Patent Application
20190249251
The present invention is drawn to an algorithm and method using the same algorithm for in vitro predicting the probability of a drug or a compound to induce a particular effect in a patient, said method using at least one target exhibiting an A-to-I editing of RNA. The present invention also relates to kits for the implementation of the method.
Mental disorders increasingly weight on health systems worldwide (1). They are common disorders in western societies and affect 1 out of 5 individuals at least once in their lifetime. Psychiatric disorders are caused by perturbed molecular pathways that affect brain circuitries, neurotransmission and neural plasticity. Recent work shows that alterations of epigenetic modifications on DNA and RNA such as methylation, acetylation and deamination are associated with for instance major depression, bipolar disorder and schizophrenia (2, 3). Recent studies also shed light on the importance of editing enzymes that catalyse adenosine deamination on RNA (A-to-I editing of RNA). This specific mechanism has been shown to directly regulate the function of genes encoding essentially for highly conserved neurotransmitters and synapse related factors (4-7). Importantly, the role in health and disease of this RNA editing machinery and cognate ADARs enzymes (Adenosine Deaminases Acting on RNA), has recently gained deeper ground by the accumulating evidence of its deregulation in brain of patients suffering from psychiatric disorders (8, 9). ADARs act on double stranded pre-mRNAs stem loops to specifically deaminate preferential adenosine residues. Deamination of residues residing in the coding sequence will lead to amino acid substitutions that produce receptor variants with different pharmacological properties (e.g. serotonin 2c receptor, glutamate receptor) (10).
Anomalies of serotonin biology in brain have been proposed to be a characteristic trait underlying depression and/or suicidal behaviour (11-13). By analyzing postmortem brain tissue of suicide victims, we and others have observed distinct alterations of the RNA editing activity on the serotonin receptor 2C (5HT2cR) pre-mRNA, known to greatly impair 5-HT2CR pharmacological properties (10, 14). Interestingly, these alterations in 5-HT2cR mRNA editing profile in human cortex of suicide victims partly overlaps with the interferon-induced changes observed in SH-SY5Y cells. We pinpointed specific biomarkers to characterize an ‘RNA editing signature’ of 5-HT2cR linked to depressed/suicide patients.
Several drugs belonging to different therapeutic classes have been reported to potentially induce severe psychiatric adverse effects, notably depression and suicidality (15-18). Today, there is no approved test to identify such molecules and the Food and Drugs Administration (FDA) can only issue general alerts concerning whole therapeutic classes.
Thus there is a need to provide with in vitro test which can determine with high accuracy and with high discriminate power the risk of a drug or a candidate drug to induce adverse side effects
We validated a previously designed innovative in vitro assay that predicts drug-induced psychiatric side effects using a carefully selected cell line (SH-SY5Y). We screened over 260 market-approved compounds to examine drug-induced alterations of 5-HT2cR editing. Compounds were selected from a wide range of therapeutic classes (antidepressant, antipsychotic, antiobesity, antiviral, antiinflammatory, antifungic, antiepileptic, mood stabilizing agents and others), known to potentially induce suicidality (having a FDA warning label and/or numerous case reports) or not (no psychiatric side effects reported). The data was used to identify ‘at risk’ compounds with high specificity and sensitivity.
In a first aspect, the present invention is directed to an algorithm for in vitro predicting the probability of a compound, particularly a drug to induce a or particular effects in a patient, wherein said algorithm is obtained by a method comprising the steps of:
a) selecting at least one target exhibiting an A-to-I editing of RNA, the pre-mRNA of which being the substrate of ADARs enzymes (Adenosine Deaminases Acting on RNA), the action of said ADARs leading to the production of different isoforms/or sites,
selecting at least one cell line which endogenously expresses said at least one target and at least the ADAR enzymes,
selecting a positive control compound capable of dose-dependently altering the relative proportion of said target isoform(s)/or editing site(s) when cells of said cell line are treated with said positive control,
selecting a collection of molecules composed of a ratio of drugs or compounds annotated with a risk score to induce said particular effects,
b) treating cells of said cell line with each single molecule of said collection of molecules, along with a negative control and said positive control,
c) analysing said at least one target RNA editing profile in each sample that have been treated with a molecule of the collection, in order to obtain the proportion of RNA editing level of said target for each of its editing isoforms/or sites and for each of the molecules of said collection,
d) i) by an univariable analysis statistical method, evaluating for each isoform/or editing site its accuracy and its power to discriminate the risk of a molecule to induce said particular effects; and/or
ii) by a multivariable analysis statistical method, evaluating for each combination of isoforms/or editing sites, its accuracy and its power to discriminate the risk of a molecule to induce said particular effects, and
iii) selecting the combination exhibiting the best discriminative performance,
e) building an algorithm using said selected combination of isoforms/or editing sites, and use said algorithm thus obtained for predicting the probability of a drug, compound or molecule to induce said particular effects in a patient.
By compounds, it is intended in the present description to designate mineral, chemical or biological compound, particularly which can be active on a human, animal patient, or in a plant.
In the present description, the wording “patient” also includes plant
The term “algorithm” also include statistical model (such as the Cart model).
In a preferred embodiment, in said algorithm according to the present invention said particular effects, or effect, are side effects, preferably selected from adverse or desired side effects, preferably adverse side effects.
In a preferred embodiment, said target exhibiting an A-to-I editing of RNA is selected from the group consisting of 5-HT2cR, PDE8A (Phosphodiesterase 8A), GRIA2 (Glutamate receptor 2), GRIA3, GRIA4, GRIK1, GRIK2, GRIN2C, GRM4, GRM6, FLNB (Filamin B), 5-HT2A, GABRA3 (GABAα3), FLNA, CYFIP2.
In a preferred embodiment, said particular effects, preferably side effects, more preferably desired or adverse side effects, are selecting from the group comprising cardiovascular, allergology, CNS, particularly psychiatric, dermatology, endocrinology, gastroenterology, hematology, infectiology, metabolism, neuromuscular, oncology, inflammatory and obesity, adverse side effects.
More preferred is the psychiatric adverse side effects.
In a preferred embodiment, the cell of said cell line according to the algorithm of the invention is from cell line which endogenously expressing said target and ADAR(s).
More preferably, said cell line is selected in the group consisting of:
human or animal cell line capable of endogenously expressing said target and displaying ADAR enzymes expression steady state similar to the one observed in human cortex,
neuroblastoma cell lines, preferably human cells lines,
neuroblastoma cell lines for which the positive control induced ADAR1a expression with a fold induction of at least 4, preferably at least 5 or 6 when normalised to negative or vehicule controls, and
the human SH-SY5Y cell line.
In a preferred embodiment, in step b) of the algorithm according to the present invention, the cells of said cell line are treated during a period of time comprised between 12 h and 72 h, more preferably during 48 h+/−4 h with the molecule or control to be tested, 48 h is the most preferred.
In a preferred embodiment, in the algorithm according to the invention, said positive control is the interferon alpha, or a compound able to reproduce the Interferon RNA editing profile curve at 100 IU/ml (as shown for example in
In a preferred embodiment, in the algorithm or the model according to the invention, the step c) comprises a step of determining the basal level of the RNA editing for each isoform or site in said cell line compared to vehicle treated control cells, in order to obtain for each molecules and each editing isoforms or editing site the mean/median relative proportion of RNA editing level of said target.
Preferably, said vehicle treated control cells are DMSO treated control cells.
In a preferred embodiment, in the algorithm or the model according to the invention, said method is a method for in vitro predicting the probability of a compound, particularly a drug to induce said particular effects, or effect, preferably side effects, preferably selected from adverse or desired side effects, preferably adverse side effects, with no or a low risk or a high risk, preferably with no risk or a high risk.
In a particular preferred embodiment, in the algorithm or the model according to the invention, said collection of molecules is composed of an equilibrated ratio of molecules annotated with a high risk and very low risk, preferably no risk, score to induce said particular effects, or effect, are side effects, preferably selected from adverse or desired side effects, preferably adverse side effects.
By an “an equilibrated ratio of molecules” it is intended to designate a collection of well annotated molecule for said desired adverse side effects, known to be at no or low risk or high risk to induce said adverse side effects, and presenting at least 3, preferably at least 4 or 5, different therapeutic classes, particularly selected from the group of cardiovascular, allergology, CNS, particularly psychiatric, dermatology, endocrinology, gastroenterology, hematology, infectiology, metabolism, neuromuscular, oncology, inflammatory and obesity therapeutic classes.
Preferably, the number of molecules including in each of said at least 3, 4, 5, 6, 7, or 8 different therapeutic classes, represent at least 10% of the total of the molecules of the collection.
In a more preferred embodiment, the therapeutic class representing the class of the desired particular effects, or effect, preferably side effects, preferably selected from adverse or desired side effects, preferably adverse side effects includes more than 20%, preferably, 25%, 30% or 35% of the total of the molecules of the collection.
In a preferred embodiment, in the algorithm according to the invention, in step c) said collection of molecules is analysed simultaneously, preferably at different concentrations for each molecules of the collection
In a preferred embodiment, in the algorithm according to the invention, step 1)d)i) comprises a step of calculating for each isoforms or sites, or a combination thereof:
the optimal threshold of sensitivity (Se %), of at least 60%, preferred 70% and preferably above 80% and specificity (Sp %) of at least 60%, preferred 70% and preferably above 80% for said particular effects, or effect, preferably side effects, preferably selected from adverse or desired side effects, preferably adverse side effects adverse side effect;
the positive (PPV, %) and negative (NPV, %) predictive values to evaluate the proportion of true presence [true positive/(true positive+false positive] and true absence [true negative/(true negative+false negative)], said method allowing the determination of the global performance of the choice of said isoform(s)/or site(s) or the combination thereof.
In a preferred embodiment, in the algorithm or the model according to the invention, in step c), the RNA editing profile is carried out by a method including:
NGS method (Next-Generation-Sequencing) comprising NGS library preparation, preferably using a 2-step PCR method to selectively sequence the sequence fragment of interest (comprising the editing site(s)) of the target(s);
the sequencing of all the NGS libraries obtained; and, optionally
the bioinformatics analysis of said sequencing data, said bioinformatics analysis preferably comprising the steps of:
pre-alignment processing and quality control of the sequences
the alignment against reference sequence; and
the editing levels calling,
to obtain the editing profile of the target.
In a preferred embodiment, in the algorithm according to the invention, in step d) i) and d)ii), and in step e), said statistical method allowing the obtaining of said algorithm is carried out by a method including:
mROC program, particularly to identify the linear combination, which maximizes the AUC (Area Under the Curve) ROC and wherein the equation for the respective combination is provided and can be used as a new virtual marker Z, as follows:
Z=a
1·(Isoform 1)+a2·(Isoform 2)+ . . . ai·(Isoform i)+ . . . an·(Isoform n)
where a1 are calculated coefficients and (Isoform i) are the relative proportion of individual RNA editing level of isoform's target; and/or
a logistic regression model applied for univariate and multivariate analysis to estimate the relative risk of molecules at different isoform(s)/or editing site(s) values; and/or
a CART (Classification And Regression Trees) approach applied to assess isoform(s)/or editing site(s) combinations; and/or
a Random Forest (RF) approach applied to assess the isoform/or editing site combinations, particularly to rank the importance of editing isoform/or site and to combine the best isoforms/or editing sites to classify the “relative risk” of molecule, and/or optionally
a multivariate analysis applied to assess the isoforms/or editing sites combination for the “relative risk” of molecules selecting from the group consisting of as
Support Vector Machine (SVM) approach;
Artificial Neural Network (ANN) approach;
Bayesian network approach;
wKNN (weighted k-nearest neighbours) approach;
said at least one target is the 5-HT2cR, and
said adverse side effects are psychiatric adverse side effects, and
the cell line is the human SH-SY5Y neuroblastoma cell line, and
the positive control is the interferon alpha, and
and wherein:
the sites combination capable of discriminating whether the test drug is at low risk or high risk to induce said psychiatric adverse side effects comprises at least a combination of at least 2, 3, 4 or 5 of the single sites selected from the group constituted of the following 5-HT2cR, sites:
A, B, C, D, and E,
preferably a combination of at least 3, 4 or 5 of said sites,
or the isoforms combination capable of discriminating whether the test drug is at low risk or high risk to induce said psychiatric adverse side effects comprises at least a combination of at least 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12 or 13 of the single isoforms selected from the group constituted of the following 5-HT2cR, isoforms:
A, B, AB, ABC, AC, C, D, AD, AE, ACD, AEC, ABCD and NE,
preferably a combination of at least 5, 6 or 7 of said isoforms,
and, optionally, wherein:
said statistical method allowing the obtaining of said algorithm or model is carried out by a method including:
mROC program, Random Forest approach and/or Cart algorithm.
In a second aspect, the present invention is directed to an in vitro method predicting the probability or the risk of a drug, a compound or a molecule, to induce particular effects in a patient, preferably side effects, more preferably adverse or desired side effects, said method using as a target exhibiting an A-to-I editing of RNA, the pre-mRNA of which being the substrate of ADARs enzymes, the action of said ADARs leading to the production of different isoforms or editing sites, wherein said method comprises the steps of:
A) Analysing the target RNA editing profile in sample that have been treated with said drug or compound or molecule, in order to obtain the proportion of RNA editing level of said target for each of its editing isoforms, and,
wherein said target RNA editing profile is obtained as obtained for a molecule of the collection of molecule in the algorithm or the model according to one of claims 1 to 15 obtained for said particular effects;
B) calculating the end value or applied the algorithm or model obtained for said drug or compound using the algorithm or model obtained for said target and said particular effects according to one of claims 1 to 15; and
C) determining whether said drug or compounds is at risk, particularly at low risk versus high risk, to induce said particular effects in a patient in view of the results obtained in step B).
In another embodiment, said in vitro method predicting the probability or the risk of a drug, a compound or a molecule, to induce particular effects in a patient according to the present invention, uses a combination of at least 2, 3 or 4 targets exhibiting an A-to-I editing of RNA, the pre-mRNA of which being the substrate of ADARs enzymes, the action of said ADARs leading to the production of different isoforms or sites, wherein said method comprises the steps of:
A) Analysing each of the targets RNA editing profile of said targets combination in sample that have been treated with said drug or compound or molecule, in order to obtain the proportion of RNA editing level for each of said targets for each of its editing isoforms or sites, and,
wherein said each of said targets RNA editing profile is obtained as obtained for a molecule of the collection of molecule in the algorithm or the model according to one of claims 1 to 15 obtained for said particular effects;
B) calculating the end value or applied the algorithm or model obtained for said drug or compound using the algorithm or model obtained for such of said targets and said particular effects according to one of claims 1 to 15; and
C) determining whether said drug or compounds is at risk, particularly at no risk or low risk versus high risk, to induce said particular effects in a patient in view of the results obtained in step B).
In another preferred embodiment, said combination of at least 2, 3 or 4 targets exhibiting an A-to-I editing of RNA, the pre-mRNA of which being the substrate of ADARs enzymes target exhibiting an A-to-I editing of RNA is selected from a combination of targets selected from the group consisting of 5-HT2cR, PDE8A (Phosphodiesterase 8A), GRIA2 (Glutamate receptor 2), GRIA3, GRIA4, GRIK1, GRIK2, GRIN2C, GRM4, GRM6, FLNB (Filamin B), 5-HT2A, GABRA3 (GABAα3), FLNA, CYFIP2.
In a third aspect, the present invention is directed to a kit for determining whether a compound, preferably a drug is at risk, particularly at low risk versus high risk, to induce said particular effects, or effect, preferably side effects, preferably selected from adverse or desired side effects, preferably adverse side effects adverse side effect adverse side effects in a patient comprising:
1) instructions for using an algorithm according to the invention, or to applied the method for predicting the probability or the risk of a compound or preferably a drug to induce said particular effects, or effect, preferably side effects, preferably selected from adverse or desired side effects, preferably adverse side effects in a patient according to the invention, in order to obtain the end value the analysis of which determining the risk to induce said adverse side effects in a patient for said test drug, said instructions comprising optionally a ROC curve or a Cart decision tree; and
2) reagents for determining the editing RNA profile obtained for said test drug according to the reagents need for obtaining the editing RNA profile for each molecules of the collection of molecules used for determining said algorithm or said model of said instructions of 1).
In a preferred embodiment, said reagents include the set of primers necessary for the 2-step PCR for NGS libraries preparation when using this method in the algorithm or model of the present invention.
In a more preferred embodiment, said reagents include oligonucleotides sequences used for obtaining RNA editing profile according to claims 1 to 17 for at least one of said targets or for a combination of at least 2, 3 or 4 targets.
In a more preferred embodiment, said reagents include one or a combination of a set of primers necessary for the 2-step PCR for NGS libraries preparation and wherein said at least one target or said combination of targets is selected from targets selected from the group consisting of 5-HT2cR, PDE8A (Phosphodiesterase 8A), GRIA2 (Glutamate receptor 2), GRIA3, GRIA4, GRIK1, GRIK2, GRIN2C, GRM4, GRM6, FLNB (Filamin B), 5-HT2A, GABRA3 (GABAα3), FLNA, CYFIP2.
In another more preferred embodiment, said reagents include one or a combination of a set of primers selected from the group consisted of:
The following examples and the figures and the legends hereinafter have been chosen to provide those skilled in the art with a complete description in order to be able to implement and use the present invention These examples are not intended to limit the scope of what the inventor considers to be its invention, nor are they intended to show that only the experiments hereinafter were carried out.
Other characteristics and advantages of the invention will emerge in the remainder of the description with the Examples and Figures, for which the legends are given herein below.
Figure legends:
(IFNα) 5-HT2cR mRNA editing ‘profile’ in SH-SY5Y human neuroblastoma cell line. Dose-response analysis of the effect of interferon alpha (IFNα) after 48 hours treatment with FNα. The relative proportion of 5-HT2cR mRNA was analysed by NGS-based sequencing. The profile was obtained by subtraction of the relative proportion of 5-HT2cR mRNA editing in vehicle treated control cells to the relative proportion of 5-HT2cR mRNA editing measured in IFNα treated cells.
Quantitative PCR (qPCR) analysis of ADAR1a expression in SH-SY5Y cells treated with the molecules for 48 hours. ADAR1a mRNA expression levels have been quantified in each sample after 48 hours of treatment with the molecule, vehicle (DMSO) or IFNα. A single biological replicate (n=1) is shown. As expected, each well treated with IFNα displayed increased ADAR1a expression (A to J). Of note, molecule 165 also displayed strong increase of ADAR1a mRNA expression levels post exposure to the molecule.
Raw Data of all vehicle controls and IFNα.-treated (100 UI/ml) SH-SY5Y cells (A) Global analysis of all 150 vehicle controls (DMSO) and IFN treated wells. (A) The tables are displaying all basic statistical characteristics of all 5-HT2cR mRNA editing isoforms. Vehicle and IFNα treated conditions obtained during the entire experiment (n=150) were pooled in the analysis to generate the standard measurement of IFN-induced RNA editing changes on 5-HT2cR.
(B) Histograms showing most significantly affected 5HT2cR editing isoforms by IFN treatment. Mean, median, standard deviation and coefficient of variation (CV expressed as percentage) is given for vehicle treated (DMSO) and IFNα-treated wells for all 5-HT2cR mRNA editing isoforms.
5HT2cR mRNA editing profile obtained by subtraction of the relative proportion of 5-HT2cR mRNA editing in vehicle treated control cells to the relative proportion of 5-HT2cR mRNA editing measured in IFNα treated cells. Mean and median value are given, error bars represent standard error of the mean (sem, n=150).
Decision rule: Z=0.121×ACD−0.142×NE.
Decision rule: Z=−0.1449×C+0.569×AE−0.1548×NE.
Decision rule: Z=0.0235×AB+0.1567×ACD+0.3880×AEC−0.1355×NE.
Decision rule: Z=0.016×AB−0.0563×ABC+0.183×ACD+0.386×AEC−0.1428×NE.
Decision rule: Z=0.0157×AB−0.0557×ABC+0.0187×D+0.1817×ACD+0.3883×AEC−0.1426×NE.
Decision rule: Z=−0.0505×B+0.0224×AB+0.001×D+0.163×ACD+0.389×AEC−0.1402×ABCD−0.1385×NE.
Decision rule: Z=0.2035×A+0.1283×B+0.1979×AB+0.1147×ABC+0.1860×AC+0.04331×C+0.1884×D+0.1259×AD+0.7739×AE+0.4295×ACD+0.4775×AEC−0.0415×ABCD+0.0245×NE.
A chemical library containing a collection of 1280 small molecules dissolved in DMSO at precisely 10 mM was purchased from Prestwick Chemicals. All the small molecules contained in the library are 100% approved drugs (FDA, EMA and other agencies), present the greatest possible degree of drug-likeness and have been selected for their high chemical and pharmacological diversity as well as for their known bioavailability in humans. At purchase of the chemical library (Prestwick Chemicals), a highly annotated database was provided containing detailed information on target, therapeutic class/effect, patent and ADMET of each single molecule. We searched for reports emitted for suicide and depression related adverse side effects of the drugs when prescribe to humans by inquiring databases that regularly update safety information and case reports (such as FDA Medwatch, EMEA, . . . ). Next, we compiled results of the queries and attributed a risk score to each drug contained in the chemical library. The scoring system was established in order to quantify the risk of the drugs to potentially induce adverse psychiatric side effects (depression and/or suicide related adverse side-effects) taking into account a variety of parameters such as number of cases reporting suicide and/or depression related adverse side effects, extent of prescription of the drug, being on the list of essential drugs according to the WHO and many more. We obtained a comprehensive database with specific information regarding risk to induce adverse psychiatric side-effects.
The SH-SY5Y human neuroblastoma cell line was used because it endogenously expresses the 5-HT2cR mRNA and displays an ADAR1 enzymes expression steady state similar to the one observed in human cortex (Cavarec et al. 2013, Weissmann et al. 2016 Translational Psychiatry, Patent TOXADAR). The SH-SY5Y human neuroblastoma cell line was purchased from Sigma Aldrich. Cells were routinely cultured in standard conditions at 37° C. in a humified atmosphere of 5% CO2. Dialysed Foetal Bovine Serum (FBS Science Tech reference number FB-1280D/500) was preferred to non-dialyzed because of desensitisation and down-regulation of the 5-HT2cR mRNA expression by serotonin often present in serum (Saucier et al. 1998). During the course of the experiments cells were cultured between passage number P8 and P22. Prior seeding of the cells into the 12 wells cell culture plate, estimation of the number of cells was performed by two independent loading of the trypsinized cell suspension into the Kovaslide (Kova International) chamber, a disposable microscope slide made of optically clear plastic with a hemocytometer counting grid. Both chambers were counted by two laboratory technicians and the average of the four independent counting results was further used for calculation of cell number and plating of the 12-wells cell culture plates.
Upon receipt, the entire Prestwick chemical library was transferred to individual tubes, codified, aliquoted and stored at −80° C. until further use. From our in-house generated drug-induced psychiatric adverse side-effects database we selected 260 molecules composed of an equilibrated ratio of drugs annotated with a high risk and very low risk score. The drugs were codified and care was taken to randomly process the molecules throughout the experimental setup. All 260 molecules were analysed simultaneously in each experiment along with a negative control (the vehicle DMSO) and a positive control (Interferon alpha). On each 12-well cell culture plate a negative control and a positive control was added leaving 10 vacant positions for testing molecules. In turn, each single replicate consisted of 27 culture plates of 12 wells (ref). The experiment was repeated five times in an exactly similar manner as such generating five independent biological replicates (n=5) for each tested molecule. Over the course of the experiment a total of 1620 samples were generated i.e. 27 (number of well plates) ×12 (number of wells per plate) ×5 (number of replicates). A preliminary experiment allowed identifying 7 molecules that were lethal for the SH-SY5Y cells at 10 μM. For these molecules the concentration was adapted and lowered until reduced toxicity could be detected. Prior experimentation, all dilutions of molecules and controls were prepared and arranged in racks. Cell density, morphology, viability and contamination of all 324 wells (27×12 wells) were controlled by microscope prior treatment. Additionally, a picture of each well was taken using a Canon EOS700 digital camera. Exactly 48 hours after treatment of the cells with the molecules a picture of each well was taken using the defined parameters with the digital camera. After carefully removing the growing medium 350 μl of RLT lysis buffer (Qiagen) containing 1% beta-mercaptoethanol was added for complete chemical lysis of the cells. The 12-well plates were stacked and stored in the freezer until RNA extraction.
Total RNA extraction was carried out following manufacturer's guidelines (Qiagen). The RNeasy Mini Kit provides fast purification of high-quality RNA from cells using silica-membrane RNeasy spin columns. All cell lysates were extracted using the fully automated sample preparation QIAcube. The extractions were processed using a standard procedure in batches of 12 samples (one complete 12-wells plate) per run, using appropriate protocol. During sample preparation and RNA extraction, standard precautions were taken to avoid RNA degradation by RNAses. All extracted RNA samples were analysed by labChipGx (Perkin Elmer) to both quantify and qualify the total RNA. Fluorescent-based quantification by Qubit was also performed to validate LabChipGx data. The RNA Quality Score (RQS score) was determined for each individual sample (Average RQS score of the 1620 samples=9.6/10). Next, samples were normalised and reverse transcription of the purified RNA was performed using the Takara kit (PrimeScript RT Takara ref#RR037A) was performed starting from 1 μg RNA material in a 20 μl final reaction volume. The cDNA synthesis was performed at 42° C. on a Peqstar 96x thermocycler for 15 minutes and reaction mixes were kept at 4° C. until further use.
Relative mRNA Expression by Quantitative PCR (qPCR)
After cDNA synthesis samples were stored at 4° C. prior analysis of ADAR1a mRNA expression by qPCR on a LC480 system (Roche). qPCR data were quantified using the standard curve method. mRNA expression of ADAR1a is known to be induced by Interferon alpha treatment (IFNα). As expected all samples that have been treated with IFNα for 48 hours displayed an increase of ADAR1a expression with a fold induction of gene expression between 6 and 7. In addition, Reserpine treatment did also consistently increase ADAR1a mRNA levels.
For NGS library preparation a 2-step PCR method was employed in order to selectively sequence exon V of the 5-HT2cR previously described and confirmed by us and others to be subjected to RNA editing. Validated PCR primers were used to amplify the region of interest by PCR. For PCR amplification the Q5 Hot Start High Fidelity enzyme (New England Biolabs) was used according to manufacturer guidelines (ref#M0494S). The PCR reaction was performed on a Peqstar 96x thermocycler using optimised PCR protocol. Post PCR, all samples were analysed by LabChipGx (Perkin Elmer) and both quantity and quality of the PCR product was assessed. Purity of the amplicon was determined and quantification was performed using fluorescent based Qubit method. After quality control, the 96 PCR reactions (microplate) were purified using magnetic beads (High Prep PCR MAGbio system from Mokascience). Post purification DNA was quantified using Qubit system and purification yield was calculated. Next, samples were individually indexed by PCR amplification using Q5 Hot start High fidelity PCR enzyme (New England Biolabs) and the Illumina 96 Indexes kit (Nextera XT index kit; Illumina). Post PCR, samples were pooled into a library and purified using Magbio PCR cleanup system. The library was denatured and loaded onto a sequencing cartridge according to Illumina's guidelines for sequencing FASTQ only on a MiSeq platform. A pool of plasmid containing determined amounts of 5HT2cR isoforms was included in each library to control for sequencing quality and error in each sequencing run. In addition, a standard RNA pool was incorporated into the libraries to determine variability between different sequencing flow cells during the course of the experiment. To sequence all 1620 samples, 18 MiSeq Reagent kits V3 were required (Illumina). All NGS libraries were sequenced at 14 pM and 10% Phix (PhiX Control V3) was spiked in to introduce library diversity.
The sequencing data was downloaded from the Miseq sequencer (Illumina) as fastq file. To evaluate sequencing quality, an initial quality of each raw fastq file was performed using FastQC software version 0.11.5. A pretreatment step was performed consisting of removing adapter sequences and filtering of the sequences according to their size and quality score (all short reads (<50 nts) and reads with average QC<30 were removed). Next, to facilitate and improve the quality of alignment of the sequences a flexible read trimming tool for Illumina NGS data was used (trimmomatic programs version 0.35). After pre-processing steps were performed an additional quality control of each cleaned fastq file was carried out prior further sequence processing.
Alignment of the processed reads was performed using bowtie2 version 2.2.5 with end-to-end sensitive mode. The alignment was done to the latest annotation of the human genome sequence (UCSC hg38) and reads multiple alignment regions, reads with poor alignment quality (Q<40) or reads containing insertion/deletion (INDEL) were taken out of the further analysis. Filtering of file alignment was carried out with SAMtools software version 1.2 that provide various utilities for manipulating alignments in the SAM format, including sorting, merging, indexing and generating alignments in a per-position format.
Next, SAMtools mpileup was used to pileup obtained alignment results data from multiple samples simultaneously. An in-house script was run to count the number of different ATGC nucleotides in each genomic location (‘base count’). So, for each genomic location, the home-made script computes the percentage of reads that have a ‘G’ [Number of ‘G’ reads/(Number of ‘G’ reads+Number of ‘A’ reads)*100]. The genomic location ‘A’ reference with percentage in ‘G’ reads >0.5 are automatically detected by the script and are considered as ‘A-to-I edition site’. The last stage was to compute the percentage of all possible combinations of ‘A-to-I edition site’ previously described to obtain the editing profile of the target.
We have analysed the 5HT2cR RNA editing profile of an extensive set of molecules (n=260). To compare molecules together, we have, in a first step, determined the basal level of the RNA editing of our target for each isoform/or sites in SH-SY5Y human neuroblastoma cell line compared to vehicle treated (DMSO) control cells. For this, we calculated, example given, the average of RNA editing level of 5HT2cR from over 150 vehicle independent experiments (replicates). Secondly, an in-house script has automatically computed the deviation of each replicates of molecule (n=5) to the control reference (CTRL).
Finally, for each molecules and each editing isoforms/or sites we obtained the mean/median relative proportion of RNA editing level of the target.
All statistics and figures were computed with the “R/Bioconductor” statistical open source software (19, 20). RNA editing values are usually presented as means±standard error of the mean (SEM). A differential analysis was carried out with the non-parametric Wilcoxon rank sum test and the Welch's t-test. With the multiple testing methodologies, it is important to adjust the p-value of each editing isoforms (as example: 32 RNA editing isoforms including the non-edited isoform (Ne) for 5HT2cR from 5 editing sites (A,B,C,E,D)) to control the False Discovery Rate (FDR). The Benjarnini and Hochberg (BH) procedure (21) was applied on all statistical tests with the “multtest package” and an adjusted p-value below 0.05 was considered as statistically significant. Relative proportion of editing levels was normally distributed and consequently no normalization was applied. All data distributions are illustrated as medians and barplots or boxplots for each significant isoforms. An editing profile curve from significant isoforms and representing the RNA editing level of 5HT2cR in SH-SY5Y human neuroblastoma cell line are also shown for each molecule. A Pearson test correlation was applied to identify isoforms correlation for all molecules groups.
The 5HT2cR editing isoform diagnostic performance could be characterised by: sensitivity, which represents its ability to detect the ‘high risk molecule’ group and specificity which represents its ability to detect the ‘no or low risk molecule’ group. The results of the evaluation of a diagnostic test can be summarised in a 2×2 contingency table comparing these two well-defined groups. By fixing a cut-off, the two groups could be classified into categories according to the results of the test, categorised as either positive or negative. Given a particular isoform, we can identify a number of molecules with a positive test result among the “high risk” group (the “True Positive”: TP) and b molecules with a positive test result among the “low risk” group (the “True Negative”: TN). In the same fashion, c molecules with a negative test result among the ‘high risk’ group (the “False Positive”: FP) and d molecules with a negative test result among the ‘low risk’ group (the “False Negative”: FN) are observed. Sensitivity is defined as TP/(TP+FN); which is herein referred to as the “true positive rate”. Specificity is defined as TN/(TN+FP); which is herein referred to as the “true negative rate”.
The accuracy of each 5HT2cR editing isoform and its discriminatory power was evaluated using a Receiving Operating Characteristics (ROC) analysis. ROC curves are the graphical visualization of the reciprocal relation between the sensitivity (Se) and the specificity (Sp) of a test for various values.
In addition, all 5HT2cR editing isoforms were combined with each other to evaluate the potential increase in sensibility and specificity using several approaches as mROC program [Comput. Methods Programs Biomed. 2001; 66:199-207], logistic regression (22) and with two supervised learning algorithms, CART (23) and RandomForest (24).
mROC is a dedicated program to identify the linear combination (25, 26), which maximizes the AUC (Area Under the Curve) ROC (27). The equation for the respective combination is provided and can be used as a new virtual marker Z, as follows:
Z=a×Isoform1+b×Isoform2+c×Isoform3,
where a, b, c are calculated coefficients and Isoform 1,2,3 are the relative proportion of individual RNA editing level of isoform's target.
A combination of 2, 3 or 4 targets can be combined with each other to evaluate the potential increase in sensibility and specificity using a multivariate approaches as for example mROC program or logistic regression. An equation for the respective combination can be calculated and can be used as a new virtual marker Zn, as follows:
Zn=n
1×target1+n2×target2+n3×target3,
where n1, n2, n3, are calculated coefficients and target 1,2,3 are for example a value correlated with the level of targets.
A logistic regression model was also applied for univariate and multivariate analysis to estimate the relative risk of molecules at different isoforms or sites values. We analysed isoforms as both continuous (data not shown) and categorical (using the tertile values as cutpoints) variables. In the last cases, the odds ratio (OR) and their 95% confidence interval are computed. A penalized version of the logistic regression (LASSO, ridge or Elastic-Net approaches) was also applied on continuous variables. For these methods the packages: glmnet version 2.0-3 of R software version 3.2.3 are used.
A CART (Classification And Regression Trees) approach was also applied to assess isoforms combinations. This decision tree approach allows to produce a set of classification rules, represented by a hierarchical graph easily understandable for the user. At each node of the tree, a decision is made. By convention, the left branch corresponds to a positive response to the question of interest and the right branch corresponds to a negative response to the question of interest. The classification procedure can then be translated as a set of rules ‘IF-THEN’ (see
A Random Forest (RF) approach was applied as previously to assess the isoform combinations. This method combines Breiman's “bagging” idea and the random selection of features in order to construct a collection of decision trees with controlled variance. So, random forests can be used to rank the importance of editing isoform and to combine the best isoforms to classify the “relative risk” of molecule (see
CART and RandomForest are supervised learning methods. These methods require the use of a training set used to construct the model and a test set to validate it. So, we have shared our data set: ⅔ of the dataset are used for the learning phase and ⅓ are used for the validation phase. This sharing has been randomized and respect the initial proportion of the various statutes in each sample. To estimate the errors prediction of these classifiers, we used the 10-fold cross-validation method, repeated 10 times in order to avoid overfitting problems. For these approaches, we used the the “rpart package 4.1-10” and the “randomForest package 4.6-12” of the R software version 3.2.3.
Another multivariate analysis may be used to assess 5HT2cR editing isoforms combination for the “relative risk” of molecules as:
Prior to the experiment, the human neuroblastoma cell line (SH-SY5Y) was treated with an increasing dose of interferon and RNA editing of 5HT2cR was measured using NGS based approach. As expected, the relative proportion of the 5HT2cR isoforms is altering and, particularly can, increase dose-dependently (
Only once the cell-line showed stable grow characteristics and responded accordingly to IFN treatment, the screen of 260 molecules was prepared. Based on in-house defined criteria a risk score was attributed to each of the 1280 molecules in the chemical library. For practical reasons 260 molecules were selected to further test on proprietary in vitro assay. During selection procedure of the molecules, care was taken to cover part of all, preferably at least 3, 4, 5, 6 or 7 of the major therapeutic classes, identified in the
Relative ADAR1a mRNA Expression
Following RNA extraction, cDNA was synthesised and ADAR1a expression was assessed on a LC480 lightcycler (Roche) in a 384-micro wells plate. In this way, all samples of the same batch could be analysed in a single qPCR run. An interferon dependent induction of ADAR1a was observed for all IFN treated cells on each 12-wells plate reflecting robustness of the response. Interestingly molecule 165 also induced ADAR1a mRNA expression (
Post the cDNA synthesis step, a 2-step PCR approach to target exon V of the 5HT2cR was applied to build NGS libraries and accurately quantify the relative proportion of each individual 5HT2cR mRNA in all samples. The mean value of all vehicle controls and IFN treated wells (n=150) is displayed in
As example, when comparing the levels of 5-HT2cR RNA editing isoforms in the presence of IFN (n=150) to vehicle control (vehicle, n=150) on SH-SY5Y cell lines, AC, ABC, AB, A, AE, ACE, D, ABCD, ABE, C, B, BC and ABCE RNA editing levels of 5-HT2cR were significantly altered (
As example, when comparing molecules with low risk (n=82) to molecules with high risk (n=61), single editing or non-edited (Ne) levels of 5-HT2cR isoforms can be significantly altered (
The accuracy of each isoforms and its discriminatory power was evaluated using a Receiving Operating Characteristics (ROC) analysis. ROC curves are the graphical visualization of the reciprocal relation between the sensitivity (Se) and the specificity (Sp) of a test for various values. AUC means area under the curve, with its confidence interval (CI). ROC Curves are based on models of prediction of relative risk of molecules by calculating optimal threshold of sensitivity (Se %) and specificity (Sp %) for single marker. Positive (PPV, %) and negative (NPV, %) predictive values for single RNA editing isoforms were calculated to evaluate the proportion of true presence [true positive/(true positive+false positive] and true absence [true negative/(true negative+false negative)] of high risk molecules in ‘suicide side-effect group’.
Multiple marker analysis with mROC (multiple Receiving-Operating-Characteristic) approach improved significantly AUC when comparing low risk to high risk molecules. The isoforms combination associated for example 2, 3, 4, 5, 6, 7 or the 13 isoforms selected from the group of the 13 isoforms of the following combination: A+B+AB+ABC+AC+C+D+AD+AE+ACD+AEC+ABCD+NE, combination obtained by the method of the present invention, has a predictive value for higher risk of suicide side-effect in high risk molecules as reported by the higher sensitivity and specificity than those obtained in Cavarec et al. (2013). The statistical analysis combining 2, 3, 4, 5, 6, 7, 8, 9, 0, 11, 12 and the 13 isoforms as identified in the combination of the present invention, generated a series of decision rules; a new virtual marker (Z) was calculated for each combination as illustrated in
The accuracy of multi-isoforms panel and its discriminatory power was evaluated using a Receiving Operating Characteristics (ROC) analysis. ROC curves are the graphical visualization of the reciprocal relation between the sensitivity (Se) and the specificity (Sp) of a test for various values. AUC means area under the curve, with its confidence interval (CI). ROC Curves are based on models of prediction of high risk of toxicity by calculating optimal threshold of sensitivity (Se %) and specificity (Sp %) for multi-isoforms panel. Positive (PPV, %) and negative (NPV, %) predictive values for combined marker were calculated to evaluate the proportion of true presence [true positive/(true positive+false positive] and true absence [true negative/(true negative+false negative)] of high risk molecule of suicide/depression inducing adverse side effects.
CART algorithm which stands for “Classification And Regression Trees” is a decision tree approach. These trees will help to build a set of classification rules, represented as a hierarchical graph easily understandable for the user. The tree consists of internal node (decision node), edge and terminal leaf. These nodes are labeled by tests and possible responses to the test match with the labels of edges from this node. If the decision tree is binary, by convention, the left edge corresponds to a positive response to test and right edge correspond to the negative response. The procedure for classification obtained will have an immediate translation in terms of decision rule.
Decision trees are popular and efficient methods of supervised classification. This method requires the use of a training set to construct the model and a test set to validate it. So, for building the dataset, we have shared our list of ‘no ambiguous’ molecules (n=143): 90% of the dataset are used for the learning phase (n=93 drugs) and 10% are used for the test phase (50 drugs). This sharing has been randomized and respects the initial proportion of the various statutes in each molecule. Moreover,
As example we have combined the 6 RNA editing isoforms in the ‘IFN profile’, with CART method for building a model of decision making (
The diagnostic performances of CART model using 5 RNA editing isoforms of 5-HT2cR on the data test can be also very interesting for discriminating the low risk molecules versus high risk molecules.
RandomForest is a popular and efficient method of supervised classification. This method requires the use of a training set to construct the model and a test set to validate it. So, for building the dataset, we have shared our list of ‘no ambiguous’ molecules (n=143): 65% of the dataset are used for the learning phase (n=93 drugs) and 35% are used for the test phase (50 drugs). This sharing has been randomized and respects the initial proportion of the various statutes in each molecule. Moreover, we have weighting the learning dataset by IFN to improve the separation power of drugs with ‘IFN profile’ and drugs with ‘basal0 profile’. So, we have added 12 IFN molecules and 8 control (basal0) taken randomly in the learning set (n=113).
As example, we have combined 7 and the 13 representatives RNA editing isoforms (See RD1 of C7, Table 9, and C13, Table 10) in the ‘IFN profile’, with RandomForest (RF) algorithm for building a model of decision making (Parameters of RF model: mtryStart=1, stepFactor=2, ntree=500, improve=0.01; Out Of Bag (OOB) estimate of RF model=0.21) (
The diagnostic performances of RF model using 7 or 13 RNA editing isoforms of 5-HT2cR are very interesting for discriminating the low risk molecules versus high risk molecules with a sensitivity, specificity and accuracy superior to 90% (for C7) and superior to 95% (for C13), high significantly superior to those disclosed in Cavarec et al (2013).
To further supplement the 5HT2cR mRNA editing in SH-SY5Y cells we analysed additional ADAR substrates (GRIA2, FLNB, PDE8A, GRIK2 and GABRA3). Interestingly, IFN treatment altered the relative proportion of the RNA editing isoforms for all three targets studied (
Compound specific RNA editing profiles have been obtained by NGS-based analysis of GABRA3, GRIA2, GRIK2 and HTR2C targets (see
A positive value (%) indicates an increase in RNA editing at the specific site that is induced by the compound compared to the vehicle treated cells. Oppositely, a negative value (%) indicates a decrease in RNA editing at the specific site as a result of treatment with the compound compared to the vehicle treated cells.
The RNA editing profiles has been obtained for two compounds with low or no risk to induce a particular effect in a patient (see
The RNA editing profiles has been obtained for two compounds with high risk to induce a particular effect in a patient like Reserpine (see
Time course analysis of RNA editing changes has been observed by Aripiprazole, Interferon (IFN) and Reserpine on HTR2C (see
The most preferred algorithm was applied to determine the risk score of each compound at each studied time point (prob(Algorithm)). While for Interferon and Reserpine risk scores were high at all time points, Aripiprazole treatment was identified positively at risk starting from 24 hours and beyond (see Table 14 below).
Dose-dependent alterations of RNA editing profiles have been obtained after treatment of SH-SY5Y cells with three different compounds, Clozapine, Sertraline and Ketamine (see
The RNA editing profiles represent the respective relative proportion of HTR2C RNA editing as compared to vehicle-treated SH-SY5Y cells.
4712678.
Increased serotonin 2C receptor niRNA editing: a possible risk factor for suicide. Molecular psychiatry. 2008 November; 13(11):1001-10. PubMed PMID: 17848916.
| Number | Date | Country | Kind |
|---|---|---|---|
| 16000600 | Mar 2016 | EP | regional |
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/IB2017/000417 | 3/13/2017 | WO | 00 |
