Method to screen, diagnose or monitor treatment using increased numbers of deviating analytes

Abstract

The present invention relates to the field of methods to screen, diagnose or monitor treatment of disease. More in particular, the present invention discloses methods wherein an increased number of analytes demonstrating deviating amounts in a biofluid or tissue sample is used to determine whether an individual is diseased.

Description

[0001] Method to screen, diagnose or monitor treatment using increased numbers of deviating analytes

[0002] Technical field of invention

[0003] The present invention relates to the field of methods to screen, diagnose or monitor treatment of disease. More in particular, the present invention discloses methods wherein an increased number of analytes demonstrating deviating amounts in a biofluid or tissue sample is used to determine whether an individual is diseased.

[0004] Background art

[0005] Diseases such as cancer are often complex and have considerable heterogeneity. Finding robust and easily accessible biomarkers is crucial to improve patient well-being and outcome. Such biomarkers can either assist in early detection of disease, which means that therapy can be started at an earlier stage and may be less invasive, or they can help diagnosis and monitoring of treatment response to guide tailored therapy decision making. As an alternative for biomarkers that require invasive tissue biopsies, liquid biopsy-based analytes such as circulating nucleic acids have emerged as a promising tool for disease diagnosis and monitoring. To date, most studies on circulating nucleic acids in cancer patients have focused on circulating tumor DNA (ctDNA)1. However, the amount of tumor DNA in circulation is highly dependent on the tumor type and burden2, which means that ctDNA can be less informative for certain cancer types or early-stage cancers.

[0006] Cell-free RNA (cfRNA) may complement ctDNA as it can reflect dynamic changes in gene expression during cancer development and progression in both healthy and diseased cells. A recent study showed that a lot of different cell types contribute the healthy cell-free plasma transcriptome and that cfRNA allows monitoring of cell-type-specific changes in diseases, such as non-alcoholic fatty liver and Alzheimer's disease3. Another study indicated the potential of using cfRNA as prognostic biomarker for pre-eclampsia and organ damage in pregnancy4. Some studies have also shown potential of using cfRNA biomarkers in certain cancer types. Larson et al.5specifically looked at genes that are "dark" (no or few detected transcripts) in plasma of healthy controls but abundant in plasma lung and / or breast cancer patients. They further defined dark channel biomarker genes as dark genes that are also differentially abundant at group level. These biomarkers thus depend on the absence of transcripts in controls and exclude genes that are less abundant in cancer than control. Chen et al.6reported potential of combining human and microbe-derived cfRNA in plasma for distinguishing healthy donor and cancer plasma samples (coIorectum, stomach, liver, lung, and esophageal cancer). The classifiers were built based on abundance profiles of RNAs that show differences at group level (rank sum test). Roskams-Hieter et al.7showed classification potential for liver cancer and multiple myeloma compared to non-cancer controls. They applied learning vector quantization (LVQ) to find the most important features that distinguished the two groups and showed that the corresponding classifiers yielded a higher accuracy than those based on differentially abundant genes. However, the authors also noted that liver and bone marrow are major contributors to the plasma transcriptome which may result in more cancer signal. US 2010 / 130377 to Vasmatzis et al. discloses a method wherein a prognostic biomarker is used to assess the aggressiveness of prostate cancer. The greater the number of said biomarker in a sample, the more aggressive is the prostate cancer and the more susceptible is the mammal to a poor outcome.

[0007] There is however still a need to design methods that take into account variability of analytes in a biofluid or tissue sample and wherein a biomarker is used to determine whether an individual is diseased.

[0008] Description of invention

[0009] The present invention relates in first instance to an in vitro method to screen, diagnose or monitor treatment of a disease comprising:

[0010] -providing a biofluid or tissue sample taken from a possibly diseased individual and from a group of healthy control individuals,

[0011] -determining in said biofluid or tissue sample taken from a possibly diseased individual the analytes having an amount which deviates more than 3 standard deviations from the mean amount of the corresponding analytes in said biofluid or tissue sample from said group of healthy control individuals, and

[0012] -determining the number of said deviating analytes, wherein an increased number of said deviating analytes is used to determine whether said individual is diseased. The number of deviating analytes above which an individual is called 'diseased' is determined based on ROC analysis using a training and test approach with optional cross validation.

[0013] The latter method clearly differs from already disclosed methods in that the total number of analytes which are present in an amount which is higher than 3 (positive) standard deviations from the mean control amount plus the analytes which are present in an amount which is lower than 3 (negative) standard deviations from the mean control amount is used to diagnose a disease.

[0014] The term 'a disease' refers to a particular abnormal condition that negatively affects the structure or function of all or part of an organism, and that is not immediately due to any external injury. Non-limiting examples of 'a disease' are cancer such as liver cancer, lymphoma or bladder cancer, neurological disorders such as Alzheimer's disease, metabolic diseases, and infectious diseases.

[0015] The term 'analyte' as used herein refers to a substance or chemical constituent which is a measurable indicator of some biological state or condition such as a disease. Said analyte can be any chemical compound such as a nucleic acid (DNA, RNA...), a peptide, a lipid, a metabolite...

[0016] The term 'a biofluid' can be any liquid such as blood, plasma, serum, urine...taken from the human body.

[0017] The term 'a tissue' can be any group of structurally and functionally similar cells and their intercellular material for example a sample of liver, kidney, lung, muscle, thyroid... taken from the human body.

[0018] The term 'a possibly diseased individual' refers to a human or an animal which is suspected to have a particular disease. The latter term stands in contrast with the term 'healthy control individuals' which refers to a group of humans or animals which are known to be healthy and / or free of any disease. The terms 'the analytes having an amount which deviates more than 3 standard deviations from the mean amount of the corresponding analytes' relate to -as explained via a none-limiting example- an analyte having an amount of less than 4pg / L or more than 16 pg / L in case the mean amount of said analyte in said biofluid sample from said group of healthy control individuals is 10 pg / L with a standard deviation of 2pg / L. The number of analytes having an amount of less than 4pg / L or more than 16 pg / L are higher in a diseased individual when compared to said number of analytes present in samples from healthy control individuals and can thus be used to determine whether an individual is diseased.

[0019] The present invention thus further specifically relates to an in vitro method as described above wherein said biofluid sample is a blood plasma sample, a urine sample, or a uterotubal lavage sample.

[0020] The present invention further specifically relates to an in vitro method as described above wherein said disease is cancer.

[0021] The present invention further more specifically relates to an in vitro method as described above wherein said cancer is ovarian cancer, prostate cancer, uterine cancer, lymphoma, colon cancer, or bladder cancer.

[0022] The present invention further more specifically relates to an in vitro method as described above wherein said cancer is early stage, locally advanced or metastatic cancer.

[0023] The present invention further specifically relates to an in vitro method as described above wherein said analyte is RNA or a metabolite such as an amino acid, nucleic acid, organic acid, lipid, or carbohydrate. The present invention more specifically relates to an in vitro method as described above wherein said RNA is mRNA.

[0024] The present invention further relates to an in vitro method as described above wherein said step of determining the number of said deviating analytes can be preceded by a filtering step to identify which analytes are specifically associated to the disease state: for each deviating analyte, the association of the deviation to a particular disease state is tested by means of a statistical contingency table test (e.g. Fisher's exact test, chi-square test of independence, or alike) and those analytes with significant associations (p<0.05) are retained in a biomarker subset. Counting the number of deviating analytes per sample belonging to this subset increases the classification potential.

[0025] Examples

[0026] Methods

[0027] Sample cohorts

[0028] Plasma cohort 1

[0029] Plasma samples of 56 human donors were obtained from Proteogenex (CA, USA), following approval by the ethics committee of Ghent University Hospital, Ghent Belgium (B670201734362). The consists plasma samples from 30 cancer-free control donors, 12 prostate cancer (PRAD), 12 ovarian cancer (OV), and 12 endometrial cancer (UCEC) patients. All cancer patients in both cohorts had locally advanced (stage 3) to metastatic cancer (stage 4). Blood was drawn before treatment.

[0030] Plasma cohort 2

[0031] This validation cohort consists of plasma samples from 65 human donors: 30 diffuse large B-cell lymphoma (DLBCL) patients, 13 primary mediastinal large B-cell lymphoma (PMBCL) patients, and 22 cancer-free control donors. Plasma samples were collected at time of diagnosis, following approval by the ethics committee of Ghent University Hospital, Ghent, Belgium (2016 / 0307).

[0032] Urine cohort

[0033] This validation cohort consists of urine samples from 12 muscle-invasive bladder cancer patients and 12 control donors (healthy donors). The cohort is described in more detail by Hulstaert et al8. Uterotubal lavage cohort

[0034] This cohort consists of uterotubal lavage fluid samples (obtained by flushing saline into the uterine cavity and fallopian tubes) from 25 ovarian cancer patients and from 48 donors with non-malignant conditions (benign ovarian mass, uterine prolapse, menorrhagia etc.) as described in more detail by Hulstaert et al10.

[0035] Metabolites cohort This cohort includes cancer and normal tissue samples from patients with colon adenocarcinoma (n=275) as described by Benedetti et al9. The preprocessed metabolite data (normalized and with missing values imputed to minimum) from this paper was used for further analysis.

[0036] Blood collection and plasma preparation

[0037] Blood of plasma cohort 1 donors was collected in EDTA vacutainer tubes (Becton Dickinson, 367525). Blood was refrigerated (4°C) until plasma preparation which started by gently inverting the EDTA tube 10 times and centrifugating for 10 min at 1500 g (without brake at 4°C). Supernatans was then transferred to 15 mL centrifuge tubes and centrifuged again for 10 min at 1500 g (without brake at 4°C). The resulting platelet depleted plasma was transferred to 2 mL cryovials, frozen and stored at - 80°C within 4 hours of blood collection. Plasma samples were shipped on dry ice and only thawed on ice immediately before RNA isolation.

[0038] For plasma cohort 2, 2.5 mL of blood was collected from each donor in PAXgene Blood DNA Tubes (BD Biosciences, 761165). Platelet-poor plasma was prepared by a one-step 15 min centrifugation at 1900 g without brake (room temperature) and stored at -80°C within 4 hours of blood collection.

[0039] RNA isolation and gDNA removal

[0040] RNA was each time isolated from 200 pL of plasma with the miRNeasy Serum / Plasma kit (Qiagen, 217184), according to the manufacturer's instructions. For plasma cohort 1, 2 pL Sequin spike-in controls (1 / 1,300,000 of stock solutions mix A, Garvan Institute of Medical Research, Australia) and 2 pL RNA extraction Control spike-ins (Integrated DNA Technologies) were added to the lysate during RNA isolation as previously described in Hulstaert et al (ref). For plasma cohort 2, 2 pL of Sequin spike-in controls were added to the sample lysates (1 / 5,000 of stock solution mix A (Garvan Institute of Medical Research) and 2 pL of External RNA Control Consortium (ERCC) spike-in controls (ThermoFisher Scientific, 4456740) were added to the RNA eluate.

[0041] Genomic DNA was removed by adding 1 pL HL-dsDNase (ArcticZymes, 70800-202) and 1.6 pL reaction buffer (ArcticZymes, 66001) to 12 pL RNA eluate, 10 min incubation at 37 °C, followed by 5 min incubation at 55 °C. RNA was stored at -80°C and thawed on ice immediately before library preparation.

[0042] Messenger RNA capture library preparation, sequencing, and quantification

[0043] MRNA capture library preparation in plasma cohort 1 started from 8.5 pL DNase treated RNA eluate and cDNA synthesis was performed using TruSeq RNA Library Prep for Enrichment (Illumina, 20020189) as described in Hulstaert et al11. Briefly, RNA was fragmented, and first strand cDNA was generated using random priming. RNA templates were subsequently removed and replaced by a newly synthesized second strand of cDNA. AMPure XP beads (Beckman Coulter Life Sciences, A63881) were used for purifying the blunt-ended double stranded cDNA. 30 pL cDNA of each sample was then used as input for Illumina DNA Prep with Enrichment (previously Nextera Flex for Enrichment; Illumina, 20025524). Tagmentation and amplification steps were done according to manufacturer's instructions. Quality of resulting pre-enriched libraries was assessed using a high sensitivity Small DNA Fragment Analysis Kit (Agilent Technologies, DNF-477-0500). The libraries were pooled per 8 or 6 samples, for pan-cancer and validation cohort respectively, based on qPCR quantification with YWHAZ, ACTB, B2M, and UBC as reference genes. Each multiplex pool was concentrated to a volume of 15 pL with AMPure XP beads (Beckman Coulter Life Sciences, A63881). Finally, enrichment was performed using probes from the Illumina Exome Panel (Illumina, 20020183), probes complementary to the spike-in controls, and blocking probes against globin (anti-CEX) as described in Hulstaert et al11. At the end of the enrichment workflow, equimolar library pools were prepared based on qPCR quantification with KAPA Library Quantification Kit (Roche Diagnostics, KK4854). Paired-end sequencing was performed (2x100 nucleotides) on a NovaSeq 6000 using a NovaSeq SI kit (Illumina, 20028318) with Xp workflow loading of 1.25 nM (2% PhiX).

[0044] Adapter trimming and removal of reads shorter than 20 nucleotides was done with cutadapt (vl.18). Read quality was assessed with FastQC (vO.11.9) and low-quality reads were filtered out: only reads where at least 80% of bases in both mates have a quality score >20 (99% accuracy) were kept (between 87 and 99% of reads / sample). Some samples obtained very few reads (< 1.6M) and were removed from all analyses. More specific, 4 samples were filtered out: 1 ovarian cancer and 3 control samples. Quality filtered reads were then mapped with STAR (v2.6.0) using the default parameters (except for -twopassMode Basic, -outFilterMatchNmin 20 and -outSAMprimaryFlag AIIBestScore). The reference files for all analyses were based on genome build GRCh38 and transcriptome build Ensembl v91, complemented with spike annotations. Finally, gene counts were determined with HTSeq (vO.11.0) in non-stranded mode, only considering uniquely mapping reads.

[0045] Total RNA library preparation, sequencing, and quantification

[0046] Total RNA sequencing libraries of plasma cohort 2 were prepared starting from 8 pL of RNA eluate using the SMARTer Stranded Total RNA-Seq Kit v3 - Pico Input Mammalian (Takara, 634487) according to the manufacturer's protocol. Equimolar library pools were prepared based on qPCR quantification with KAPA Library Quantification Kit (Roche Diagnostics, KK4854). The libraries were paired-end sequenced (2x100 nucleotides) on a NovaSeq 6000 instrument using a NovaSeq S2 kit (Illumina, 20028315) with standard workflow loading of 0.65 nM (2% PhiX). After adapter trimming with cutadapt (vl.18) and quality control with FASTQC (vO.11.9), reads were mapped with STAR (v2.7.3) using default options and the GRCh38 reference described above. Resulting BAM files were deduplicated using UMI-tools (vl.0.0) based on the unique molecular identifier (UMI) sequences in the Pico v3 SMART UMI adapters. In the end, gene counts were determined with HTSeq (vO.ll.O) in reverse stranded mode, only considering uniquely mapping reads.

[0047] Data Availability

[0048] Raw RNA-sequencing data was deposited in the European Genome-phenome Archive (EGA). The plasma cohort 1 FASTQs can be found under study EGAS00001006755, dataset EGAD00001009713. Plasma cohort 2 FASTQs are gathered in EGAS00001007127, dataset EGAD00001010259. The FASTQs of the urine cohort are available in EGAS00001003917, dataset EGAD00001005439.

[0049] Data analyses

[0050] To identify tail genes, transcript counts of every sample were first converted to a customized z score based on the mean and standard deviation in the reference group, including all control samples except the sample of interest. More specifically, a pseudocount was added to all normalized transcript counts followed by Iog2 transformation. Then, for each gene, the mean and standard deviation of these transformed counts in the reference group was used to scale the transformed counts in the sample of interest.

[0051] Tail genes were then defined as abundant genes, at least 40 counts after DESeq2 normalization, that deviated more than 3 standard deviations from the mean of the control reference ( | z | >3).

[0052] For each gene in a sample of interest: are respectively the mean and standard deviation of Iog2 transformed counts for this gene over all (other) control samples.

[0053] Fisher's exact tests were used to select a subset of tail genes, named biomarker tail genes, that are specifically associated to the disease state. To make sure single samples did not influence results, the Fisher's exact test was done iteratively for each tail gene, each time leaving out a different sample, and a tail gene needed to be significant (p<0.05) in every iteration to be called a biomarker tail gene. Binary logistic regression, R glm function with binomial family, was used to classify a sample as cancer or control based on the number of tail genes, belonging to a certain subset of tail genes in case of pre-filtering. Leave-one-out cross-validation was used by iteratively leaving out a different sample from the training set and testing the model on this sample. Based on these predictions, the Receiver Operating Characteristic (ROC) curve and Area Under Curve (AUC) were calculated and visualized using pROC vl.18.0.

[0054] For internal validation, all sample labels were randomly reshuffled using the sample function in R without replacement. Next, tail gene identification and evaluation was repeated based on the new class labels (using the new control samples as reference). This process was repeated 20 times. Statistics

[0055] Kruskal-Wallis tests were used to compare multiple groups, and Wilcoxon rank-sum tests were used to compare two groups. For individual testing, p-values smaller than 0.05 were considered significant. In case of multiple testing, Benjamini-Hochberg procedure was used to calculate false discovery rate adjusted p-values (q-values) and significance was defined as q<0.05.

[0056] Results:

[0057] Number of mRNAs (as an example of an analyte and denominated as 'tail genes') distinguish cancer samples from controls

[0058] Analyses in our plasma cohort showed that - apart from liver cancer - plasma samples of patients with solid cancer types did not reveal evidence of a tumor (or tissue of origin) signal among mRNAs that were more abundant in cancer patient versus control plasma at group level. Moreover, these higher abundant mRNAs were less recurrent across cancer types and showed lower validation rates between cohorts compared to mRNAs with lower abundance in cancer versus control plasma. To capture differences in the cfRNA profile of an individual cancer patient, we compared mRNA abundance of a cancer plasma sample to the mRNA abundance distribution in the entire control group, which served as a reference. More specifically, a z-score was calculated for each mRNA based on the mean and standard deviation of the Iog2 normalized count distribution in the control group. Abundant mRNAs with an absolute z-score of at least 3 were defined as "tail genes" for that sample, referring to their position in one of the tails of the control z-score distribution.

[0059] We applied this tail gene analysis to every cancer sample in plasma cohort 1 and used all control samples of the same cohort as the reference group. The tail gene analysis was also applied to every individual control sample, each time using all other control samples as reference. We identified a total of 3312 unique tail genes and observed a significantly higher number of tail genes in plasma samples from cancer patients compared to controls (Kruskal-Wallis test p=0.0012). 74% of tail genes were not identified as differentially abundant mRNAs in cancer samples versus controls, and the direction of the tail gene in a specific sample could be opposite to the direction reported by the differential abundance analysis in that cancer type, which also highlights the heterogeneity between patients. Notably, the recurrence of tail genes was higher in cancer samples than controls given that 98% of the control sample tail genes were identified in one control sample only while 22 to 40% of cancer sample tail genes were also identified in other samples of the same cancer type. We then randomly swapped sample labels prior to tail gene identification and repeated this procedure 20 times. Contrary to the procedure with the original sample labels, we did not observe a significant difference in number of tail genes between the newly labeled cancer samples and controls in any of the repeats (Kruskal-Wallis tests p>0.05). This suggests that an increased number of tail genes is specifically associated to cancer samples.

[0060] Given the higher number of tail genes in cancer samples versus controls, we further assessed the binary (cancer / control) classification potential of the number of tail genes per sample. First, biomarker tail genes were selected using Fisher's exact tests (based on | z | >3 and donor cancer / control status). For each tail gene, iterative Fisher's exact tests were performed with each time one sample excluded and only consensus genes (p<0.05 in all Fisher's exact tests) were considered to avoid gene selection being biased by individual samples. 108 biomarker tail genes were obtained when comparing all cancer samples to controls and the difference in number of tail genes between cancer samples and controls was more pronounced when only considering these biomarker tail genes (Kruskal-Wallis test p=4.9E-10) instead of all tail genes. Still, 46% of biomarker tail genes were not identified as differentially abundant in any cancer type. The biomarker tail genes were found to be higher abundant in 426 cases and lower abundant in 430 cases when comparing the cancer samples to the control reference. We then built a classifier for discriminating cancer and control samples based on the number of consensus tail genes in a particular sample. After leave-one- out cross-validation, we obtained an AUC of 0.9799. When selecting biomarker tail genes based on individual cancer types versus controls, the binary classification was perfect for ovarian cancer samples and prostate cancer samples compared to controls (AUC=1, based on 143 and 46 biomarker tail genes, respectively), and still high for uterine cancer samples (AUC=0.9167, based on 13 biomarker tail genes). The type-specific biomarker tail gene sets showed significant (p<0.002) but limited overlap, which illustrates between cancer variability.

[0061] Applying the biomarker tail gene selection strategy to the dataset with randomly swapped samples labels resulted in considerably lower numbers of consensus tail genes, between 0 and 10 biomarker tail genes with a mean of 2.5 after 20 repeats, compared to 108 biomarker tail genes without label swapping. Together, these analyses suggest that individual cancer samples can be distinguished from controls based on the number of tail genes in their plasma cfRNA profiles.

[0062] To independently validate these findings, we generated plasma cfRNA profiles for a second plasma cohort consisting of plasma samples of 22 cancer-free control donors and 43 lymphoma patients, more specifically 30 diffuse large B-cell lymphoma (DLBCL) and 13 primary mediastinal large B-cell lymphoma (PMBCL) patients. We identified 5373 unique tail genes, of which 1111 were biomarker tail genes, and again observed a significant increase in the number of tail genes in cancer samples compared to controls (Kruskal-Wallis p=3.7E-9). When swapping sample labels prior to tail gene identification (20 repeats), we did not observe significant differences between the number of biomarker tail genes in cancer samples versus controls in any of the repeats (Kruskal-Wallis tests p>0.05). Biomarker tail genes detected in the control group were always unique for an individual control sample whereas 45.9% of DLBCL and 44.2% of PMBCL biomarker tail genes were shared across multiple DLBCL or PMBCL samples. A binary classifier (cancer / control) based on biomarker tail genes with leave-one-out cross-validation resulted in an AUC of 0.9567. Of note, 74.2% of these biomarker tail genes were not differentially abundant in any lymphoma subtype versus controls. The biomarker tail genes were found to be higher abundant in 2527 cases and lower abundant in 2072 cases when comparing cancer samples to the control reference. Notably, early-stage lymphoma patients also displayed an increased number of biomarker tail genes, indicating the potential for early cancer detection. Cancer diagnosis (i.e. DLBCL or PMBCL) was the only factor significantly associated to the number of biomarker tail genes (p = 1.8E-4 and 3.1E-5 for DLBCL and PMBCL, respectively) in a generalized linear model including diagnosis and tumor stage.

[0063] To test if our cancer / control classification strategy based on biomarker tail genes is also applicable to other biofluids besides plasma, we tested it on a previously generated urine cancer-control cohort8. This cohort included urine samples of 12 bladder cancer patients (BLCA) and 12 cancer-free control donors. We identified 1152 unique tail genes of which 39 biomarker tail genes. These biomarker tail genes were found to be higher abundant in 119 cases and lower abundant in 18 cases when comparing cancer samples to the control reference. The number of biomarker tail genes per sample was significantly higher in cancer samples versus controls (Wilcoxon rank-sum p-value=6.0E-5) and 61.5% of consensus tail genes were not differentially abundant between the cancer and control group. In line with the other datasets, sample label swapping prior to tail gene identification did not result in significant differences in the number of tail genes between cancer samples and controls. A binary classifier (cancer / control) based on biomarker tail genes with leave-one-out cross-validation resulted in an AUC of 0.903.

[0064] Similarly, we identified 15 biomarker tail genes in a uterotubal lavage fluid cohort comprising of samples from 25 ovarian cancer patients and from 48 donors with non-malignant gynecological conditions. The biomarker tail genes were found to be higher abundant in 27 cases and lower abundant in 33 cases when comparing cancer samples to the control reference. The number of biomarker tail genes per sample significantly differed between the cancer and benign group (Wilcoxon rank-sum p-value=1.7E-5). Though the overall AUC was only 0.565, a threshold in number of biomarker tail genes was possible at which 32% of the samples with ovarian cancer were identified as such (sensitivity) without misclassifying samples from patients with benign gynecological conditions (100% specificity), suggesting potential for a minimally invasive screening test with high positive predictive value. Finally, we tested the approach for another analyte and sample type: metabolites in tissue samples of patients with colon adenocarcinoma (n=275, covering all cancer stages) and healthy controls (n=275). Here, we identified 30 biomarker tail metabolites with 418 cases of higher abundance and 138 cases of lower abundance in colon cancer compared to the control reference. The number of biomarker tail metabolites per sample was again higher in the cancer versus control group (Wilcoxon rank-sum p<2.2E-16) with an AUC of 0.8143.

[0065] Taken together, these results confirm the cancer / control discriminative potential of the number of biomarker tail genes in cfRNA profiles in plasma and suggest that the concept may also be applicable to other sample types and analytes.

[0066] To validate the classification performance of the method, analyses are expanded towards additional biofluid cohorts (blood plasma, seminal plasma, blood, utero-tubal lavage, saliva, CSF) of cancer patients with solid tumors (high and low stage), healthy controls and patients with non-malignant disease including but not limited to infectious disease, neurological disease, metabolic disease, cardiac disease. In addition, longitudinal biofluid samples obtained from cancer patients with solid tumors before, during and after therapy are evaluated to demonstrate the performance of the method for therapy response monitoring and relapse prediction.

[0067] References

[0068] 1. Cescon, D. W., Bratman, S. V., Chan, S. M. & Siu, L. L. Circulating tumor DNA and liquid biopsy in oncology. Nat Cancer 1, 276-290 (2020).

[0069] 2. Pessoa, L. S., Heringer, M. & Ferrer, V. P. ctDNA as a cancer biomarker: A broad overview. Critical Reviews in Oncology / Hematology 155, 103109 (2020).

[0070] 3. Vorperian, S. K., Moufarrej, M. N. & Quake, S. R. Cell types of origin of the cell-free transcriptome. Nat Biotechnol 40, 855-861 (2022).

[0071] 4. Moufarrej, M. N. et al. Early prediction of preeclampsia in pregnancy with cell-free RNA. Nature 602, 689-694 (2022).

[0072] 5. Larson, M. H. et al. A comprehensive characterization of the cell-free transcriptome reveals tissue- and subtype-specific biomarkers for cancer detection. Nat Common 12, 2357 (2021).

[0073] 6. Chen, S. et al. Cancer type classification using plasma cell-free RNAs derived from human and microbes. eLife 11, e75181 (2022).

[0074] 7. Roskams-Hieter, B. et al. Plasma cell-free RNA profiling distinguishes cancers from pre-malignant conditions in solid and hematologic malignancies, npj Precis. One. 6, 1-11 (2022).

[0075] 8. Hulstaert, E. et al. Charting Extracellular Transcriptomes in The Human Biofluid RNA Atlas. Cell Reports 33, 108552 (2020).

[0076] 9. Benedetti, E. et al. A multimodal atlas of tumour metabolism reveals the architecture of genemetabolite covariation. Nat Metab 5, 1029-1044 (2023).

[0077] 10. Hulstaert, E. et al. RNA biomarkers from proximal liquid biopsy for diagnosis of ovarian cancer. Neoplasia 24, 155-164 (2022).

[0078] 11. Hulstaert, E. et al. Messenger RNA capture sequencing of extracellular RNA from human biofluids using a comprehensive set of spike-in controls. STAR Protocols 2, 100475 (2021).

Claims

Claims1. An in vitro method to screen, diagnose or monitor treatment of a disease comprising: -providing a biofluid or tissue sample taken from a possibly diseased individual and from a group of healthy control individuals,-determining in said biofluid or tissue sample taken from a possibly diseased individual the analytes having an amount which deviates more than 3 standard deviations from the mean amount of the corresponding analytes in said biofluid or tissue sample from said group of healthy control individuals, and-determining the number of said deviating analytes, wherein an increased number of said deviating analytes is used to determine whether said individual is diseased.

2. An in vitro method according to claim 1 wherein said biofluid sample is a blood plasma sample, a urine sample or a uterotubal lavage sample.

3. An in vitro method according to claims 1-2 wherein said disease is cancer.

4. An in vitro method according to claim 3 wherein said cancer is ovarian cancer, prostate cancer, uterine cancer, lymphoma, colon cancer, or bladder cancer.

5. An in vitro sample according to claims 3-4 wherein said cancer is an early stage, a locally advanced or a metastatic cancer.

6. An in vitro method according to claims 1-5 wherein said analyte is RNA or a metabolite.

7. An in vitro method according to claim 6 wherein said RNA is mRNA.

8. An in vitro method according to claims 1-7 wherein said step of determining the number of said deviating analytes can be preceded by a filtering step to define the analytes which are specifically associated to the disease state.

Images