Methods for analyzing peritoneal dialysis samples

Abstract

The present invention provides a method for analyzing a peritoneal dialysis analysis sample, comprising the following steps: a) providing an analysis sample from a subject, wherein the subject undergoes peritoneal dialysis and the analysis sample is based on peritoneal dialysis effluent from the subject; b) in a spectroscopic analysis step using Fourier transform infrared spectroscopy (FTIR), ‑1 Up to 400cm ‑1 c) in a comparison step, determining a similarity value by comparing the sample spectrum with at least one reference spectrum obtained by measuring at least one reference sample in the manner defined in step b); d) assigning a clinical parameter to the analysis sample based on the similarity value.

Description

Technical Field

[0001] The present invention relates to a method for analyzing a peritoneal dialysis sample. Background Art

[0002] Patients with end-stage renal disease often rely on life-saving dialysis to remove salt and water from their uremic bodies. Peritoneal dialysis (PD) offers particular advantages, such as greater distance from hospital dialysis centers, better preservation of residual renal function, and survival at the start of renal replacement therapy.

[0003] Unfortunately, these advantages are offset by the limitations of PD, including the risk of peritonitis, inflammation, and membrane failure as the main physiological cause of technical failure. The presence of bacterial infection requires rapid and accurate decision-making based on a profound diagnosis.

[0004] Events leading to unexpected discontinuation of PD can be divided into three categories: i) the patient suffers from peritonitis (i.e., inflammatory event); ii) the patient suffers from peritoneal exacerbation or iii) technical failure occurs, where technical failure is a collective term for events leading to unexpected discontinuation of PD treatment (including transplantation, death, switching to hemodialysis, etc.).

[0005] The management of PD patients currently relies mainly on systemic parameters (i.e., routine laboratory blood, plasma, or serum assessments) and the peritoneal equilibrium test (PET), which is a functional test of peritoneal transport and assesses local / peritoneal parameters. Systemic parameters may be too late to be conclusive in diagnosis and too nonspecific. PET is a laborious, time-consuming, and cost-intensive method to guide the mechanism of treatment in PD patients. Long-term monitoring of patients on home therapy and only intermittent routine examinations in tertiary care centers is also a challenge for PD patients, often resulting in late detection of complications.

[0006] In peritoneal dialysis, a special solution (PD fluid) is introduced in the lower abdomen (i.e., the peritoneal cavity), and the peritoneum acts as a membrane for exchange with the blood. Subsequently, the introduced volume is removed from the peritoneal cavity. The removed solution is the so-called PD effluent. The PD effluent is therefore a fluid obtained directly from the patient. The PD effluent contains a large amount of information about the biology of the peritoneum and the state of the membrane as well as dissolved molecules.

[0007] Various studies have led to a deeper characterization of PD effluent at the level of proteins, metabolites, and cells contained (Kratchwill 2016, Aufricht 2017, Herzog 2017, Herzog 2018, Wiesenhofer 2019), none of which has been accepted as a predictive biomarker (Aufricht 2017). For example, proteomic analysis of peritoneal dialysis fluid in patients with dialysis-associated peritonitis identified the protein β2-microglobulin as a biomarker for CAPD peritonitis (Wei-Tung 2008). Although these approaches allow for broader biological insights, they are more expensive and time-consuming than clinically applicable diagnostic tests.

[0008] Araújo et al. studied PD effluent samples using mass spectrometry and principal component analysis (PCA) to classify patients according to their kidney disease, including chemical depletion of protein components in PD effluent samples (Araújo 2016).

[0009] Spectroscopic methods have been suggested for diagnosis and analysis in dialysis. US patent application US2004 / 0204634 proposes a method for monitoring hemodialysis, comprising obtaining a Raman spectrum of a sample (i.e., a patient's blood). Raman spectroscopy is a scattering technique. Molecules that show Raman activity require a change in polarizability. Another method based on Raman spectroscopy is described in WO2015 / 164620A1 (US2017 / 045455A1). The method requires determining an analyte pattern of at least two analytes present in a sample. In an embodiment, the sample is a dialysate present in hemodialysis. In the hemodialysis fluid, urea, glucose, and phosphate are determined using Fourier transform infrared spectroscopy (FTIR) (Jensen 2004). FTIR has also been suggested as a method for cancer screening in blood samples (Barlev 2015).

[0010] To date, there are no published methods for analyzing biofluids of PD patients that allow for the assignment of clinical parameters to provide clinicians with better tools to monitor / manage dialysis regimes and / or predict PD outcomes. Summary of the invention

[0011] The present invention provides a method for analyzing a peritoneal dialysis analysis sample, comprising the following steps:

[0012] a) providing an analytical sample from a subject, wherein the subject undergoes peritoneal dialysis and the analytical sample is based on peritoneal dialysis effluent from the subject;

[0013] b) In a spectroscopic analysis step using Fourier transform infrared spectroscopy (FTIR), at 4000 cm -1Up to 400cm -1 measuring a sample spectrum of the analysis sample within a spectral range of

[0014] c) determining a similarity value in a comparison step, in particular a calculation step, by comparing the sample spectrum with at least one reference spectrum obtained from at least one reference sample measured in the manner defined in step b);

[0015] d) assigning a clinical parameter to said analyzed sample based on said similarity value.

[0016] The inventors have found that FTIR technology allows the analysis of PD effluent samples and the prediction (i.e., assignment) of clinically important parameters in a straightforward manner. A preliminary study showed that the method can assign concentrations of PD-related parameters and physiological parameters such as immune-related biomarkers with high accuracy, even based on a small data set (Example 1).

[0017] Therefore, the method according to the invention provides a rapid, non-invasive and cost-effective tool to obtain important clinical parameters in PD treatment, which allows to assess the patient status and guide PD management.

[0018] The chemometrics-assisted spectroscopy approach has many advantages over conventionally used molecular tests such as enzyme-linked immunosorbent assays (ELISAs) because it relies on a "holistic view" of the sample by studying a range of different molecules simultaneously, rather than identifying and quantifying specific compounds. Using FTIR technology, a spectral profile of the overall chemical composition of the main biochemical components in the sample (also known as a molecular spectral fingerprint) is generated.

[0019] Since FTIR also allows high-throughput analysis with low running costs (reagent-free, label-free, reusable sample holders), it is well suited for large-scale research as well as for translation into routine analysis performed every day for clinical use. DETAILED DESCRIPTION

[0020] In the methods of the present invention, the "sample" is based on peritoneal dialysis effluent (PDE) of a subject undergoing PD. PDE is a byproduct of PD treatment and the sample collection is straightforward and convenient for the patient.

[0021] In a preferred embodiment, the sample is peritoneal dialysis effluent.

[0022] Alternatively, the PDE may be subjected to sample preparation, wherein the preparation may include steps such as centrifugation and separation, dilution, freezing, thawing cooling and / or drying, etc. The sample may be prepared as a liquid or dry sample.

[0023] Thus, in another embodiment, the sample is prepared from peritoneal dialysis effluent, in particular by diluting peritoneal dialysis effluent and / or freezing and thawing peritoneal dialysis effluent.

[0024] As used herein, the term "analysis sample" refers to the sample to be analyzed using the method of the present invention and provided in step a.

[0025] On the other hand, as used herein, "at least one reference sample" refers to a sample used for comparison.

[0026] Both analytical samples and reference samples are based on the PDE of the subject matter. Therefore, the embodiments and preferred modes for preparing analytical samples apply respectively or similarly to reference samples. Preferably, at least one reference sample is prepared similarly to the analytical sample (i.e., the same or similar preparation steps are applied).

[0027] In certain embodiments, the reference sample is based on a PDE:

[0028] - a PD subject different from the subject from whom the analytical sample PDE was collected (i.e. the reference sample was from a reference subject); or

[0029] - The same subject as the subject from whom the PDE sample was collected for analysis.

[0030] In the first case, the method relies on the differences (or similarities) between individual subjects with different physiological and / or pathological states.

[0031] In the second case, it is particularly preferred that the reference samples come from the same subject, but are taken at different time points. Therefore, the method relies on different states of the subject that may change over time. The second case is of particular interest and implies a longitudinal comparison. In the case of a longitudinal comparison, similar values ​​can also be determined and clinical parameters assigned based on the relative differences with the reference samples over time, particularly in the case of an algorithm trained using a pair of reference samples collected at different time points.

[0032] Preferably, the reference sample is associated (or pre-assigned) with a certain clinical parameter. The assignment or relationship of the reference sample to a defined clinical parameter can be accomplished by routine diagnostics, for example, including evaluation of system parameters, PET data, patient history and medical documentation.

[0033] In step b) of the method of the present invention, in a spectroscopic step using Fourier transform infrared (FTIR) spectroscopy, at 4000 cm -1 Up to 400cm -1 The spectrum is measured in the spectral range.

[0034] The sample spectrum and the reference spectrum are measured independently of each other based on the analysis sample and the reference sample, respectively. However, the reference sample is measured in the manner defined in step b, so that the reference spectrum is measured similarly to the sample spectrum. Therefore, the same or similar measurement conditions (i.e., technology, spectral range, ...) are applied to measure at least one reference spectrum, and the embodiments related to the sample spectrum are similarly applicable to the at least one reference spectrum.

[0035] "FTIR spectroscopy" refers to the absorption spectroscopy technique. FTIR depends on changes in the molecular dipole moment (the opposite of Raman spectroscopy). Based on the principles of infrared spectroscopy, Fourier transforms are used to simultaneously collect and process high spectral resolution data in the infrared region of the spectrum.

[0036] The FTIR spectrum used in this article includes FTIR transmission sample technology and FTIR attenuated total reflection (FTIR-ATR). Both techniques rely on the FTIR principle. The difference between the transmission sample technology and ATR is the light path taken by the light to the sample during measurement. In the FTIR transmission sample technology, the light is transmitted through the sample. Therefore, it is preferred or required to obtain thin samples such as dry samples from liquids or fluids. The ATR technology can reflect light on the sample without transmission. Therefore, ATR can be used to analyze liquid or dry samples. In the method of the present invention, it can be preferably used as a preparation of a liquid sample, because the PDE itself is a liquid, and if the sample is a liquid, the drying process as a time-consuming preparation step can be omitted. On the other hand, compared with liquid samples, dry samples have a higher concentration of biomolecules and can obtain better signal intensity.

[0037] Therefore, in one embodiment, the spectrum in step b) is obtained by applying FTIR transmission sample technique, wherein in step a), the sample is dry. In another embodiment, the spectrum in step b) is obtained by FTIR attenuated total reflectance (ATR), wherein in step a, the sample is provided in liquid form or dry form.

[0038] The term "spectrum" as used herein refers to an optical spectrum. As understood herein, the term spectrum includes one or more sample spectral components, the spectral components including wavenumbers and absorbance values. In the spectral step b), a plurality of optical values ​​are determined within the spectral range, wherein each of the plurality of optical values ​​is associated with a certain wavenumber. Therefore, any spectrum can be a data set, i.e. a function, comprising a plurality of pairs of data points (wavenumber / absorbance value).

[0039] According to the present invention, at 4000cm -1 Up to 400cm -1 The sample spectrum is measured in a spectral region of 3000 cm. Within this range, certain sub-ranges are of particular interest because they are associated with biomolecules.-1 Up to 2800cm -1 The range of 1500cm is of interest because the signals associated with fatty acids appear in this spectral range (fatty acid region). -1 Up to 1200cm -1 The 1800 cm-1 range is of interest because signals associated with proteins and fatty acids appear in this spectral range (mixed region). -1 Up to 1500cm -1 The range of 1200cm is of interest because protein-related signals appear in this spectral range (protein region). -1 Up to 800cm -1 The range of 1800 cm is of interest because signals associated with carbohydrates appear in this spectral range (polysaccharide region). Preliminary data determined that -1 Up to 1500cm -1 、1200cm -1 Up to 800cm -1 Range, their combination (1800 to 1500cm -1 +1200 to 800cm -1 ) and 1800cm -1 Up to 800cm -1 The full range of is the preferred region for analysis of PDE based samples (Examples 1 and 3).

[0040] Therefore, in one embodiment, in step b), the sample spectrum is measured in at least one spectral range selected from the group consisting of the following spectral ranges: 3000 cm -1 Up to 2800cm -1 、1500cm -1 Up to 1200cm -1 、1800cm -1 Up to 1500cm -1 、1200cm -1 Up to 800cm -1 、1800cm -1 Up to 800cm -1 and combinations thereof (e.g., 1800 to 1500 cm -1 +1200 to 800cm -1 ), preferably 1800cm -1 Up to 800cm -1 、1800cm -1 Up to 1500cm -1 and / or 1200cm -1 Up to 800cm -1 .

[0041] In step c), a similarity value is determined by comparing the sample spectrum with at least one reference spectrum.

[0042] In certain embodiments, in step c), the comparing step comprises preprocessing the sample spectrum, the reference spectrum, or both.

[0043] As understood herein, "pre-processing" may include calculating the nth derivative of the spectrum, where n is a natural number. In the context of this specification, the term "derivative" is used in its known sense in the field of mathematics. The term "derivative of the spectrum" means the derivative of a function, where the independent variable is the wavelength or wave number, where the dependent variable is an optical value, where the optical value determined by the spectral measurement of the sample is approximated by the function. In certain embodiments, the first, second, third, fourth or fifth derivative is calculated, preferably the second derivative.

[0044] Pretreatment can also include normalization of spectrum, preferably vector normalization. Vector normalization is carried out as follows: multiple optical values ​​are determined by liquid sample, wherein each optical value is determined at a specific wave number. Subsequently, a function is provided, wherein the function includes multiple x values ​​and multiple y values, wherein each x value corresponds to a wave number, and each y value corresponds to an optical value, wherein the function describes the relationship between x values ​​and y values. In addition, the mean value of multiple y values ​​is determined, and the mean value is subtracted from each y value, thereby producing multiple corrected y values. Subsequently, the square of each corrected y value is calculated, and the sum of all squares is determined. Finally, each corrected y value is divided by all square sums, thereby forming multiple vector normalized y values, wherein the vector normalized y value and the wave number determining each y value constitute a vector normalized spectrum. Advantageously, vector normalization allows better analysis of spectrum, particularly comparison of spectrum.

[0045] In one embodiment, the pretreatment is selected from the group consisting of:

[0046] - Savitzky-Golay smoothing;

[0047] -Multiple Scatter Correction (MSC);

[0048] - Standard Normal Transformation (SNV);

[0049] - spectral derivatives, e.g. first, second, third, fourth or fifth order derivatives;

[0050] - Baseline correction, e.g., rubber band, automatic weighted least squares, Whittaker filter;

[0051] - Standardization;

[0052] - Scaling, e.g., mean centering, block scaling; and

[0053] - A combination of these.

[0054] In a preferred embodiment, said pre-processing comprises calculation of the second order derivative of said spectrum and subsequent vector normalization of the derivative to obtain a derived vector normalized spectrum.

[0055] To determine the similarity value, the analytical spectrum is compared with at least one reference spectrum measured in the manner defined above. In some embodiments, the at least one reference spectrum is stored as a data set. In this way, the reference spectrum can be used multiple times to compare sample spectra of different analytical samples.

[0056] Preferably the reference spectrum is preprocessed in a similar manner to the sample spectrum. In this embodiment, preferably, the preprocessing is not performed in step c), but the at least one reference spectrum is stored for preprocessing, for example as a derived vector normalized reference spectrum.

[0057] As used herein, a "similarity value" is a result from a comparison step, in particular a calculation step, wherein the analysis sample is directly or indirectly compared to at least one reference sample. For example, if the sample spectrum is more similar to the reference spectrum, the similarity value is greater. Depending on the comparison step, the similarity value may be a real number or a natural number. The similarity value may also be a category, such as "yes / no"; or a numerical value converted to a category, such as "low", "high" or "no result".

[0058] Thus, in certain embodiments, the similarity value is a numerical variable and / or a categorical variable.

[0059] In certain embodiments, the similarity value is determined by calculating a statistical parameter of the analysis spectrum, particularly a univariate statistical parameter, more particularly a mean or a median or a multivariate statistical parameter, and comparing the reference spectrum with the statistical parameter.

[0060] In certain embodiments, an algorithm is used to determine the similarity value in the comparing step, particularly in the calculating step, wherein the algorithm comprises the application of a machine learning model.

[0061] Algorithms based on machine learning models are preferred algorithms due to their dynamic nature, i.e., the reference sample set can be incrementally expanded to further improve the algorithm over time.

[0062] A person skilled in the art may select a method for calculating a similarity value based on the nature of the clinical parameters, the quantity and quality of the reference samples. For example, in certain embodiments, the machine learning model is selected from the group consisting of:

[0063] - Neural networks, e.g., artificial neural networks (ANN), convolutional neural networks (CNN);

[0064] - Support vector machines;

[0065] - Discriminant analysis, e.g. linear / quadratic / Mahalanobis discriminant analysis, partial least squares discriminant analysis, canonical variate analysis, discriminant function analysis;

[0066] -k-nearest neighbor algorithm (also referred to as k-NN);

[0067] - Regression analysis, e.g., linear regression, multiple linear regression, principal component regression, partial least squares regression, logistic regression;

[0068] - Evolution-based algorithms, e.g., genetic algorithms, genetic programming / computation, evolutionary programming;

[0069] - Regression and decision tree learning, e.g., Classification and Regression Trees (CART), Gradient Boosted Trees, Bootstrapped Aggregate Decision Trees, including Random Forests;

[0070] - Adaptive enhancement; and

[0071] - A combination of these.

[0072] In more specific embodiments, the algorithm is based on principal component analysis - linear discriminant analysis (PCA-LDA) or principal component analysis - Mahalanobis discriminant analysis (PCA-MDA).

[0073] As shown in the examples, clinical parameters can be successfully assigned to the investigated samples by determining similarity values ​​based on calculation of principal components (PCA), i.e., preprocessing the multivariate parameters of the spectra (second derivatives, normalization) and applying algorithms based on linear discriminant analysis (LDA) or Mahalanobis discriminant analysis (MDA).

[0074] The machine learning model is created by a process called training (supervised or unsupervised). In the method of the present invention, the training can rely on a set of reference spectra. The reference spectrum set is obtained by measuring multiple reference samples in the manner defined in step b). That is, the reference spectrum set includes multiple reference spectra. The reference spectra are measured from reference samples that are similar or similar to the sample spectra. Preferably, each reference sample is a sample in which certain clinical parameters are defined.

[0075] In embodiments where an algorithm is used, for example based on a machine learning model in calculation step c, a set of reference spectra is used to train the algorithm (or machine learning model).

[0076] When a machine learning model is applied in step c, the similarity value is determined as an output value of the model. In these cases, the comparison of the sample spectrum with at least one reference spectrum can be indirect. The machine learning algorithm includes information about the reference spectra during the training process and then applies it when calculating the similarity value based on the (processed) sample spectrum.

[0077] In step d), the analysis sample is assigned to a clinical parameter according to the similarity value determined in step c. Thus, the result of the method for analyzing a peritoneal dialysis analysis sample is a prediction of a clinical parameter of the analysis sample.

[0078] As used herein, the term "clinical parameters" includes demographic parameters, physiological and pathological parameters, including biomarker concentrations, dialysis-related parameters, and outcome parameters.

[0079] It should be understood that the clinical parameters assigned in step d) have the same properties as the specific clinical parameters pre-assigned to the reference sample. Thus, the clinical parameters that can be analyzed for the analysis sample depend on the pre-assignment of at least one reference sample or the reference spectrum set used to train the algorithm.

[0080] In a preferred embodiment, in step d), the assigned clinical parameters are selected from physiological parameters and pathological parameters including biomarker concentrations, dialysis-related parameters and outcome parameters. Assigning such clinical parameters by a method based on FTIR technology can enable clinicians to obtain parameters faster than other methods. In addition, complex outcome parameters will help guide clinicians (and / or patients) in their decision-making during PD.

[0081] In addition, other descriptive demographic parameters such as age have been shown to be discriminative for PDE spectra (Example 1). Therefore, these parameters may also be included in the step of determining the similarity value and / or may be assigned these parameters for comparison according to the method of the present invention.

[0082] Thus, in one embodiment, the clinical parameters are descriptive demographic parameters, preferably selected from dialysis regimen (eg, automated cycling PD vs. manual CAPD), PD duration, co-morbidities, blood pressure, body mass index, sex, and age.

[0083] In one embodiment, the clinical parameter is a dialysis-related parameter selected from the group consisting of residual urine volume, ultrafiltration, residual clearance, dialysate to plasma creatinine ratio, residual glomerular filtration rate, peritoneal small solute transport rate, mass transfer area coefficient, effective lymphatic absorption rate, transcapillary ultrafiltration rate, free water transport and sodium leaching.

[0084] In certain embodiments, the clinical parameters are demographic parameters selected from the group consisting of: dialysis regime (e.g., auto-cycling PD vs. manual CAPD) and PD duration, and / or dialysis-related parameters selected from the group consisting of: ultrafiltration (peritoneal ultrafiltration), dialysate to plasma creatinine ratio, peritoneal small solute transport rate, mass transfer area coefficient, effective lymphatic absorption rate, transcapillary ultrafiltration rate, free water transport, and sodium leaching. These clinical parameters are specific to peritoneal dialysis and have special significance for peritoneal dialysis management.

[0085] In another embodiment, the clinical parameter is a demographic parameter selected from the group consisting of: comorbidity, blood pressure, body mass index, sex and age; and / or a dialysis-related parameter selected from the group consisting of: residual urine volume, residual clearance and residual glomerular filtration rate. These parameters are not particularly relevant to peritoneal dialysis, but are also important, for example, during hemodialysis. Parameters such as residual urine volume, residual clearance, residual glomerular filtration rate are also of general significance for the management of water and electrolyte balance.

[0086] In another embodiment, the clinical parameter is a biomarker concentration, wherein the biomarker is selected from: proteins such as cytokines and chemokines (e.g., interleukin 8, interleukin-6) and heat shock proteins (HSPs; e.g., Hsp72) or metabolites such as amino acids (e.g., glutamine), and biogenic amines (e.g., methionine sulfoxide). Typically, the biomarker concentration represents the concentration in the PD effluent.

[0087] In another embodiment, the clinical parameter is an outcome parameter indicative of:

[0088] - The subject is at risk of having or developing peritonitis;

[0089] - The subject has or is at risk of developing peritoneal exacerbations; and / or

[0090] - Risk of technical failure.

[0091] Thus, the method according to the invention allows the analysis of peritoneal dialysis samples in risk categories for predicting the outcome of peritoneal dialysis. The outcome may be the risk of a subject suffering from or developing peritonitis, the risk of a subject suffering from or developing peritoneal deterioration and / or the risk of peritoneal dialysis technical failure.

[0092] The invention is further illustrated by the following examples, which are not to be construed as limiting the invention to the specific procedures described therein.

[0093] Example

[0094] Example 1 - Distribution of Dialysis-Related Parameters and Biomarker Concentrations

[0095] patient

[0096] PD effluent samples were obtained in a prospective, randomized, open-label, two-period crossover phase I / II study conducted at the Department of Nephrology, Medical University of Vienna (Austria). The study and study protocol were approved by the local ethics committee of the Medical University of Vienna (EK 867 / 2010, EK 1167 / 2013, EK 2035 / 2015) and registered at www.clinicaltrials.gov (NCT01353638). This study was conducted in accordance with the Declaration of Helsinki. All patients provided written informed consent before trial participation. The study design, eligibility criteria, randomization, clinical methods, patient characteristics, and adverse events have been described previously (Kratchwill 2016). In total, 20 patients with stable PD (13 men / 7 women, mean age 58 years, mean PD years 2.4 years) were treated according to the protocol. Patients were clinically stable and had no serious comorbidities (five patients had a history of peritonitis for more than 3 months before sample collection).

[0097] Effluent samples were collected in standard collection tubes without any anticoagulant (Vacuette, Bio-Greiner-One, Kremsmünster, Austria) immediately after instillation of the dialysate into the patient's peritoneal cavity (= time point 0) and 4 hours (= time point 4 hours), i.e. after two standard peritoneal equilibration tests, one with commercial PD fluid (Dianeal, 3.86% glucose, Baxter, Deerfield, IL, USA) and the other with the same PD fluid supplemented with 8 mM alanine-glutamine (Ala-Gln) (N(2)-alanine-L-glutamine 200 mg / mL, Fresenius Kabi, Bad Homburg, Germany), in random order. The two peritoneal equilibration tests were separated by a washout period of 28 to 35 days.

[0098] Sample collection and preparation

[0099] PD effluent was centrifuged immediately after collection (250°C, 10 min), and cell-free supernatant samples were aliquoted and stored at -80°C until further analysis. Biochemical measurements and transport kinetics between peritoneal and systemic circulation were performed on serum (PET test 2 h) and dialysate specimens by standard methods at the Vienna General Hospital Clinical Laboratory. PD effluent creatinine concentrations were determined by Jaff kinetic measurements and corrected for hyperglycemia by determining a correction factor for unused PD fluid measured by the same method. PD effluent interleukin 6 (IL-6) and interleukin 8 (IL-8) concentrations were determined using an immunoassay (Siemens, Vienna, Austria) or a Bio-Plex bead array (Bio-Rad, Hercules, CA).

[0100] Sample preparation and FTIR spectroscopy

[0101] PD effluent samples were randomly divided into groups and subjected to blind testing and blind analysis. First, PD effluent samples were thawed and then 30 μL of undiluted or diluted NaCl (1:2 or 1:5) was spotted on a zinc selenide (ZnSe) optical plate and dried at 40 °C for 30 min. Infrared spectra were collected in transmission mode using an HTS-XT microplate adapter coupled to a Tensor 27 FTIR spectrometer (Bruker Optics, Ettlingen, Germany) with the following parameters: 4000-500 cm-1 spectral range, 6 cm-1 spectral resolution, zero filling factor 4, BlackmannHarris 3-term apodization, and 32 interferograms, with background subtraction (Grunert 2013) for each spectrum for averaging. A total of 79 spectra from 20 PD effluent samples were recorded (0 h effluent, n=39, 4 h effluent, n=40).

[0102] Spectral preprocessing and analysis

[0103] Spectral preprocessing and subtraction spectral analysis were performed using the software OPUS 7.2 (Bruker Optics GmbH). To improve the robustness of the subsequent chemometric analysis, spectral preprocessing was performed to improve spectral resolution, minimize baseline shifts, and adjust for biomass variations between different sample preparations. The raw absorption spectra were analyzed over the entire spectral range (4000–500 cm -1) were preprocessed as follows: (1) vector normalization followed by baseline correction; (2) second-order derivative preprocessing of the raw spectra using a 9-smoothing point Savitzky-Golay filter followed by vector normalization, the latter method being used to calculate subtraction spectra, unsupervised chemometric analysis, and supervised classification (Grunert 2014, Grunert 2016).

[0104] The following supervised learning-based classification algorithm was used to classify proteins using high-resolution (1800-1500 cm -1 ) or polysaccharide (1200~800cm -1 ) spectral regions were used to classify the spectral data: principal component analysis-linear discriminant analysis (PCA-LDA); principal component analysis-Mahalanobis discriminant analysis (PCA-MDA).

[0105] Unsupervised and supervised chemometrics

[0106] All multivariate analyses were performed using the software Uncombler X (CAMO Software, Oslo, Norway). Unsupervised chemometrics were performed using principal component analysis (PCA) on preprocessed data.

[0107] PCA calculation is based on the singular value decomposition (SVD) algorithm and is performed on proteins (1800-1500 cm -1 ) or polysaccharide (1200~800cm -1 ). Score plots are used for exploratory data analysis and show possible distinctions between classes, while loading plots provide information about the spectral region.

[0108] The supervised chemometric analysis of the preprocessed data was performed using principal component analysis linear discriminant analysis (PCA-LDA) and principal component analysis-Mahalanobis discriminant analysis (PCA-MDA). Linear discriminant analysis (LDA) was used when the difference between the two classes was represented by a linear function; whereas Mahalanobis method (MDA) used an ellipse to define the distance. For PCA-LDA and PCA-MDA, the following parameters were used: 7 components of the PCA projection, equal prior probabilities were assumed, and equal weights for all variables.

[0109] The samples were divided into two different data sets: 1) PD effluent samples without Ala-Gln after a 4-hour residence time (n=20) to reflect the situation in daily PD procedures; or (2) samples of both treatments without and with Ala-Gln were combined after a 4-hour residence time (n=40) to obtain sufficient data per group to perform PCA-LDA and PCA-MDA analyses.

[0110] Clinical parameters

[0111] All available numerical datasets related to clinical parameters were divided into two balanced groups and transformed in categorical variables to implement a binary classification problem that allows the assignment of class similarity values. The classification rate (%) was used as the classification performance measure, which was defined as the average between the individual accuracies (%) for each of the two classes that accounted for the imbalanced dataset, rather than calculating the overall correct guesses (Gajjar 2013).

[0112] result

[0113] All results of principal component analysis--linear discriminant analysis (PCA-LDA) and principal component analysis--Mahalanobis discriminant analysis (PCA-MDA) were plotted as confusion matrices, and all overall accuracies are summarized in the following table, where the abbreviations represent the following: Treatment A: PD fluid supplemented with 8 mM Ala-Gln; Treatment B: standard PD fluid, CAPD: continuous ambulatory peritoneal dialysis, UF: ultrafiltration, IL: interleukin, D / P Crea: dialysate to plasma creatinine ratio, Glu: PDF dialysis history based on glucose only, Ico: PDF dialysis history based on icodextrin.

[0114] Table 1. Classification rate of patient-related parameters (%)

[0115]

[0116]

[0117] The classification results based on FTIR spectral data associated with patient characteristics, age and sex are shown in Table 1. In the protein or polysaccharide spectral region, high classification rates were obtained for the two datasets (n = 20 and n = 40) using PCA-MDA for patient age (over 60 years / under 60 years). The differences between PD samples extracted from female and male patients showed no favorite spectral range or classifier, and the highest overall accuracy was 78.3% (n = 40, PCA-MDA in the protein region). Therefore, modeling of FTIR spectral data provided evidence for the existence of age-related molecular signatures of PD effluent samples, capable of distinguishing patients younger or older than 60 years old.

[0118] Further demographic data related to treatment history, for example, treatment modality (97.5%) or duration of PD treatment (88.3%). The allocation of these data may be relevant for patient stratification and understanding of clinical phenotypes.

[0119] Table 2. Classification rate of dialysis-related parameters (%)

[0120]

[0121]

[0122] The distribution of physiological parameters related to PD treatment (Table 2) may have diagnostic value. The dialysate to plasma creatinine ratio (100.0%) and ultrafiltration (100.0%) obtained the highest classification rate in the polysaccharide area, while the protein area had the highest classification rate in the residual urine volume (100.0%) and residual clearance (89.0%) sample parameters.

[0123] Currently, parameters such as ultrafiltration and the ratio of dialysate to plasma creatinine are determined by time-extensive peritoneal equilibrium tests. Residual urine volume and residual clearance require further urine collection. It would be desirable to obtain similar information more quickly, for example, by analyzing peritoneal dialysis samples using spectroscopic methods. Furthermore, these parameters could provide a good reflection of the technical performance of the dialysis.

[0124] Table 3. Classification rate (%) of physiological parameters such as biomarker concentrations

[0125]

[0126]

[0127] In addition, referring to Table 3, clear correlations between PD spectral features and biomarkers indicative of inflammation (IL-8, IL-6) and stress response (HSP) can be shown. Heat shock protein HSP can be well predicted using protein and polysaccharide regions (100.0%). The classification rate of IL-6 prediction using polysaccharide regions was 88.1%. The protein and polysaccharide spectral regions had a discrimination rate of 83.3% for samples from patients with or without peritonitis. However, PCA-MDA was not significantly different from that of PCA-MDA by narrowing the spectral window used to 850-800 cm -1 In the polysaccharide spectral region, the overall accuracy could be further improved to 92.5% (data not shown).

[0128] Biomarker concentrations can be obtained by other methods, and some individual biomarkers have been shown to correlate with peritoneal membrane status (e.g., IL-6 effluent concentrations and peritoneal small solute transport rate (PSTR)) (Lambie 2013). However, spectral-based assignments may provide a more accessible way to obtain biomarker information.

[0129] Furthermore, these data clearly demonstrate that the sample spectra contain valuable information about the subject's physiological state (e.g., inflammation).

[0130] Example 2 - Reference Samples with Result Parameters

[0131] Based on the proof of concept presented in Example 1, it is plausible that direct correlation with patient disease status can be obtained based on sample spectra, without the need for identification and quantification of individual biomarkers, given a set of pre-assigned reference samples.

[0132] PD samples were available for a cohort of patients treated in Austria (approximately n=340), mainly patients treated at the Medical University of Vienna between 2010 and 2019.

[0133] In this group, approximately 85 experienced peritonitis episodes (total of approximately 180 infection episodes), approximately 500 peritoneal equilibration tests (characterizing membrane status and / or membrane failure and transport properties) were performed, and approximately 400 technical failure events were recorded.

[0134] In total, approximately 7800 samples were available. In some cases, multiple samples (longitudinal profiles) were available per patient, in other cases only one sample per patient was available.

[0135] Based on these data, the sampled reference samples can be assigned to clinical outcome parameters. Thus, this data forms the basis for training an algorithm that calculates similarity values ​​for the subject's PDE samples, and the PDE samples can be assigned to clinical parameters indicating the following outcomes:

[0136] - The subject is at risk of having or developing peritonitis;

[0137] - the subject has or is at risk of developing peritoneal membrane exacerbations; and / or

[0138] -Risk of technical failure.

[0139] Example 3 - Allocation of Disease Status

[0140] A subset of samples was selected from the data described in Example 2. 18 PDE samples assigned with "acute peritonitis" and 22 PDE samples taken from patients without acute peritonitis were selected as a control group. The control group was selected so that it matched the group of PDE samples with peritonitis in terms of standard gender distribution, PD treatment time and age.

[0141] Patients with peritonitis received PD for an average of 15 months (range, 0-77 months) and had a median age of 48 years (range, 1-85 years), with a similar gender distribution relative to PD effluent samples. Control samples were of the same gender, had similar duration of PD treatment (median difference, 0.2 months (IQR: 0.06-0.71)), and were of similar age (median difference, 5.9 years (IQR: 3.1-11.6)).

[0142] All PDE samples were prepared and measured by FTIR spectroscopy similar to the procedure of Example 1. The samples were measured three times, resulting in a total of 120 FTIR spectra, of which 54 spectra were classified as "peritonitis" and 66 spectra were classified as "peritonitis negative".

[0143] Analogously to Example 1, principal component analysis (PCA) was performed on the data based on preprocessed spectra (second derivatives, normalization) and an algorithm based on linear discriminant analysis (LDA) or Mahalanobis discriminant analysis (MDA) was applied.

[0144] Table 4. Classification rate of acute peritonitis based on the spectral range of specific biomolecules (%)

[0145]

[0146] Table 5. Classification rates of acute peritonitis based on a combined and larger spectrum (%)

[0147]

[0148] Tables 4 and 5 show the classification rates for different spectral ranges. The range including polysaccharide and phosphorus-containing molecule signals performed slightly better than the protein range by both analytical models (Table 4). However, the data from both ranges indicate that the peritoneal dialysis analysis method involving FTIR spectroscopy can be used to assign clinical outcome parameters to the analyzed samples.

[0149] As can be seen from the comparison of Tables 4 and 5, a wider spectral range yields even better results with respect to overall prediction accuracy. It is believed that the overall information obtained by FTIR fingerprinting helps the method used to analyze PD samples to predict complex clinical parameters even without distinguishing and identifying individual marker molecules.

[0150] Non-patent references

[0151] Araújo, J. S., Jorge, S., Magrico, R., e Costa, T., Ramos, A., Reboiro-Jato, M., Fdez-Riverola, F., Lodero, C., Ceplo JL, and Santos, HM (2016) “Classifying patients in peritoneal dialysis by massspectrometry-based profiling” Talanta 152: 364-370.

[0152] Aufricht, C., R. Beelen, M. Eberl, M. Fischbach, D. Fraser, A. Jorres, K. Kratochwill, M. Lopez Cabrera, P. Rutherford, CP Schmitt, N. Topley, and J. Witowski (2017). “Biomarker research to improve clinical outcomes of peritoneal dialysis: consensus of the European Training and Research in Peritoneal Dialysis (EuTRiPD) network.” Kidney Int 92(4):824-835.

[0153] Barlev, E., Zelig, U., Bar, O., Segev, C., Mordechai, S., Kapelushnik, J., Nathan, I., Flomen, F., Kashtan, H., Dickman, R., Madhala-Givon, O. and Wasserberg, N. (2015) “A novel method for screening colorectal cancer by infrared spectroscopy of peripheral blood mononuclear cells and plasma”, J Gastroenterol (2016) 51: 214-221.

[0154] Gajjar, K., J. Trevisan, G. Owens, P. J. Keating, N. J. Wood, H. F. Stringfellow, P. L. Martin-Hirsch, and F. L. Martin (2013). “Fourier-transform infrared spectroscopy coupled with a classification machine for the analysis of blood plasma or serum: a novel diagnostic approach for ovarian cancer.” Analyst 138(14): 3917–3926.

[0155] Grunert, T., A. Monahan, C. Lassnig, C. Vogl, M. Muller, and M. Ehling-Schulz (2014). “Deciphering host genotype-specific impacts on the metabolic fingerprint of Listeria monocytogenes by FTIR spectroscopy.” PLoS One 9(12): e115959.

[0156] Grunert, T., R. Stephan, M. Ehling-Schulz, and S. Johler (2016). “Fourier Transform Infrared Spectroscopy enables rapid differentiation of fresh and frozen / thawed chicken.” Food Control 60: 361-364.

[0157] Grunert, T., M. Wenning, M. S. Barbagelata, M. Fricker, D. O. Sordelli, F. R. Buzzola, and M. Ehling-Schulz (2013). “Rapid and reliable identification of Staphylococcus aureus capsular serotypes by means of artificial neural network-assisted Fourier transform infrared spectroscopy.” J Clin Microbiol 51(7): 2261-2266.

[0158] Herzog, R., M. Boehm, M. Unterwurzacher, A. Wagner, K. Parapatics, P. Majek, A. C. Mueller, A. Lichtenauer, K. L. Bennett, S. L. Alper, A. Vychytil, C. Aufricht, and K. Kratochwill (2018). “Effects of Alanyl-Glutamine Treatment on the Peritoneal Dialysis Effluent Proteome Reveal Pathomechanism-Associated Molecular Signatures”, Mol Cell Proteomics 17(3):516-532.

[0159] Herzog, R., L. Kuster, J. Becker, T. Gluexam, D. Pils, A. Spittler, M.K. Hasin, S.L. Alper, A. Vychytil, C. Aufricht, and K. Kratochwill (2017). “Functional and Transcriptomic Characterization of Peritoneal Immune-Modulation by Addition of Alanyl-Glutamine to Dialysis Fluid.” Sci Rep 7(1):6229.

[0160] Jensen, P. S., Bak, J., Ladefoged, S. and Andersson-Engels, S. (2003) "Determination of urea, glucose and phosphate in dialysate with Fourier transform infrared spectroscopy", Spectromchimica Acta Part A 60:899-905.

[0161] Kratochwill, K., M. Boehm, R. Herzog, K. Gruber, A. Lichtenauer, L. Kuster, D. Csaicsich, A. Gleiss, S. L. Alper, C. Aufricht, and A. Vychytil (2016). “Addition of Alanyl-Glutamine to Dialysis Fluid Restores Peritoneal Cellular Stress Responses-A First-In-Man Trial.” PLoS One 11(10): e0165045.

[0162] Lambie, M., J. Chess, KL Donovan, YL Kim, JY Do, HB Lee, H. Noh, PF Williams, AJ Williams, S. Davison, M. Dorval, A. Summers, JD Williams, J. Bankart, SJ Davies, N. Topley, and I. Global Fluid Study (2013). “Independent effects of systemic and peritoneal inflammation on peritoneal dialysis survival”, J Am Soc Nephrol 24(12):2071-2080.

[0163] Wiesenhofer, FM, R. Herzog, M. Boehm, A. Wagner, M. Unterwurzacher, DC Kasper, SL Alper, A. Vychytil, C. Aufricht, and K. Kratochwill (2019). “Targeted Metabolomic Profiling of Peritoneal Dialysis Effluents Shows Anti-oxidative Capacity of Alanyl-Glutamine,” Frontiers in Physiology 9 (1961).

[0164] Wei-Tung, L., Chung-Chang, T., Chen-Y., Wei-Jen, L., Ching-Chyuan, S. and Yu-Jen, W. (2008). “Proteomic Analysis of Peritoneal Dialysate Fluid in Patients with Dialysis-Related Peritonitis”, Renal Failure 30(8): 772-777.

Claims

1. A method for analyzing a peritoneal dialysis analysis sample, comprising the following steps: a) providing an analytical sample from a subject, wherein the subject undergoes peritoneal dialysis and the analytical sample is based on peritoneal dialysis effluent from the subject; b) In the spectral analysis step using Fourier transform infrared spectroscopy FTIR, at 4000 cm -1 Up to 400cm -1 measuring a sample spectrum of the analysis sample within a spectral range of c) determining a similarity value in a comparison step by comparing the sample spectrum with at least one reference spectrum obtained by measuring at least one reference sample in the manner defined in step b); d) assigning a clinical parameter to said analyzed sample based on said similarity value, Wherein, the similarity value is a numerical variable and / or a categorical variable, and the clinical parameter is selected from: demographic parameters; physiological and pathological parameters, including biomarker concentrations.

2. The method for analyzing a peritoneal dialysis sample according to claim 1, wherein: In the calculation step, a similarity value is determined by comparing the sample spectrum with at least one reference spectrum obtained from at least one reference sample measured in the manner defined in step b).

3. The method for analyzing a peritoneal dialysis sample according to claim 1 or 2, wherein: In step a), the analysis sample is peritoneal dialysis effluent; or, wherein the analysis sample is prepared from peritoneal dialysis effluent.

4. The method for analyzing a peritoneal dialysis sample according to claim 3, wherein: The analysis samples are prepared from peritoneal dialysis effluent by diluting the peritoneal dialysis effluent and / or freezing and thawing the peritoneal dialysis effluent.

5. The method for analyzing a peritoneal dialysis sample according to claim 1 or 2, wherein: The spectrum in step b) is obtained by applying FTIR transmission sample technology, wherein, in step a), the sample is dry; or, the spectrum in step b) is obtained by applying FTIR attenuated total reflection, wherein, in step a), the sample is provided in liquid form or dry form.

6. The method for analyzing a peritoneal dialysis sample according to claim 1 or 2, wherein: In step b), the sample spectrum is measured in at least one spectral range selected from the group consisting of: 3000 cm -1 Up to 2800cm -1 、1500cm -1 Up to 1200cm -1 、1800cm -1 Up to 1500cm -1 、1200cm -1 Up to 800cm -1 、1800cm -1 Up to 800cm -1 and their combinations.

7. The method for analyzing a peritoneal dialysis sample according to claim 1 or 2, wherein: In step b), the sample spectrum is measured in at least one spectral range selected from the group consisting of: 1800 cm -1 Up to 800cm -1 、1800cm -1 Up to 1500cm -1 、1200cm -1 Up to 800cm -1 and their combinations.

8. The method for analyzing a peritoneal dialysis sample according to claim 1 or 2, wherein: In step c), the comparing step comprises pre-processing the sample spectrum.

9. The method for analyzing a peritoneal dialysis sample according to claim 1 or 2, wherein: In the comparison step c), an algorithm is used to determine the similarity value in the comparison step, wherein the algorithm comprises the application of a machine learning model.

10. The method for analyzing a peritoneal dialysis sample according to claim 9, wherein: In the comparison step c), an algorithm is used to determine the similarity value in the calculation step, wherein the algorithm comprises the application of a machine learning model.

11. The method for analyzing a peritoneal dialysis analysis sample according to claim 9, wherein: The machine learning model is selected from neural networks, support vector machines, discriminant analysis, k-nearest neighbor algorithms, regression analysis, evolution-based algorithms, regression and decision tree learning, adaptive boosting, and combinations thereof.

12. The method for analyzing a peritoneal dialysis sample according to claim 9, wherein: The algorithm is based on principal component analysis-linear discriminant analysis or principal component analysis-Mahalanobis discriminant analysis.

13. The method for analyzing a peritoneal dialysis sample according to claim 9, wherein: The algorithm is trained using a set of reference spectra obtained by measuring a plurality of reference samples in the manner defined in step b).

14. The method for analyzing a peritoneal dialysis sample according to claim 1, wherein: The clinical parameter is the concentration of a protein biomarker indicative of inflammation.

15. The method for analyzing a peritoneal dialysis analysis sample according to claim 1, wherein: The clinical parameter is a biomarker concentration, wherein the biomarker is selected from a protein, a metabolite or a biogenic amine.

16. The method for analyzing a peritoneal dialysis analysis sample according to claim 15, wherein: The biomarker is a protein selected from the group consisting of: cytokines and chemokines.

17. The method for analyzing a peritoneal dialysis analysis sample according to claim 16, wherein: The clinical parameter is a biomarker concentration, the biomarker being a protein selected from the group consisting of: interleukin-8 or interleukin-6; and heat shock protein.

18. The method for analyzing a peritoneal dialysis analysis sample according to claim 17, wherein: The heat shock protein is HSP72.

19. The method for analyzing a peritoneal dialysis sample according to claim 1 or 2, wherein: The reference sample is based on peritoneal dialysis effluent from the same subject, which was collected at a different time than the peritoneal dialysis effluent on which the analytical sample is based, and wherein the determination of the similarity value involves a longitudinal comparison.

Images