Salivary diagnostic method by raman spectroscopy

Abstract

The object of the present invention is a method of differential diagnosis and phenotyping for respiratory diseases based on salivary analysis by Raman spectroscopy. In particular, the present invention relates to a method for analyzing a salivary sample from a subject affected or potentially affected by a chronic respiratory disease such as chronic obstructive pulmonary disease (COPD), comprising the following steps: a) providing a salivary sample from a subject on a support; b) acquiring a plurality of Raman spectra from said salivary sample and obtaining an average Raman spectrum of said salivary sample; c) performing a digital processing of said average Raman spectrum obtained in step b); d) comparing said average Raman spectrum processed according to step c) with a classification model of salivary sample categories and generating a value or a set of values of comparison; e) deriving diagnostic and / or prognostic indications related to said COPD disease based on said value or set of values of comparison.

Description

[0001] SALIVARY DIAGNOSTIC METHOD BY RAMAN SPECTROSCOPY

[0002] Description

[0003] Field of the technique of the invention

[0004] The object of the present invention is a method of differential diagnosis and phenotyping for respiratory diseases based on salivary analysis by Raman spectroscopy.

[0005] Prior art

[0006] Chronic Obstructive Pulmonary Disease (COPD) is a chronic and debilitating lung condition, with varying degrees of exacerbation, which causes limitations in physiological airflows, leading to airway remodeling, pulmonary emphysema and, in 20% of cases, death due to accelerated decline in pulmonary functions. Today, COPD is the fourth leading cause of death worldwide with more than 3 million deaths annually and an incidence between 4% and 10% with symptoms ranging from chronic cough, excessive saliva production, shortness of breath and the so-called "air hunger" (WHO, COPD). As the disease progresses, four different types of pathological phenotypes can be identified, based on the Spanish guidelines for COPD: asthma-COPD overlap (aCOPD), non-exacerbating (neCOPD), frequently exacerbating with pulmonary emphysema (eeCOPD) and frequently exacerbating with chronic bronchitis (ebCOPD). The four described types can be distinguished based on a two-year follow-up in which the clinician evaluates the number of exacerbations, the High-Resolution Computed Tomography (HRCT) results, the presence of chronic bronchitis and a-1 antitrypsin deficiency, in addition to classic pulmonary exams such as dyspnea, low FEV1 and PaCk. In particular, exacerbations are defined as events of clinical instability. It is estimated that patients affected by COPD suffer on average between 1 and 4 episodes a year, defining two categories of patients: non-exacerbators (0 events a year) and frequent exacerbators (more than two events a year).

[0007] The methodologies used to date for the phenotypic determination of patients affected by COPD, the pharmacological therapy to be adopted and the identification of an exacerbation index, require long periods of time (2-4 years), thereby limiting the applications of correct therapies in the early stages of the disease and presenting controversial results based on the different existing phenotypic classes. Further, the characterization of a phenotype is not clinically associated with a single biomarker of the patient affected by COPD, potentially leading to the assignment of generic, non-personalized pharmacological therapies. A further limit is given by the inability to predict patient exacerbations based on the phenotypic and pathological profile of the subject, except with follow-ups lasting 2 or more years. These factors, as mentioned, require long periods of time to be evaluated, drastically reducing the opportunity for prompt pharmacological intervention that can act in the initial stages of the disease.

[0008] These patients also present specific symptoms related to the different phenotype, for which different pharmacological and rehabilitative treatments are necessary, to halt the pathological progression and, at the same time, to increase the patient's quality of life. The continuous research of new therapies and new diagnostic and monitoring tools has led to embark on a global initiative for COPD called GOLD (Global Initiative For Chronic Obstructive Lung Disease), aimed at regularizing a combinatory therapy to prevent or reduce symptoms, using Inhaled Bronchodilator Steroids (IBS) and long-acting p2-agonists / long-acting muscarinic antagonists (LABA / LAMA) . While the main object of the therapy is to reduce the impact of symptoms, effective command of COPD symptoms has demonstrated to reduce factors such as the degree of exacerbation, the periodic hospitalization and mortality due to worsening symptoms, and to produce an improvement of the patient's quality of life.

[0009] As with all chronic diseases, the non-adherence to the therapy by the patient is quite common in COPD as well, leading to deteriorations in health status, increased mortality risk, and the failure of the ongoing therapy. In the same way, personalized respiratory rehabilitation plays an essential role in alleviating symptoms and slowing down pathological progression.

[0010] To date, there are no analytical methods or biomarkers to be associated with the quantitative evaluation of efficacy of the rehabilitation process, excluding the so-called walking test or "6 Minute Walk Test" which measures the distance a subject can walk as quick as possible on a flat surface in six minutes, including all the breaks the subject deems necessary . The relevance of pharmacological and rehabilitative treatments in patients affected by COPD has a substantial social and economic impact, with direct and indirect medical costs. Therapeutic non-adherence in patients affected by COPD results in an economic burden of about 300 million dollars per year in the United States, leading to the formation in 2001 of the World Health Organization Adherence Project, whose intent is to promote the research for new methodologies for monitoring and administering inhalation therapy. This initiative arose from the scarcity and low efficiency of direct and indirect methodologies to evaluate therapeutic adherence in patients affected by COPD, such as, for example, biochemical blood and urine tests, reports compiled by the patients themselves or by the pharmacists, therapeutic results associated with clinical deterioration, electronic monitoring, pill counting and inhaler weighing. All the described methodologies are inadequate for a constant, rapid, continuous, sensitive and economical therapeutic monitoring due to the limitations widely described in the literature. Further, there are no specific test to predict or evaluate the risk of exacerbation related to the individual clinical phenotype or specific subject.

[0011] Therefore, there is a need to overcome current diagnostic and prognostic limits and provide a rapid and sensitive diagnostic method, capable of evaluating and monitoring therapeutic adherence in patients affected by COPD, while simultaneously collecting useful information regarding the levels of key molecules during the physiological or pathological state. Raman spectroscopy (RS) is a vibrational spectroscopy that allows obtaining a specific and comprehensive characterization of a given biological sample in a rapid, sensitive, and non-destructive manner, without the use of additional markers for identification or special procedures for the preparation of the sample to be analyzed. RS has already been proposed in the literature for the specific analysis of different pathological states such as neoplasms, neurodegenerative diseases and pulmonary syndromes, achieving high specificity and diagnostic accuracy in all cases.

[0012] In RS the entire spectrum obtained from the sample can be used as a highly specific "fingerprint" for the selected sample (for example saliva, blood, serum, cerebrospinal fluid) that represents the diagnostic biomarker. Thanks to the high sensitivity of the methodology, even very slight biochemical variations or low concentrations of pharmacological molecules can be identified in the selected fluid.

[0013] In particular, saliva represents an ideal biofluid due to the minimal invasiveness of the collection process and the richness of information contained within it in terms of proteins, lipids, nucleic acids, hormones and metabolites.

[0014] The inventors of the present patent application have therefore conceived to apply RS in a diagnostic method for the analysis of salivary samples from subjects affected or potentially affected by

[0015] COPD. The methodology based on salivary analysis by RS is potentially capable of drastically reducing the time necessary to attribute a specific COPD phenotype to the patient under examination, determining in advance, compared to the methods used to date, the exacerbations the patient will face and the most appropriate pharmacological therapy for the subject. The procedure is also performed by analyzing a biofluid like saliva, which can be collected in minimally invasive manner allowing for easily repeated longitudinal analyses over time. The biochemical analysis of saliva can also determine specific COPD phenotypes based on a repeatable and specific signature of the biofluid, definitively determining the number of identifiable COPD phenotypes.

[0016] The patent publication US 2021 / 0307614 Al discloses the analysis of saliva samples using RS for the detection of the presence of infectious agents. The authors describe that, thanks to the use of "machine learning" it is possible to assign a value / score to the spectral data indicating the presence of the pathogen in the saliva. The method can be used to determine if a subject has been infected by the virus responsible for COVID-19.

[0017] The method involves the use of a small volume of saliva and the execution of the test in a short period of time, which can be reduced to as little as 40 minutes. The analysis of the saliva sample is conducted within an appropriate disposable support after having calibrated the Raman spectroscope. The Raman spectrum of the saliva sample is acquired using specific software and entered into a database. In parallel, the patient is tested with a gold standard methodology to verify the presence of pathogens and to perform the diagnosis, for example, of COVID-19 (for example, rapid test). The spectra obtained from the Raman analysis are saved in a specific database for subsequent comparison and machine learning analysis. Following the machine learning analysis, the classification model that provides the best sensitivity and specificity in sample classification is selected. Using the proposed classification model, the analyzed saliva samples are classified based on the score generated by the model and assigned to the group of subjects positive or negative for COVID-19.

[0018] Summary of the invention

[0019] The present invention provides a diagnostic solution that uses a new method based on saliva analysis by Raman spectroscopy. The method developed by the inventors of the present patent application provides information regarding salivary biochemical modifications associated with chronic respiratory diseases, particularly COPD, asthma and OSA (Obstructive Sleep Apnea), to be compared with the Raman signature obtained from healthy subjects. Further, the method allows to highlight the biochemical changes related to ongoing and combinatory LAMA / LABA therapy, such as the presence of the drug, its concentration, or effects on other biochemical species, and consequently, regarding adherence or non-adherence to pharmacological therapy, with the aim of providing a differential diagnosis and optimized and personalized management of patients with chronic respiratory disease.

[0020] A second aim of the present invention is to provide a method of analysis by RS which can be optimized and translated to a point-of- care platform with portable Raman.

[0021] Therefore, an object of the present invention is a diagnostic method for subjects affected or potentially affected by COPD, asthma or OSA as outlined in a any one of the appended claims. Further, the present invention could also be used for the diagnosis and monitoring of other viral diseases and neurological diseases.

[0022] In particular, the invention relates to a method for analyzing a salivary sample from a subject affected or potentially affected by a chronic respiratory disease, comprising the following steps: a) providing a salivary sample from a subject on a support; b) acquiring a plurality of Raman spectra from said salivary sample and obtaining an average Raman spectrum of said salivary sample; c) performing a digital processing of said average Raman spectrum obtained in step b); d) comparing said average Raman spectrum processed according to step c) with a classification model of salivary sample categories and generating a value or set of values of comparison; e) deriving diagnostic and / or prognostic indications related to said chronic respiratory disease based on said value or set of values of comparison. The text of the appended claims forms an integral part of the present description for the purposes of sufficiency of disclosure.

[0023] Compared to the prior art, the method of the invention allows for:

[0024] • Rapid identification of the COPD phenotype of a subject

[0025] • Performing a differential diagnosis of chronic respiratory diseases

[0026] • Pre-emptive evaluation of the number of exacerbations / year

[0027] • Obtaining indications on the appropriate pharmacological therapy for the patient

[0028] • Achieving a decisive speed of analysis

[0029] • Having minimal invasiveness of the saliva collection method

[0030] • Having reduced costs per individual analysis

[0031] • Providing immediate and readily interpretable results.

[0032] The described advantages would lead to a reduction in timeframes, an increase in the specificity of the diagnosis, and a reduction in healthcare costs.

[0033] Further characteristics and advantages of the procedure according to the invention will become apparent from the description of preferred examples of embodiments reported hereinafter, provided by way of example and not limitation, with reference to the appended Figures.

[0034] Brief description of the Figures

[0035] Figure 1 represents a block diagram of the analysis of a salivary sample from a subject; Figure 2 represents a block diagram of the "machine learning procedure for obtaining a comparison database;

[0036] Figure 3 shows the superimposition of the average Raman spectra of the COPD, ASTHMA, CTRL experimental groups;

[0037] Figure 4 shows a scatter plot relating to the values of canonical variables 1 and 2 obtained with the multivariate PCA-LDA analysis, in which each symbol represents the average spectrum of a control subject (CTRL, empty circles), a subject with COPD (filled square) or with asthma (empty triangle);

[0038] Figure 5 shows a scatter plot relating to the values of canonical variables 1 and 2 obtained with the multivariate PCA-LDA analysis, in which each symbol represents the average spectrum of a control subject (CTRL, empty circles), a subject with COPD (filled square) or with asthma (empty triangle) and wherein the 95% confidence ellipses for the three experimental groups and the values of the center and vertices are shown in the graph; a = length of the major axis; b = length of the minor axis, xc = x coordinate of the center; yc = y coordinate of the center;

[0039] Figure 6 shows the superimposition of the average Raman spectra of the OSA and CTRL experimental groups obtained by using the Raman Aramis benchtop spectroscope (Horiba Jobin-Yvon, France);

[0040] Figure 7 shows a box plot with whiskers (box-plot) relating to the values of canonical variable 1 obtained with the multivariate

[0041] PCA-LDA analysis, in which each symbol represents the value assigned to the average spectrum of each analyzed control subject (CTRL) or subject with OSA.

[0042] Detailed description of the invention

[0043] The present invention relates to a method for analyzing a salivary sample from a subject affected or potentially affected by a chronic respiratory disease, in particular COPD, asthma or OSA (Obstructive Sleep Apnea), comprising the following steps: a) providing a salivary sample from a subject; b) acquiring a plurality of Raman spectra from said salivary sample and obtaining an average Raman spectrum of said salivary sample; c) performing a digital processing of said average Raman spectrum obtained in step b); d) comparing said average Raman spectrum processed according to step c) with a classification model of salivary sample categories and generating a value or a set of values of comparison e) deriving diagnostic and / or prognostic indications related to said chronic respiratory disease based on said value or set of values of comparison.

[0044] Step a) of the method comprises a step of preparing the salivary sample.

[0045] The salivary sample is preferably acquired by having the subject chew on a swab. Appropriate commercial swabs, for example, those known under the trademark Salivette® (Sarstedt, Germany), can be used for this purpose. The swab is then centrifuged to collect the saliva sample.

[0046] Although the described method is preferred, other methods of collecting the salivary sample may also be used.

[0047] The salivary sample is deposited in the form of a drop (about 3 pl) on a support coated with a layer of aluminium. The support is typically a slide for Raman analysis. This support can be coated on at least one side with a sheet of aluminium, with the matte side facing outwards and the shiny side adhering to the support. Alternatively, commercial supports can be used, such as a slide coated with a uniform layer of aluminium (1" x 3" x 1.1 mm Microscope Slides) with a thickness of -100 nm.

[0048] Therefore, step a) comprises the following steps: al) extracting the salivary sample from a chewing swab; a2) depositing said extracted salivary sample on said support, said support being coated with a layer of aluminium, preferably matte aluminium .

[0049] Step b) of acquiring the Raman spectrum can be performed using either a benchtop Raman instrument or a portable Raman instrument. Commercial instruments are used, for example, the Aramis Raman microspectroscope (Horiba Jobin-Yvone, France) as the benchtop instrument and the iRaman Plus spectroscope (B&W Tek) as the portable instrument .

[0050] The acquisition of the Raman spectrum is performed along the edges of the drop of salivary sample deposited on the aluminium surface of the slide. In this way, it has been found that a Raman spectrum with greater informational content can be obtained.

[0051] In particular, step b) comprises the following steps: bl) providing a Raman spectroscope having a source of monochromatic electromagnetic radiation; b2) providing said support, said support being coated with aluminium, preferably matte, and comprising at least one drop of said salivary sample; b3) arranging said source of monochromatic electromagnetic radiation above the edges of said at least one drop of salivary sample; b4) acquiring a plurality of Raman spectra from said salivary sample; b5) deriving from said plurality of Raman spectra of said salivary sample an average spectrum representative of said salivary sample; wherein step b4) of acquiring said plurality of Raman spectra comprises: i) defining a first square area including a first portion of said edge of the salivary sample; ii) acquiring up to 10 Raman spectra at up to 10 different points within said first square area; iii) if at least one of said spectra is saturated or at plateau, defining a second square area including a second portion of said edge and acquiring up to 10 Raman spectra at up to 10 different points within said second square area; iv) if at least one of said acquired spectra in said second square area is saturated or at plateau, repeating step iii) until obtaining 10 unsaturated Raman spectra for each salivary sample.

[0052] In preferred embodiments, steps iii) and / or iv) are performed as soon as a saturated spectrum is acquired in the previous step, such that if, after acquiring a plurality of unsaturated spectra in step ii), the subsequent spectrum results saturated, step iii) is performed without completing the acquisition of the 10 spectra of step ii), and so on until acquisition of 10 unsaturated spectra. In other words, if after acquiring, for example, three unsaturated spectra in step ii), the fourth spectrum results saturated, step iii) is performed without completing the acquisition of the 10 spectra of step ii). The same applies for step iv) in relation to step iii) and so on.

[0053] In preferred embodiments, the unsaturated spectra from each step ii), iii) and iv) are grouped together until obtaining the plurality of 10 Raman spectra necessary for the analysis.

[0054] The source of monochromatic electromagnetic radiation preferably emits radiation with a wavelength between 780 and 790 nm, more preferably of 785 nm. Typically, the source of monochromatic electromagnetic radiation is a laser source.

[0055] The acquisition of the Raman spectrum is preferably performed in a spectral range comprised between 400 and 1600 cm-1. Step c) of digital processing of said average Raman spectrum comprises the following steps: cl) removal of artifact peaks (spikes) in said average Raman spectrum; c2) subtraction of the baseline in said average Raman spectrum; c3) peak-to-peak alignment of said average Raman spectrum with a reference peak; c4) optionally, adjustment of the resolution; c5) normalization of the peak intensities in said average Raman spectrum .

[0056] Step cl) is performed by applying the Savitzky-Golay despiking algorithm.

[0057] In step c2), the type of baseline to be fitted is polynomial, with the ability to set both the degrees of the polynomial and the maximum number of points of the baseline to be interpolated for construction and noise points. Preferably, a polynomial baseline between 4 and 5 degrees is used, with 60-80 maximum points for its construction and with 20-30 noise points, preferably without removing the aluminium spectrum.

[0058] Step c3) is performed by translating / shifting, resizing and cropping the spectra so as to align them for the subsequent comparison. The alignment is performed by comparing the position of the reference peak at 1001 cm-1and by translating / shifting the spectra that have the reference peak shifted relative to 1001 cm-1. Due to the shift during alignment, it is necessary to crop the spectra by setting the spectral range to 402-1598 cm-1.

[0059] Step c4) of resolution adjustment is optional because, only for the data acquired with the benchtop Raman spectroscope, it is important to verify that the spectra all have the same resolution. If some have lower resolution, it is necessary to identify the spectrum with the lowest resolution and set it for all. The acquired spectra according to the described method have a resolution comprised between 950 and 990 points.

[0060] Since the spectra have been aligned, it is possible to use a single frequency column (Raman shift) corresponding to the X values.

[0061] Step c5) is performed by preferably applying Unit Vector normalization to the acquired spectrum.

[0062] Steps d) and e) consist in the comparison of the average spectrum processed according to step c) with a classification model of salivary sample categories. This comparison is performed by generating one or more values / scores of comparison indicative of descriptive parameters of said average Raman spectrum.

[0063] The categories of salivary sample can be, for example, the following:

[0064] - Healthy subjects

[0065] - Subjects affected by COPD

[0066] - Subjects affected by asthma

[0067] - Subjects affected by OSA Subjects affected by particular COPD phenotypes

[0068] (subpopulations of COPD patients prone to a high rate of exacerbation).

[0069] The classification model is obtained as described hereinafter.

[0070] A plurality of Raman spectra obtained from healthy subjects and subjects with respiratory diseases are subjected to steps cl) to c5) described above.

[0071] Additionally, within each category of subjects (experimental group) an additional step c6) of calculating the mean and standard deviation of said plurality of Raman spectra can be performed.

[0072] In step c6), the spectra normalized according to step c5) are used for calculating the mean and standard deviation.

[0073] To highlight the spectral differences between experimental groups, it is possible to calculate the subtraction spectrum and visualize it graphically.

[0074] The spectra thus processed are then subjected to multivariate analysis.

[0075] Calculation of the principal components (PCs) is performed along with the relative "scores" and "loadings". Subsequently, linear discriminant analysis (LDA) is conducted. To determine the number of principal components to be used for the creation of the LDA classification model, attempts are made by selecting an increasing number of "PC scores" until the result of the classification error after "leave one out" cross-validation (error rate for cross validation) starts increasing instead of decreasing. Once the number of PCs to be used for the classification model is determined, the linear discriminant analysis is performed.

[0076] For the creation of the classification model using machine learning based on multivariate analysis (PCA-LDA), the average spectra obtained from the considered subjects are used. The information deriving from the clinical diagnosis according to standard methodology is entered into the database along with the average spectrum of each subject obtained from the Raman analysis of the salivary sample.

[0077] For the creation of the classification model using LDA, said number of PCs derived from the PCA analysis that provided the lowest classification error after "leave one out" cross-validation is used.

[0078] Therefore, obtaining the classification model comprises the following steps: i) obtaining a plurality of Raman spectra ("Training data") from healthy subjects and subjects with respiratory diseases through steps cl) to c5) as defined above; ii) submitting the spectra obtained from step i) to a multivariate analysis comprising the calculation of principal components (PCA) and their relative "scores" and "loadings" and the linear discriminant analysis (LDA), preferably using the first five principal components derived from the PCA analysis.

[0079] Further details are provided in the specific examples reported hereinafter .

[0080] EXAMPLE OF ACQUISITION AND PREPARATION OF A SALIVARY SAMPLE A swab as described above is chewed by a subject for one minute at least two hours after the last meal or the last tooth brushing. All demographic and clinical data of the subject is recorded, including saliva collection parameters (temperature, time, date).

[0081] Before saliva collection, it is important to verify: that at least two hours have passed from the last meal; that at least one hour has passed from tooth brushing; that at least one hour has passed from the last coffee or sugary beverage; ask and note any oral infections (for example, candidiasis, gingivitis, etc.); ask and note any recent dental operations; ask and note any gum bleeding; note ongoing, recent, or chronic respiratory system diseases; note if the subject is a smoker or former smoker; record, when possible with the help of a clinician, comorbidities, clinical scales, pharmacological and rehabilitative therapies of the patient.

[0082] Saliva can be stored up to 72 hours at +4°C. For longer storage periods, freezing and storage at -20°C is necessary.

[0083] The swab is then thawed (if stored at -20°C) or acclimated to room temperature (if stored a +4°C), and subsequently centrifuged to collect the saliva of the subject (preferably, 2 minutes at 1000 g). The weight of an empty Salivette is approximately 6.5 g. If the subject's Salivette has a weight less or equal to 6.5 g, the centrifugation step must be repeated increasing the number of revolutions and the minutes of centrifugation. If saliva is still not collected in this case, it is necessary to add, directly on the swab of the Salivette, 400 pl of deionized water, noting the applied dilution on the sample to account for during analysis (in most of these instances, it will not be possible to analyze the sample due to the scarcity of collected biological material).

[0084] A drop of saliva (about 3 pl) is deposited with an automatic pipette onto a commercial aluminium coated slide sized 2 x 4 cm, so that the matte part of the sheet will be facing up (for the portable Raman method) or onto an aluminium-coated slide (1" x 3" x 1.1 mm Microscope Slides) with a thickness of -100 nm for the benchtop Raman method and left to dry under a biological hood for 10 / 15 minutes. It is essential to avoid any contact of hands / gloves and adhesive tape with the aluminium part where the drops of saliva will be deposited. The same slide can be reused for several analyses, depositing one after another the samples to be analyzed on the same day, and at the end of the analyses, storing it in a suitable container protected from dust.

[0085] EXAMPLE OF ACQUISITION OF A RAMAN SPECTRUM WITH BENCHTOP RAMAN INSTRUMENT

[0086] The Aramis Raman microspectroscope (Horiba Jobin-Yvone, France) is used. The acquisitions are made with the software provided with the instrument: LabSpec6 (HORIBA-Scientific) (version 6.4.3.35). Before starting the analyses, it is important to verify that the temperature of the CCD is at -60°C for the correct functioning of the instrument and select "Laser position" to verify that the laser is centered with the cursor that will be used for orientation on the drop. Before proceeding with the analysis of the samples, the calibration of the microspectroscope is performed using the appropriate function of the Labspec program and a silicon-coated slide with a reference band at 520.7 cm-1. The calibration is performed using the acquisition parameters used for the subsequent analysis of the samples.

[0087] Specifically, the acquisition parameters are:

[0088] • Laser 785 nm

[0089] • Acquisition time = 30 sec

[0090] • Accumulation = 2

[0091] • Delay = 2 sec

[0092] • Grating = 600 gr / mm

[0093] • Hole = 400 pm

[0094] • Objective MPlan N 50x (numerical aperture 0.75 N.A., tube length °°, with maximum slide thickness equal to 0, FN22; Olympus, Japan)

[0095] • Spectral range: 400-1600 cm-1

[0096] The spectra are acquired along the edges of the drop according to the methodology previously described, where the formation of a more intensely colored edge ("coffee ring") is visible to the naked eye, containing the molecules characterizing the sample.

[0097] A map formed by 10 spectra is acquired in an automated way, for example, by setting a map sized 5 x 2 points or a linear map centered on the coffee ring area. To limit errors, it is advisable to deposit and analyze about 15 samples on the aluminium slide at a time, without leaving the dry, non-acquired samples on the slide for more than 24 hours, as the molecular components of the saliva tend to degrade, and dust could settle on the slide, reducing the collected Raman signal. The collected data is saved in TXT format.

[0098] Then proceed with the export of the average spectrum for the individual patient. Since during the analysis a single spectrum for each subject will be used, it is necessary to calculate and save the average spectrum for each map / drop of saliva directly with the Labspec6 software or to calculate the average spectrum with a data analysis program.

[0099] EXAMPLE OF ACQUISITION OF A RAMAN SPECTRUM WITH A PORTABLE RAMAN INSTRUMENT

[0100] The iRaman Plus spectroscope (B&W Tek) is used. This instrument has a resolution of < 4.5 cm-1and is associated with an optical microscope with PL 40x objective (numerical aperture 0.60 N.A., tube length 160 mm, with slide thickness of approximately 1 mm). To perform the Raman analysis, the sample must be positioned away from light. For this purpose, a cover tent for fluorescence microscopy (Sfa-Tent Eclipse Micro Tent, Nightsea) is used.

[0101] Next, the spectra of various samples are acquired in the range comprised between 400-1600 cm-1, by launching the BWSpec application (version 4.11_1 or later) and by setting the following acquisition parameters : Acquisition Time = 30 sec

[0102] • Accumulation = 2

[0103] • Laser power = 100%.

[0104] Using the camera, the laser (785 nm with 100% power) is positioned over the edges of the drop according to the methodology previously described, where the formation of a more intensely colored edge (coffee ring) is already visible to the naked eye, containing the molecules characterizing the sample, or alternatively on the present accumulations, then the brightfield image and the laser beam are focused.

[0105] Before analyzing each sample, or if requested by the instrument, the "dark" spectrum is acquired, which, subtracted from the sample spectra, nullifies all light interferences. That is, the instrument's CCD records the contribution of light in the surrounding environment before acquiring the spectrum of the biological sample, such spectrum defined as "dark" will be saved and considered to obtain the sample spectrum with the removal of the contribution of light in the environment ("dark subtracted").

[0106] Next, 10 "dark subtracted" spectra are acquired for each sample at different and distant points, possibly equidistant from each other.

[0107] It is very important not to analyze the samples following the series of experimental groups to avoid bias, it is advisable to analyze about 6 samples at a time, without leaving the dry, unanalyzed samples on the slide for more than 24 hours as the molecular components of the saliva tend to degrade and dust could settle on the slide, reducing the collected Raman signal.

[0108] The collected data is saved in TXT format, indicating the X-axis as the Raman shift and saving the spectrum in dark subtracted version.

[0109] EXAMPLE OF PROCESSING OF RAMAN SPECTRA FROM ONE OR MORE SUBJECTS

[0110] For the portable Raman spectroscope, before proceeding with data analysis, the spectra are exported and saved individually, not in map format, using a data analysis program such as, for example, Origin and the Export function.

[0111] Next, the average spectrum is calculated using a data analysis program .

[0112] Subsequently, two software programs are used for data analysis, for data acquired both with portable Raman and with benchtop Raman: LABSPEC6 with the MVA package and ORIGIN2021 or later versions (OriginLab) with the Principal Component Analysis for Spectroscopy App and then moving on with the following steps.

[0113] Removal of the spikes (i.e., "artifact" peaks): this is performed with the Despike command in Labspec6 (Processing_Correction_Despike_3 ). Alternatively, Origin2021 or later versions can be used: Processing_Smoothing_Filtering and applying the minimum degree (from 1 to 3 is sufficient), in Type, select Despike to apply the Savitzky-Golay despiking algorithm. It is essential not to remove the spectrum of the aluminium slide. Subtraction of the baseline, the command for which is Baseline

[0114] Correction in the Processing tab of Labspec6. The type of baseline to be fitted is polynomial (Poly) with the possibility to set both the degrees of the polynomial and the maximum number of points of the baseline to interpolate for construction and the noise points. Generally, for saliva spectra, a baseline between 4 and 5 degrees is used, with 60-80 maximum points for its construction and with 20- 30 noise points (for example, 5 - 70 - 25).

[0115] Spectra alignment: it is necessary to translate / shift the spectra, resize, and crop them so as to align them for the subsequent comparison. The alignment is performed by comparing the position of the reference peak at 1001 cm-1and by translating / shifting the spectra that have the reference peak shifted relative to 1001, as previously described.

[0116] To do this the "Data Calibration" command is used in "Processing", and the exact value of the reference peak is set in "Ref. Position". Due to the shift during the alignment, it may be necessary to crop the spectra in the "Data Range" section by setting the spectral range to 402-1598 cm-1.

[0117] Resolution: only for data acquired with the benchtop Raman spectroscope, it is important to verify that the spectra all have the same resolution. If some have lower resolution, it is necessary to identify the spectrum with the lowest resolution, set it for all by clicking on the "scale" in "data range". The acquired spectra according to the described method have a resolution comprised between 950 and 990 points.

[0118] The file is exported in .txt format and imported into the Origin2021 software or into another data analysis software.

[0119] Since the spectra were aligned, it is possible to use a single frequency column (Raman shift) corresponding to the X values.

[0120] Normalization: it is possible to perform the normalization of the intensity of the spectra (the Y values) preferably using the Normalization_By Norm function, which applies Unit Vector Normalization to the acquired spectra.

[0121] Calculation of mean and standard deviation for each experimental group: the data thus normalized can be used for calculating the mean and standard deviation of each experimental group under analysis. To do this, the function Statistics— - Descriptive statistics— - Statistic on rows_ Mean and Standard deviation is used.

[0122] Calculation of the subtraction spectrum: to highlight the spectral differences between experimental groups, it is possible to calculate the subtraction spectrum and visualize it graphically.

[0123] To do this, it is necessary to use the Set Column Values command and set the calculation for the subtraction between two average spectra.

[0124] Calculation of the area under the curve in the 898-947 interval: to highlight the contribution of lactate in the saliva spectrum, the area under the curve in the interval corresponding to lactate is calculated, that is, in a spectral range centered around the 918-920 cm-1peak attributable to lactate as per literature

[0125] (Movasaghi 2007).

[0126] This calculation is functional for distinguishing between respiratory disease groups, in particular between the group including COPD and asthma, and, on the other hand, Obstructive Sleep Apnea (OSA).

[0127] If the value of the area under the curve in the 898-947 cm-1range is lower than 0.1, COPD or asthma is suspected and METHODOLOGY 1 described hereinafter is followed; if instead the area is greater than 0.1, OSA is suspected and METHODOLOGY 2 is followed (Figure 2).

[0128] EXAMPLE OF OBTAINING A CLASSIFICATION MODEL OF SALIVARY SAMPLE

[0129] CATEGORIES BY MULTIVARIATE ANALYSIS

[0130] METHODOLOGY 1

[0131] For principal components analysis (PC), the previously normalized spectra are considered, and the Origin "PCA for spectroscopy" application or analogous data analysis software is used.

[0132] Each spectrum must be identified with the name of the individual sample and the name of the group to which it belongs. In the case of the Origin "PCA for spectroscopy" application, in the pop-up menu under "Frequency / Wavelength", the column with the values of the X axis, i.e. the Raman Shift, is entered. In "Group Info", instead, the box containing the information about the name of the group to which it belongs is selected, and in "Spectral Names" the box containing the information about the name of the individual samples is selected.

[0133] In the "Setting"_"Number of components to extract" tab, the number of PCs to be displayed (preferably a number greater or equal to 15) is entered. The software calculates the Principal Components and the related scores and loadings.

[0134] In the "Plots"_"Number of components to plot" tab, a value of 2 or 3 can be entered (3 to obtain a 3D graph) and after verifying that all other boxes are checked, click "Ok".

[0135] The "PcaSpecl" output tab is selected, where the data related to values / scores of the PCs (Principal Components) with the respective Loadings and plots regarding the loading and the Score plots, which are important for the evaluation of the spectral differences in the analyzed groups and for the distribution of the extracted PCs, is present. Each spectrum is assigned a score / value for each PC. The analysis continues by adding, in the "PCASpecResult1" tab, a new column renamed "Training Results", where the list of belonging groups is entered in the order in which the analyzed spectra are listed (for example, SOIBPCO => COPD). In this way, a category is assigned to each PC score.

[0136] Linear discriminant analysis (LDA) is then performed. As previously described, to determine the number of principal components to be used for the creation of the LDA classification model, attempts are made by selecting an increasing number of PC

[0137] Scores until the result from the "Error rate for cross validation starts increasing instead of decreasing. Once the number of PCs to be used for the model is determined, the analysis is performed.

[0138] Linear discriminant analysis is performed on the scores / values assigned to the individual spectra that constitute the "training data" group by selecting the following parameters in the software: equal prior probability, linear or quadratic discriminant function. The belonging categories of the Training data are selected under the "Group For Training Data" entry. The "Cross Validation" function is selected to perform the Leave-One-Out cross-validation.

[0139] Once the analysis is performed, the "Error Rate For Cross- Validation Of Training Data" value in the "Cross-Validation Summary For Training Data" tab indicates the model's accuracy, that is, it defines the percentage error in classifying the Training data group during the cross-validation test.

[0140] The Classification Summary Plot shows the main errors in the assignments .

[0141] Based on the classification errors, i.e., based on the number of individual spectra misclassified, it is possible to calculate the values of accuracy, precision, sensitivity and specificity of the model. The values of True Negatives (TN), False Negatives (FN), False Positives (FP) and True Positives (TP) allow the calculation of the following:

[0142] • Sensitivity: TP / TP+FN

[0143] • Specificity: TN / TN+FP

[0144] Precision: TP / TP+FP Accuracy: TP+TN / TP+TN+FP+FN

[0145] In the "Training Result" workbook tab, the names of the misclassified spectra and the probabilities associated with each sample in assigning a belonging category are reported. The multivariate LDA analysis also provides as an output result the values defined "Canonical Scores", which are the values related to the variables defined as Canonical Variable. In this example, from 3 categories of subjects, n-1 variables are produced, i.e., Canonical Variable 1 and Canonical Variable 2. To verify the different distribution of the Canonical Variables (CVs) associated with the different categories of subjects, they are preferably presented in a two-dimensional scatter plot.

[0146] In Figure 3, an example of Raman spectra of saliva obtained using the benchtop Raman Aramis spectroscope (Horiba Jobin-Yvon, France), according to the described protocol above, is reported. The average of the spectra obtained from the salivary samples of 70 subjects affected by COPD (solid black line), 77 subjects with ASTHMA (solid grey line) and 33 CTRL subjects (dashed black line) was calculated. The diagnosis was performed according to the guidelines of standard clinical practice.

[0147] For the creation of the classification model using machine learning based on multivariate analysis (principal component analysis and linear discriminant analysis; PCA-LDA), the average spectra obtained from the 10 spectra collected for each subject were considered. The information deriving from clinical diagnosis according to standard methodology was entered into the database along with the average spectrum of each subject obtained from the Raman analysis of the saliva using the Aramis Raman spectroscope (Horiba Jobin-Yvone, France).

[0148] For the creation of the classification model using LDA, the first 15 principal components derived from the PCA analysis were used.

[0149] After cross-validation, the data obtained demonstrates the ability of the classification model to discriminate the spectra of subjects affected by COPD from CTRL and ASTHMA subjects with an accuracy equal to 77%. Similarly, the accuracy of the model in discriminating ASTHMA group subjects from COPD and CTRL is 77.6%. The accuracy exceeds 98% when considering the ability of the model to identify subjects in good health status (CTRL) compared to subjects with respiratory diseases (COPD and ASTHMA). Overall, the classification model has an accuracy of 84.4% in discriminating CTRL, COPD and ASTHMA, with a sensitivity of 75.2% and a specificity equal to 88.6%.

[0150] More specifically, the PCA analysis (Principal Component Analysis) was performed on the average spectra of the analyzed subjects defined as "training data" and grouped according to the following order:

[0151] 1. COPD

[0152] 2. ASTHMA

[0153] 3. CTRL

[0154] Parameters of the LDA analysis (Linear Discriminant Analysis): Prior probability: equal

[0155] • Discriminant function: linear

[0156] • Cross-validation: leave one out

[0157] On the data obtained from the multivariate PCA-LDA analysis, 95% confidence ellipses were calculated.

[0158] Hereinafter are reported the coordinates related to the center and vertices of the major and minor axis of the 95% confidence ellipses with reference to the two-dimensional scatter plot shown in Figure 4 and in Figure 5 and obtained by plotting the values of Canonical Variable 1 on the x-axis, and the values of Canonical Variable 2 on the y-axis derived from the aforementioned PCA-LDA classification model.

[0159] CTRL: center (-5.261; -0.003); major axis vertices (-3.83; 1.82) (-6.81; -1.73); minor axis vertices (-6.22; 0.88) (-4.3; -0.89)

[0160] COPD: center (1.172; 0.561); major axis vertices (-1.23; 2.74) (3.57; -1.62); minor axis vertices (2.91; 2.49) (-0.57; -1.37)

[0161] ASTHMA: center (1.189; -0.509); major axis vertices (2.96; 1.39) (-0.58; -2.41); minor axis vertices (-0.41; 0.98) (2.79; -1.99).

[0162] In an analogous study performed with the portable iRaman Plus spectroscope (B&W Tek), used according to the described protocol, for the analysis of saliva samples obtained from 32 subjects, 16 of whom affected by chronic obstructive pulmonary disease (COPD) and 16 affected by ASTHMA, after cross-validation, yielded data demonstrating the ability of the classification model to discriminate the spectra of subjects affected by COPD from subjects with ASTHMA with an accuracy equal to 71.88%, sensitivity equal to 73.33%, specificity equal to 70.59% and precision of 68.75%.

[0163] METHODOLOGY 2

[0164] For the principal components analysis (PC) the previously normalized spectra are considered and the Origin "PCA for spectroscopy" application or analogous data analysis software is used.

[0165] Each spectrum must be identified with the name of the individual sample and with the name of the group to which it belongs. In the case of using the Origin "PCA for spectroscopy" application, in the pop-up menu under "Frequency / Wavelength" the column with the values of the X axis, i.e., the Raman Shift, is entered. In "Group Info", instead, the box containing the group name information to which it belongs is selected, and in "Spectral Names" the box containing the information about individual sample names is selected.

[0166] In the "Setting" tab_"Number of components to extract" the number of PCs to be displayed (preferably a number greater or equal to 15) is entered. The software calculates the principal components and the related scores and loadings.

[0167] In the "Plots"_"Number of components to plot" tab a value of 2 or 3 can be entered (3 to obtain a 3D graph) and proceed by verifying that all other boxes are selected, then click "Ok".

[0168] The "PcaSpecl" output tab is selected where the data related to the values / scores of the PCs (Principal Components) with their respective Loadings and the plots regarding the loadings, and the Score plot are present, which are important for the evaluation of the spectral differences in the analyzed groups and for the distribution of the extracted PCs. Each spectrum is assigned a score / value for each PC. The analysis continues by adding, in the "PCASpecResult1" tab, a new column renamed "Training Results" where the list of belonging groups is entered in the order in which the analyzed spectra are listed (for example, SOIOSA => OSA). In this way a category is assigned to each PC score.

[0169] Linear discriminant analysis (LDA) is then performed. As previously described, to determine the number of principal components to be used for the creation of the LDA classification model, attempts are made by selecting an increasing number of PC Scores until the result for "Error rate for cross-validation" starts increasing instead of decreasing. Once the number of PCs to be used for the model is determined, the analysis is performed.

[0170] The linear discriminant analysis is performed on the scores / values assigned to the individual spectra that make up the "training data" group by selecting the following parameters in the software: equal prior probability, linear or quadratic discriminant function. The belonging categories of the Training data are selected under the "Group For Training Data" entry. The "Cross Validation" function is selected to perform the Leave-One-Out cross-validation.

[0171] Once the analysis is performed, the "Error Rate For Cross- Validation Of Training Data" value in the "Cross-Validation Summary For Training Data" tab, indicates the model's accuracy, i.e., it defines the percentage classification error of the Training data group during the cross-validation test.

[0172] The Classification Summary Plot shows the main errors in the attributions.

[0173] Based on the classification errors, i.e., based on the number of individual spectra misclassified, it is possible to calculate the values of accuracy, precision, sensitivity and specificity of the model. The values of True Negatives (TN), False Negatives (FN), False Positives (FP) and True Positives (TP) allow the calculation of the following:

[0174] • Sensitivity: TP / TP+FN

[0175] • Specificity: TN / TN+FP

[0176] • Precision: TP / TP+FP

[0177] • Accuracy: TP+TN / TP+TN+FP+FN

[0178] In the "Training Result" workbook tab, the names of the misclassified spectra and the probabilities associated with each sample in assigning a belonging category are reported. The multivariate LDA analysis also provides as output results the values defined as "Canonical Scores", i.e., the values related to the variables defined as Canonical Variables. In this example, from 2 categories of subjects, n-1 variables are produced, i.e., Canonical Variable 1. To verify the different distribution of the Canonical Variable (CV1) associated with the different categories of subjects, a box and whiskers plot is preferably used. In Figure 6, an example of Raman spectra of saliva obtained by using the benchtop Raman Aramis spectroscope (Horiba Jobin-Yvon, France), according to the described protocol above, is reported. The average of the spectra obtained from the salivary samples of 25 subjects affected by OSA (solid black line), and 25 CTRL subjects (dashed black line) were calculated. The diagnosis was performed according to the guidelines of standard clinical practice.

[0179] For the creation of the classification model using machine learning based on multivariate analysis (principal component analysis and linear discriminant analysis; PCA-LDA), the average spectra obtained from the 10 spectra collected for each subject were considered. The information derived from the clinical diagnosis according to standard methodology was entered into the database along with the average spectrum of each subject obtained with the Raman analysis of the saliva with Aramis Raman spectroscope (Horiba Jobin- Yvone, France).

[0180] For the creation of the classification model using LDA, the first 15 principal components derived from the PCA analysis were used.

[0181] Following a cross-validation, the obtained data demonstrates the ability of the classification model to discriminate the spectra of subjects affected by OSA from the CTRL subjects with an accuracy equal to 96%. Overall, the classification model has an accuracy of 96% in discriminating between CTRL and OSA, with a sensitivity of 96% and a specificity equal to 88.6%. More specifically, the PCA analysis (Principal Component Analysis) was performed on the average spectra of the analyzed subjects, defined as "training data" and grouped according to the following order:

[0182] 1. OSA

[0183] 2. CTRL

[0184] Parameters of the LDA analysis (Linear Discriminant Analysis):

[0185] • Prior probability: equal

[0186] • Discriminant function: linear

[0187] • Cross-validation: leave one out

[0188] For the Canonical variable, the 95% confidence intervals were calculated on the data obtained from the multivariate PCA-LDA analysis.

[0189] Hereinafter are reported the minimum and maximum values of the 95% confidence interval with relation to the graph reported in Figure 7 and obtained by plotting the values of Canonical Variable 1 derived from the aforementioned PCA-LDA classification model on the X-axis.

[0190] OSA: 2.87; 3.68

[0191] CTRL: -3.69; -2.85

[0192] ~k~k~k ~k~k~k

[0193] It is evident that only certain specific embodiments of the present invention have been described, to which the skilled in the art will be able to apply all those modifications necessary for its adaptation to particular applications, without, however, departing from the scope of protection of the present invention as defined in the appended claims.

Claims

CLAIMS1. A method of analyzing a salivary sample from a subject affected or potentially affected by a chronic respiratory disease, comprising the following steps: a) providing a salivary sample from a subject on a support; b) acquiring a plurality of Raman spectra from said salivary sample and obtaining an average Raman spectrum of said salivary sample; c) performing a digital processing of said average Raman spectrum obtained in step b); d) comparing said average Raman spectrum processed according to step c) with a classification model of salivary sample categories and generating a value or a set of values of comparison; e) deriving diagnostic and / or prognostic indications related to said COPD disease based on said value or set of values of comparison.

2. The method according to claim 1, wherein step a) comprises the following steps: al) extracting the salivary sample from a chewing swab; a2) depositing said extracted salivary sample on said support, said support being coated with a layer of aluminium.

3. The method according to claim 1 or 2, wherein step b) comprises the following steps: bl) providing a Raman spectroscope having a source of monochromatic electromagnetic radiation;b2) providing said support, said support being coated with aluminium and comprising at least one drop of said salivary sample; b3) arranging said source of monochromatic electromagnetic radiation above the edges of said at least one drop of salivary sample; b4) acquiring a plurality of Raman spectra from said salivary sample; b5) deriving, from said plurality of Raman spectra of said salivary sample, an average spectrum representative of said salivary sample; wherein step b4) of acquiring said plurality of Raman spectra comprises: i) defining a first square area including a first portion of said edge of the salivary sample, ii) acquiring 10 Raman spectra at 10 different points within said first square area, iii) if at least one of said 10 spectra is saturated or at plateau, defining a second square area including a second portion of said edge and acquiring 10 Raman spectra at 10 different points within said second square area, iv) if at least one of said 10 acquired spectra in said second square area is saturated or at plateau, repeating step iii) until obtaining 10 unsaturated Raman spectra for each salivary sample.

4. The method according to claim 3, wherein steps iii) and / or iv) are performed as soon as a saturated spectrum is acquired in theprevious step, in such a way that, if after the acquisition a plurality of unsaturated spectra in step ii), the subsequent spectrum results saturated, the step iii) is performed without completing the acquisition of the 10 spectra of step ii), and so on until the acquisition of 10 unsaturated spectra.

5. The method according to claim 3 or 4, wherein the unsaturated spectra of each step ii), iii) and iv) are grouped together until obtaining said plurality of 10 unsaturated Raman spectra.

6. The method according to any one of claims 3 to 5, wherein said source of monochromatic electromagnetic radiation is a laser source that emits a radiation with wavelength comprised between 780 and 790 nm, preferably of 785 nm.

7. The method according to any one of claims 1 to 6, wherein the acquisition of the Raman spectrum is performed in a spectral range comprised between 400 and 1600 cm-1.

8. The method according to any one of claims 1 to 7, wherein step c) of digital processing of said average Raman spectrum comprises the following steps: cl) removal of artifact peaks (spikes) in said average Raman spectrum, c2) subtraction of the baseline in said average Raman spectrum, c3) peak-to-peak alignment of said average Raman spectrum with a reference spectrum, c4) optionally, adjustment of the resolutionc5) normalization of the peak intensities in said average Raman spectrum .

9. The method according to claim 8, wherein step cl) is performed by applying the Savitzky-Golay despiking algorithm and preferably without removing the aluminium spectrum.

10. The method according to claim 8 or 9, wherein the baseline is polynomial between 4 and 5 degrees, with 60-80 maximum points for its construction and with 20-30 noise points.

11. The method according to any one of claims 8 to 10, wherein in step c3) the alignment is performed by comparing the position of the reference peak at 1001 cm-1and by translating / shifting the spectra that have the reference peak shifted relative to 1001 cm-1and by setting the spectral range 402-1598 cm-1.

12. The method according to any one of claims 8 to 11, wherein step c5) is performed by applying a Unit Vector Normalization to the acquired spectrum.

13. The method according to any one of claims 1 to 12, wherein steps d) and e) consist in the comparison of the average spectrum obtained according to step c) with a classification model of salivary sample categories, wherein said comparison is performed generating one or more values of comparison indicative of descriptive parameters of said average Raman spectrum.

14. The method according to any one of claims 1 to 13, wherein the salivary sample categories are the following:- healthy subjects- subjects affected by COPD- subjects affected by asthma- subjects affected by OSA- subjects affected by particular COPD phenotypes (subpopulations of COPD patients prone to a high rate of exacerbation) .

15. The method according to any one of claims 8 to 14, wherein obtaining the classification model comprises the following steps: i) submitting a plurality of Raman spectra obtained from healthy subjects and subjects with respiratory diseases to steps cl) to c5) as defined above; ii) optionally, performing a further step c6) of calculating the mean and standard deviation of said plurality of Raman spectra; iii) submitting the spectra obtained from step i) to a multivariate analysis comprising the calculation of the principal components (PCs) and their relative "scores" and "loadings" and the linear discriminant analysis (LDA), using the first five principal components derived from the PCA analysis.

Images