA1 APOA4 IGHA1 IGHG4 LRG1 SERPINA1 SERPINC1 TF model10 all 10 x x x x x x x x x x proteins model6A First set of x x x x x x 6 proteins model6B Second set of x x x x x x 6 proteins model6C Third set of x x x x x x 6 proteins

TABLE-US-00007 TABLE 7 Results for all 4 models fitted for the training dataset and the validation dataset. Model name TP TN FP FN Sensitivity Specificity PPV NPV Training set model10 57 182 7 3 0.95 0.96 0.89 0.98 model6A 57 179 10 3 0.95 0.95 0.85 0.98 model6B 57 178 11 3 0.95 0.94 0.84 0.98 model6C 57 178 11 3 0.95 0.94 0.84 0.98 Validation set model10 41 263 85 2 0.95 0.76 0.33 0.99 model6A 36 243 105 7 0.84 0.70 0.26 0.97 model6B 36 241 107 7 0.84 0.69 0.25 0.97 model6C 40 243 105 3 0.93 0.70 0.28 0.99 TP: true positives; TN: true negatives; FP: false positives; FN: false negatives; PPV: positive predictive value; NPV: negative predictive value.

[0147] Finally, the diagnostic performance of at least two random proteins of said set of 10 proteins is shown in FIG. 6. Unexpectedly, it was shown that at least two random proteins of said set of 10 already provide for a correlation with ABMR of >87%. It is shown that a plateau in diagnostic performance is achieved already with six proteins.