WO2023198176A1

WO2023198176A1 - Prediction of the treatment response to bumetanide in subject with autism spectrum disorder

Info

Publication number: WO2023198176A1
Application number: PCT/CN2023/088307
Authority: WO
Inventors: Fei Li
Original assignee: Xinhua Hospital Affiliated To Shanghai Jiaotong University School Of Medicine
Priority date: 2022-04-15
Filing date: 2023-04-14
Publication date: 2023-10-19

Abstract

Provided is a method of predicting a response of a subject afflicted with an autism spectrum disorder (ASD) to bumetanide based on baseline expression levels of a set of cytokines in the subject, baseline behavioral performance of the subject and clinical information of the subject and the use thereof. Also provided are a prediction device and a computer readable medium for performing the prediction method.

Description

PREDICTION OF THE TREATMENT RESPONSE TO BUMETANIDE IN SUBJECT WITH AUTISM SPECTRUM DISORDER

CROSS-REFERENCE TO RELATED APPLICATIONS

The present application claims priority to PCT/CN2022/087139 entitled “PREDICTION OF THE TREATMENT RESPONSE TO BUMETANIDE IN SUBJECT WITH AUTISM SPECTRUM DISORDER” filed on April 15, 2022, which is incorporated herein by reference in its entirety.

FIELD

The present invention relates to treatment of autism spectrum disorder (ASD) , specifically, biomarkers for predicting the response of a subject with autism spectrum disorder (ASD) to bumetanide.

BACKGROUND

Autism spectrum disorder (ASD) affects about 1%children around the world¹ and can cause lifelong disability and elevate premature mortality². Currently, no medication that can cure ASD or all of its core symptoms is available¹. The recent success of repurposing drugs for novel treatments in psychiatry has been highlighted³, with one of the examples given being the use of bumetanide to improve the core symptoms in ASD^4-8. The most frequent adverse events were hypokalemia, increased urine elimination, loss of appetite, dehydration and asthenia. The heterogeneity in the treatment effect of bumetanide among ASD patients was significant, ranging from 37.3%to 47.62%in the randomized clinical trials (RCTs) in China^5, 8 and 51.80%⁶ or from 26.3%to 45.2%⁷ in the RCTs in France. Besides, there were also studies reported nonsignificant treatment effect of bumetanide for patients with ASD ⁹. Understanding this heterogeneity is essential for its clinical applicability and requires further investigation of its underlying mechanism of action to achieve precision medicine for ASD children.

The use of bumetanide as a potential drug to improve symptoms in ASD is based on a hypothesized pathoetiology of ASD, namely the delayed developmental switch of the gamma-aminobutyric acid (GABA) functioning from excitatory to inhibitory ^10-12. In the valproate and fragile X rodent models of autism, this GABA-switch can be facilitated by the reduction of intracellular chloride concentration which is mediated by a sequential expression of the main chloride transporters, such as the potassium (K) -Cl co-transporters 2 (KCC2) and the importer Na-K-Cl cotransporter 1 (NKCC1) ¹². Therefore, bumetanide as an NKCC1 inhibitor has been tested for its ability to restore GABA function in ASD ^{5-7, 13, 14}. However, these transporters can also be influenced by other molecules, such as cytokines, which are a number of small cell-signaling proteins closely interacting with each other to modulate the immune reactions. The cytokines have been implicated not only in brain development ¹⁵, but also in GABAergic transmission^16-18. It has been reported that the interferon (IFN) -γ can decrease the levels of NKCC1 and the α-subunit of Na⁺-K⁺-ATPase, contributing to the restore of inhibitory GABA function ¹⁶. In mice subjected to maternal deprivation, the interleukin (IL) -1 has also been found to reduce the expression of KCC2, delaying the developmental switch of the GABA function and thereby possibly contributing to the pathophysiology of developmental disorders such as ASD^17, 18. Therefore, a question naturally arises that whether the treatment effect of bumetanide for ASD can be affected by the immune responses in the patients.

Indeed, compared with healthy controls, changes of the cytokine levels have already been reported in patients with ASD ^19-22. Recent meta-analyses showed that the levels of anti-inflammatory cytokines IL-10 and IL-1 receptor antagonist (Ra) were decreased ²⁰, while proinflammatory cytokines IL-1β, IL-6 and anti-inflammatory cytokines IL-4, IL-13 were elevated in blood of patients with ASD ²¹. The levels of IFN-γ, IL-6, tumor necrosis factor (TNF) -α, granulocyte-macrophage colony-stimulating factor (GM-CSF) and IL-8 were observed to be elevated ²² in postmortem brain tissues of ASD patients, and increased level of IFN-γ, monocyte chemotactic protein (MCP) -1, IL-8, leukemia inhibitory factor (LIF) and interferon-gamma inducible protein (IP) -10 were found in another study ²³. These widely spread changes suggest that the cytokine signaling in ASD may be better characterized by multivariate patterns of cytokines. In literatures, many associations had been reported between the levels of cytokines (e.g., MCP-1, IL-1β, IL-4, IL-6, etc. ) and both core symptoms and adaptive functions in children with ASD ^24-26. Therefore, it has been suggested that cytokines may be used as biomarkers to identify different subsets within ASD. In each of these subsets the patients with ASD may share a commonly immune-related pathoetiology and therefore may have similar profiles of response to treatment ²⁷ . Based on these previous findings, we analyzed data acquired through the Shanghai Xinhua ASD registry, China, that began in 2016 to test the hypothesis that the immune activity of patients might help to identify the best responders to bumetanide in ASD.

SUMMARY

An aspect of the present invention provides a method of predicting a response of a subject afflicted with an autism spectrum disorder (ASD) to bumetanide, the method comprising:

obtaining characteristic information of the subject, wherein the characteristic information includes: (i) baseline expression levels of a set of cytokines in the subject; (ii) baseline behavioral performance of the subject; and (iii) clinical information of the subject, including gender and age;

predicting the response of the subject to bumetanide based on the characteristic information.

In some embodiments, the set of cytokines include three or more cytokines selected from the group consisting of IL1β, IL6, IL8, IFNγ, TNFα, MCP1, Eotaxin, IL17, IL4, IL2Rα, MIG, MIP1β, IFNα2, SDF1α, IL16, LIF, TNFβ, MIF, RANTES, IL18, PDGFβ, IP10, IL13, MIP1α, GCSF, GROα, HGF, IL1α, SCF, TRAIL, MCSF, CTACK, IL7, IL9 and SCGFβ.

In some embodiments, the set of cytokines include a group selected from the following:

(i) IL16, GROα and IL7;

(ii) IL16, GROα and TNFβ;

(iii) IL16, GROα, IL7, TNFβ and CTACK; and

(iv) IL16, GROα, IL7, TNFβ, LIF and MIF.

In some embodiments, the set of cytokines include IL1β, IL6, IL8, IFNγ, TNFα, MCP1, Eotaxin, IL17, IL4, IL2Rα, MIG, MIP1β, IFNα2, SDF1α, IL16, LIF, TNFβ, MIF, RANTES, IL18, PDGFβ, IP10, IL13, MIP1α, GCSF, GROα, HGF, IL1α, SCF, TRAIL, MCSF, CTACK, IL7, IL9 and SCGFβ.

In some embodiments, the behavioral performance includes one or more scores of screening tools or diagnostic tools for ASD.

In some embodiments, the behavioral performance includes 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12 or 13 scores selected from the group consisting of CAR_total, CAR_S, CAR_N, CAR_D, ADOS_S, ADOS_C, ADOS_P, ADOS_I, SRS_total, SRS_AWA, SRS_COG, SRS_COM, SRS_MOT and SRS_MANN.

In some embodiments, the behavioral performance includes a group of scores selected from the following:

(i) ADOS_S, ADOS_C, ADOS_P, SRS_COM and CARS_D;

(ii) ADOS_S, ADOS_C, ADOS_P and CARS_total;

(iii) ADOS_S, ADOS_C, ADOS_P, SRS_COM, CARS_D, CARS_total and SRS_MOT;

(iv) ADOS_S, ADOS_C, ADOS_P, SRS_COM, CARS_D and CARS_total; and

(v) ADOS_S, ADOS_C, ADOS_P and CARS_total.

In some embodiments, the method comprises predicting the response of the subject to bumetanide based on the characteristic information by using a classifier.

In some embodiments, the classifier is selected from the group consisting of Oblique Random Forest (ORF) , Partial Least Squares (PLS) , sparse Linear Discriminant Analysis (sLDA) , Neural Networks (NN) and Support Vector Machine (SVM) .

In some embodiments, the classifier has been trained.

In some embodiments, the characteristic information includes: (i) baseline expression levels of IL16, GROα and IL7; (ii) scores of ADOS_S, ADOS_C, ADOS_P, SRS_COM and CARS_D; (iii) gender and age; and the classifier is support vector machine;

the characteristic information includes: (i) baseline expression levels of IL16, GROαand TNFβ; (ii) scores of ADOS_S, ADOS_C, ADOS_P and CARS_total; (iii) gender and age; and the classifier is partial Least Squares;

the characteristic information includes: (i) baseline expression levels of IL16, GROα, IL7, TNFβ and CTACK; (ii) scores of ADOS_S, ADOS_C, ADOS_P, SRS_COM, CARS_D, CARS_total and SRS_MOT; (iii) gender and age; and the classifier is Neural Networks;

the characteristic information includes: (i) baseline expression levels of IL16, GROα, IL7, TNFβ, LIF and MIF; (ii) scores of ADOS_S, ADOS_C, ADOS_P, SRS_COM, CARS_D and CARS_total; (iii) gender and age; and the classifier is sparse Linear Discriminant Analysis; or

the characteristic information includes: (i) baseline expression levels of IL16, GROαand TNFβ; (ii) scores of ADOS_S, ADOS_C, ADOS_P and CARS_total; (iii) gender and age; and the classifier is Oblique Random Forest.

In some embodiments, the method comprises obtaining characteristic information of the subject by measuring the expression levels of the cytokines in a sample from the subject.

In some embodiments, the sample is plasma.

Another aspect of the present invention provides a method of predicting a response of a subject afflicted with an autism spectrum disorder (ASD) to bumetanide, the method comprising:

training a classifier with a training data set containing characteristic information of a plurality of individuals with ASD and the response of the individuals to bumetanide that has been determined select a subset of relevant characteristic information, wherein the characteristic information includes: (i) baseline expression levels of cytokines of IL1β, IL6, IL8, IFNγ, TNFα, MCP1, Eotaxin, IL17, IL4, IL2Rα, MIG, MIP1β, IFNα2, SDF1α, IL16, LIF, TNFβ, MIF, RANTES, IL18, PDGFβ, IP10, IL13, MIP1α, GCSF, GROα, HGF, IL1α, SCF, TRAIL, MCSF, CTACK, IL7, IL9 and SCGFβ; (ii) baseline scores of CAR_total, CAR_S, CAR_N, CAR_D, ADOS_S, ADOS_C, ADOS_P, ADOS_I, SRS_total, SRS_AWA, SRS_COG, SRS_COM, SRS_MOT and SRS_MANN; and (iii) gender and age;

predicting the response of the subject to bumetanide based on the subset of relevant characteristic information of the subject.

In some embodiments, the expression levels of the cytokines are obtained by measuring expression level of the cytokines in a sample from the subject.

In some embodiments, the sample is plasma.

Another aspect of the present invention provides a method of treating autism spectrum disorder (ASD) in a subject, the method comprising:

predicting a response of a subject afflicted with ASD to bumetanide using the any one of the methods as described above;

administering an effective amount of bumetanide to the subject that is identified to have response to bumetanide.

Another aspect of the present invention provides a prediction device comprising:

an input module which is configured to receive characteristic information of a subject with autism spectrum disorder (ASD) , wherein the characteristic information includes: (i) baseline expression levels of a set of cytokines in the subject; (ii) baseline behavioral performance of the subject; and (iii) clinical information of the subject, including gender and age;

a classifying module which is configured to comprise a classifier, wherein the classifier can predict the response of the subject to bumetanide using a classifier based on the characteristic information.

In some embodiments, the prediction device further comprises a training module which is configured to train the classifier using a training data set.

Another aspect of the present invention provides a computer readable medium comprising computer executable instructions recorded thereon for performing the operation comprising:

receiving characteristic information of a subject with autism spectrum disorder (ASD) , wherein the characteristic information includes: (i) baseline expression levels of a set of cytokines in the subject; (ii) baseline behavioral performance of the subject; and (iii) clinical information of the subject, including gender and age;

predicting the response of the subject to bumetanide using a classifier algorithm based on the characteristic information.

In some embodiments, the operation further comprises training the classifier using a training data set.

(i) IL16, GROα and IL7;

(ii) IL16, GROα and TNFβ;

(iii) IL16, GROα, IL7, TNFβ and CTACK; and

(iv) IL16, GROα, IL7, TNFβ, LIF and MIF.

(i) ADOS_S, ADOS_C, ADOS_P, SRS_COM and CARS_D;

(ii) ADOS_S, ADOS_C, ADOS_P and CARS_total;

(iii) ADOS_S, ADOS_C, ADOS_P, SRS_COM, CARS_D, CARS_total and SRS_MOT;

(iv) ADOS_S, ADOS_C, ADOS_P, SRS_COM, CARS_D and CARS_total; and

(v) ADOS_S, ADOS_C, ADOS_P and CARS_total.

In some embodiments, the classifier has been trained.

Another aspect of the present invention provides a kit for predicting a response to bumetanide in a subject with autism spectrum disorder (ASD) , the kit comprising agents for measuring expression levels of any one of the set of cytokines as described above and instructions describing any one of the predicting methods as described above.

Another aspect of the present invention provides a kit for treating autism spectrum disorder (ASD) in a subject, the kit comprising:

agents for measuring expression levels of any one of the set of cytokines as described above;

bumetanide; and

instructions describing any one of the predicting methods as described above and that an effective amount of bumetanide is administered to the subject that is identified to have response to bumetanide.

Another aspect of the present invention provides the characteristic information as described above for use in predicting a response to bumetanide in a subject with autism spectrum disorder (ASD) .

Another aspect of the present invention provides use of the characteristic information as described above in the preparation of a set of features for predicting the response to bumetanide in a subject with autism spectrum disorder (ASD) .

Another aspect of the present invention provides use of agents for measuring the expression levels of any one of the set of cytokines as described above in preparation of an agent or a kit for predicting a response to bumetanide in a subject with autism spectrum disorder (ASD) , wherein the prediction is performed through any one of the prediction methods as described above.

Another aspect of the present invention provides use of the characteristic information as described above in built a prediction model for predicting the response to bumetanide in a subject with autism spectrum disorder (ASD) .

BRIEF DESCRIPTION OF THE DRAWINGS

Figure 1. Sparse canonical correlation analysis. We carried out sparse canonical correlation analysis in [A] Discovery Set and [B] Validation Set. The canonical scores between CARS and cytokines, which were min-max normalized and log transformed, were significant related in both data sets.

Figure 2. Differences in three immuno-behavioural groups. K-means cluster plot on the immuno-behavioural plane. K-means cluster analysis was carried out in Discovery Set [A]and the patients from the Validation Set were mapped to this immuno-behavioural plane [B]. [C] Radar chart for the ratios of changes to baseline of CARS and cytokine levels in 3 immuno-behavioural groups. [D] Boxplot for the significant changes of CARS and cytokine levels in 3 immuno-behavioural groups; wherein the boxes from left to right are of best-responding group, least-responding group, and medium-responding group, respectively.

Figure 3. ROC curve for prediction for the immuno-behaviourally defined responding group. The classifiers included the Oblique Random Forest (ORF) model, Partial Least Squares (PLS) model, Support Vector Machine (SVM) model, sparse Linear Discriminant Analysis (sLDA) model and Neural Networks (NN) model. Based on the the immuno-behavioural covariation plane, the models were trained to predict the response to bumetanide for the children with ASD. As described in the main text, the models were trained using the Discovery Set, and tested using the Validation Set. The performances of the classification accuracy in the testing data set were reported in this figure. [A] Models with the cytokine levels at the baseline for predicting patients with ASD in the best responding group. [B]Models without the cytokine levels at the baseline for predicting patients with ASD in the best responding group. [C] Models with the cytokine levels at the baseline for predicting patients with ASD in the least responding group. [D] Models without the cytokine levels at the baseline for predicting patients with ASD in the least responding group.

Figure 4. Adjustment for batch effects. PCA plots show the impact of ‘ComBat’ algorithm adjustment on batch effect in two batches. Both before and after ComBat adjustment, scatter plots of the first two principal components are displayed (PC1 vs PC2) . In these plots, individual patient samples are represented by dots and color-coded according to their batch of origin. Post-ComBat adjustment the batches demonstrated more homogeneity, as evidenced by increased overlap in PCA scatter plots.

Figure 5. Pairwise association between CARS total score and cytokine. Partial correlation heatmaps are showed for pairwise association between [A] the baseline CARS total score and the baseline cytokine levels, [B] the baseline CARS total score and the changes of cytokine levels, [C] the change of CARS total score and the baseline cytokine levels, and [D] the change of CARS total score and the changes of cytokine levels. The associations were carried out FDR adjustment.

Figure 6. The scree plot for selection of the optimal number of clusters. We generated different clusters, with the number of clusters (k) ranging from 2 to 14. To validate clustering outcomes, we used internal cluster quality measures. For each of the possible number of clusters, an index is calculated reflecting the between-subject similarity within clusters and the dissimilarity between clusters. This index usually increases monotonically with increasing number of clusters, and the optimal value is determined to be at the elbow of its plot, where the change in index (difference with k-1 and k+1) is at a maximum. There are several possible indices available. We used the cluster number most of them chose. The following plots show the Hubert statistic for k of 2 to 14 possible clusters and the change (delta) in index compared with k-1. The results from the internal cluster quality indexes produce a good signal at 3 clusters.

Figure 7. Boxplot for the changes of CARS and cytokine levels in 3 immuno-behavioural groups. The boxes from left to right are of best-responding group, least-responding group, and medium-responding group, respectively.

Figure 8. ROC curve for the prediction of treatment response defined by CARS. The classifiers included the Oblique Random Forest (ORF) model, Partial Least Squares (PLS) model, Support Vector Machine (SVM) model, sparse Linear Discriminant Analysis (sLDA) model and Neural Networks (NN) model. Based on the behavioural assessments at the baseline before treatment, the models were trained to predict the response to bumetanide for the children with ASD. The responders were identified as their CARS total score decreased greater than 2.5 points (left) or 2 points (right) after a 3-month treatment of bumetanide. As described in the example, the models were trained using the Discovery Set, and tested using the Validation Set. The performances of the classification accuracy in the test data set were reported in this figure.

Figure 9. ROC curve for the prediction of best treatment response defined by CARS. The five classifiers (from left to right) included the Support Vector Machine (SVM) model, Partial Least Squares (PLS) model, Neural Networks (NN) model, sparse Linear Discriminant Analysis (sLDA) model, and Oblique Random Forest (ORF) model.

DETAILED DESCRIPTION

Bumetanide, a drug being studied in autism spectrum disorder (ASD) may act to restore gamma-aminobutyric acid (GABA) function, which may be modulated by the immune system. However, the interaction between bumetanide and the immune system remains unclear. Seventy-nine children with ASD were analyzed from a longitudinal sample for a 3-month treatment of bumetanide. The covariation between symptom improvements and cytokine changes was calculated and validated by sparse canonical correlation analysis. Response patterns to bumetanide were revealed by clustering analysis. Five classifiers were used to test whether including the baseline information of cytokines could improve the prediction of the response patterns using an independent test sample. An immuno-behavioural covariation was identified between symptom improvements in the Childhood Autism Rating Scale (CARS) and the cytokine changes among interferon (IFN) -γ, monokine induced by gamma interferon and IFN-α2. Using this covariation, 3 groups with distinct response patterns to bumetanide were detected, including the best (21.5%, n=17; Hedge’s g of improvement in CARS=2.16) , the least (22.8%, n=18; g=1.02) and the medium (55.7%, n=44; g=1.42) responding groups. Including the cytokine levels significantly improved the prediction of the best responding group before treatment (the best area under the curve, AUC=0.832) compared with the model without the cytokine levels (95%confidence interval of the improvement in AUC was [0.287, 0.319] ) . Cytokine measurements can help in identifying possible responders to bumetanide in ASD children, suggesting that immune responses may interact with the mechanism of action of bumetanide to enhance the GABA function in ASD.

It should be understood that the specific methods and conditions described in embodiments of the present invention are for the purpose of describing specific embodiments only and are not meant to be limiting, and that any methods and conditions similar or equivalent to those described herein may be used in the practice or testing of the present invention. The explanations of the relevant theories or mechanisms in the present invention are intended only to aid in the understanding of the invention and should not be considered a limitation of the embodiments protected by the present invention.

Unless otherwise noted, terms used in the present invention have the meanings commonly understood in the art and may be understood by reference to standard textbooks, references, and literature known to those skilled in the art. All publications referred to herein are incorporated herein by reference in their entirety.

Terminology

It must be noted that as used herein and in the appended claims, the singular forms "a, " "an, " and "the" include plural referents unless the context clearly dictates otherwise. It is further noted that the claims may be drafted to exclude any optional element. As such, this statement is intended to serve as antecedent basis for use of such exclusive terminology as "solely, " "only" and the like in connection with the recitation of claim elements, or use of a "negative" limitation.

Unless otherwise stated, the term “comprise” , “include” , “contain” and variations of these terms, such as comprising, comprises and comprised, are not intended to exclude further members, components, integers or steps. These terms also encompass the meaning of “consist of” or “consisting of” . The term “consist of” or “consisting of” is a particular embodiment of the term “comprise” , wherein any other non-stated member, component, integer or step is excluded.

The term "about" refers to a range equal to the particular value plus or minus ten percent (+/-10%) .

The term “and/or” refers to any one, several or all of the elements connected by the term.

The term “expressed” or “expression” , as used herein, refers to the translation from the RNA molecule to give a protein, a polypeptide, or a portion thereof.

The term “biomarker” , as used herein, refers to a gene or protein or a combination of multiple genes or multiple proteins whose levels of expression or concentration in a sample are altered after treatment with a therapeutic agent (e.g., bumetanide) or are indicative of disease (e.g., ASD) responsiveness to a therapeutic agent (e.g., bumetanide) . The biomarkers disclosed herein are genes and/or proteins whose expression level or concentration or change of expression or concentration correlates with the responsiveness of ASD to bumetanide.

The terms “subject” and “patient” may be used interchangeably herein. The term “subject” , as used herein, refers to any organism to which the method of the present invention may be applied, e.g., for experimental, diagnostic, prophylactic, and/or therapeutic purposes. Typical subjects include animals (e.g., mammals such as mice, rats, rabbits, non-human primates such as chimpanzees and other apes and monkey species, and humans) . The subject may be a mammal, particularly a human, including a male or female, and including a neonatal, infant, juvenile, adolescent, adult or geriatric, and further is inclusive of various races and ethnicities. In some examples, a subject refers to an individual in need of diagnosis, treatment or prevention of a disease or condition, said subject may have said disease or condition, or be at risk of developing said disease or condition.

The term “sample” , as used herein, refers to any sampling of cells, tissues, or bodily fluids in which expression of a biomarker can be detected. Examples of such samples include, but are not limited to, biopsies, smears, blood, lymph, urine, saliva, or any other bodily secretion or derivative thereof. Blood can, for example, include whole blood, plasma, serum, or any derivative of blood. Samples can be obtained from a subject by a variety of techniques, which are known to those skilled in the art.

The term “treatment” , as used herein, refers to both therapeutic treatment and prophylactic or preventative measures, wherein the object is to prevent or slow down (lessen) the targeted pathologic condition or disorder. Those in need of treatment include those diagnosed with the disorder as well as those prone to have the disorder (e.g., a genetic predisposition) or those in whom the disorder is to be prevented. The terms “prevent, ” “preventing, ” and “prevention” refer to reducing the likelihood of the onset (or recurrence) of a disease, disorder, condition, or associated symptom (s) .

The term “response” or “responsiveness” , as used herein, refers to the effectiveness of a treatment or therapy in relieving the disease or alleviating the symptoms. A beneficial response can be assessed using any endpoint indicating a benefit to the subject, including, without limitation, (1) inhibition, to some extent, of disease progression, including slowing down and complete arrest; (2) amelioration (e.g., reduction in number, frequency and/or intensity) of one or more symptoms of the disease; (3) stabilization of the condition, e.g., prevention or delay of deterioration expected or typically observed to occur absent the treatment; etc. A beneficial response of a subject with ASD to a treatment may be characterized by improvements in one or more behaviors associated with ASD, such as the behaviors listed in the common screening tools or the diagnostic tools.

The term “classification” refers to a procedure and/or algorithm in which individual items are placed into groups or classes based on quantitative information on one or more characteristics inherent in the items (referred to as traits, variables, characters, features, etc. ) and based on a statistical model and/or a training set of previously labeled items.

The term “administering” with respect to the methods of the invention, means a method for therapeutically or prophylactically preventing, treating or ameliorating a syndrome, disorder or disease (e.g., ASD) as described herein. Such methods include administering an effective amount of therapeutic agent (e.g., bumetanide) during the course of a therapy. The ways of administration are to be understood as embracing all known suitable therapeutic treatment regimens.

The term “pharmaceutically acceptable salt” , as used herein, refers to a relatively nontoxic, inorganic or organic acid salt of a compound of the invention. These salts may be prepared in situ during the final isolation and purification of the compounds or by reacting the purified compound in its free form separately with a suitable organic or inorganic acid and isolating the salt thus formed. Representative acid salts include, but are not limited to, acetate, adipate, aspartate, benzoate, besylate, bicarbonate/carbonate, bisulphate/sulphate, borate, camsylate, citrate, cyclamate, edisylate, esylate, formate, fumarate, gluceptate, gluconate, glucuronate, hexafluorophosphate, hibenzate, hydrochloride/chloride, hydrobromide/bromide, hydroiodide/iodide, isethionate, lactate, malate, maleate, malonate, mesylate, methylsulphate, naphthylate, 2-napsylate, nicotinate, nitrate, orotate, oxalate, palmitate, pamoate, phosphate/hydrogen phosphate/dihydrogen phosphate, pyroglutamate, saccharate, stearate, succinate, tannate, tartrate, tosylate, trifluoroacetate and xinafoate salts. In one embodiment, the pharmaceutically acceptable salt is a hydrochloride/chloride salt.

The term “solvate” , as used herein, refers to a complex of variable stoichiometry formed by a solute (e.g., the active agent of the present invention) and a solvent. Such solvents for the purpose of the invention may not interfere with the biological activity of the solute. Examples of suitable solvents include, but are not limited to, water, methanol, ethanol and acetic acid.

The term “tautomer” , as used herein refers to two (or more) compounds that differ from each other only in the position of one (or more) mobile atoms and in electron distribution, for example, keto-enol and imine-enamine tautomers.

The term “effective amount” or “therapeutically effective amount” means that amount of active compound or pharmaceutical agent, a combination of therapeutic compounds or pharmaceutical compositions thereof provided herein, that elicits the biological or medicinal response in a tissue system, animal or human, that is being sought by a researcher, veterinarian, medical doctor, or other clinician, which includes preventing, treating or ameliorating a syndrome, disorder, or disease being treated, or the symptoms of a syndrome, disorder or disease being treated (e.g., ASD) .

The term “computer” , as used herein, includes at least one hardware processor that uses at least one memory. The at least one memory may store a set of instructions. The instructions may be either permanently or temporarily stored in the memory or memories of the computer. The processor executes the instructions that are stored in the memory or memories in order to process data. The set of instructions may include various instructions that perform a particular task or tasks, such as those tasks described herein.

Prediction of the response of autism spectrum disorder (ASD) to bumetanide

Autism spectrum disorders (ASD) are clinically diagnosed disorders with both single and complex multi-gene etiology and can be diagnosed according to Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition, DSM-5. Autism spectrum disorders, which are characterized by varying degrees of impairment in communication skills, social interactions, and restricted, repetitive and stereotyped patterns of behavior, include autism, PDD-NOS (pervasive developmental disorder not otherwise specified) , Asperger syndrome, Rett syndrome and childhood disintegrative disorder (CDD) . The ASD as described in the present invention may also refer to any of the specific disorders or syndromes mentioned above. In some embodiments, the ASD may be an autism spectrum disorders in child.

ASD can be screened and diagnosed using a variety of tools, such as various scales or subscales designed for screen or diagnosis of ASD. Some common screening tools that can help with screening for and diagnosing of ASD include the Checklist of Autism in Toddlers (CHAT) , the modified Checklist for Autism in Toddlers (M-CHAT) ₅ the Screening Tool for Autism in Two-Year-Olds (STAT) , the Social Communication Questionnaire (SCQ) (for children 4 years of age and older) , the Autism Spectrum Screening Questionnaire (ASSQ) , the Australian Scale for Asperger's Syndrome, and the Childhood Asperger Syndrome Test (CAST) .

Typical diagnostic tools include, but are not limited to, the Autism Diagnosis Interview-Revised (ADI-R) , the Autism Diagnostic Observation Schedule (ADOS) , the Childhood Autism Rating Scale (CARS) , Autism Behavior Checklist (ABC) and Social Responsiveness Scale (SRS) . Behavioral performance can be tested using these tools and a person skilled in the art knows how to perform the behavioral test and obtain the score of these tools, for example, the scores for each item in a scale or subscale may be added together to obtain the score for that scale or subscale.

CARS consisted of 15 items rated on a 7-point scale from one to four; higher scores are associated with a higher level of impairment (Schopler E, Reichler R, DeVellis R. Toward objective classification of childhood autism: Childhood autism rating scale (CARS) Journal of Autism and Developmental Disorders. 1980; 10: 91–103; Schopler E, Reichler R, Rochen Renner B. The childhood autism rating scale. Western Psychological Services; 1988) . Total scores can range from a low of 15 to a high of 60; scores below 30 indicate that the individual is in the non-autistic range, scores between 30 and 36.5 indicate mild to moderate autism, and scores from 37 to 60 indicate severe autism. The items in CARS can be further categorized into three subscales: Social impairment (SI) , Negative emotionality (NE) , and Distorted sensory response (DSR) (DiLalla DL, Rogers SJ. Domains of the childhood autism rating scale: Relevance for diagnosis and treatment. Journal of Autism and Developmental Disorders 1994; 24 (2) : 115-128) . In the present invention, the CARS scores of the total items, and three subscales of SI, NE and DSR are designated as CAR_total, CAR_S, CAR_N and CAR_D, respectively, as shown in Table 1.

ADOS (also called Autism Diagnostic Observation Schedule-Generic, ADOS-G) is a semi-structured assessment of social interaction, communication, play, and imaginative use of materials for individuals who may have autism or an autism spectrum disorder (ASD) . The ADOS consists of four “modules” . Module 1 is used for children who are preverbal or have single-word language. Module 2 is appropriate for individuals with phrase speech abilities. Module 3 is used for children and adolescents who are verbally fluent. Verbally fluent adolescents and adults are assessed with Module 4. (Lord, et al., The autism diagnostic observation schedule-generic: a standard measure of social and communication deficits associated with the spectrum of autism, Journal of Autism and Developmental Disorders, 2000, 30 (3) : 205-223; Lord, C., Rutter, M., DiLavore, P.C., & Risi, S. (2008) . Autism Diagnostic Observation Schedule Manual. Los Angeles: Western Psychological Services. ) In the present invention, the ADOS scores of the subscales of social interaction, communication, play, and imaginative use of materials are designated as ADOS_S, ADOS_C, ADOS_P and ADOS_I, respectively, as shown in Table 1. Higher scores represent greater autism symptom severity.

SRS is a 65-item questionnaire and is a standardized measure of the core symptoms of autism. Each item is scored on a 4-point Likert scale. The score of each individual item is summed to create a total raw score. Total score of 0-62 is within normal limits, total score of 63-79 indicates mild range of impairment, total score of 80-108 indicates moderate range of impairment, and total score of 109-149 indicates severe range of impairment. Five subscales are also provided: Social Awareness (AWA) , Social cognition (COG) , Social Communication (COM) , Social Motivation (MOT) , and Autistic Mannerism (MANN) . (Constantino JN, Gruber CP. Social responsiveness scale: SRS-2. Western Psychological Services Torrance, CA, 2012. ) In the present invention, the SRS scores of the total items and the subscales of AWA, COG, COM, MOT and MANN are designated as SRS_total, SRS_AWA, SRS_COG, SRS_COM, SRS_MOT and SRS_MANN, respectively, as shown in Table 1.

ADI-R is a standardized, semi-structured clinician led parent interview (Lord, et al., "Autism Diagnostic Interview-Revised: a revised version of a diagnostic interview for caregivers of individuals with possible pervasive developmental disorders, " J Autism Dev Disord, 1994, 24 (5) : 659-685) . The ADI-R includes 93 items focusing on Early Development, Language/Communication, Reciprocal Social Interactions, and Restricted, Repetitive Behaviors and Interests. Higher scores represent greater autism symptom severity.

ABC is a symptoms rating checklist used to assess and classify problem behaviors of children and adults in a variety of settings (Aman et al., Psychometric characteristics of the aberrant behavior checklist. Am J Ment Deflc. 1985 Mar; 89 (5) : 492-502. ) . The ABC includes 58 items that resolve into five subscales: (1) irritability, (2) lethargy/social withdrawal, (3) stereotypic behavior, (4) hyperactivity/noncompliance, and (5) inappropriate speech. Higher scores represent greater autism symptom severity.

Bumetanide has been reported to improve the core symptoms of ASD, but only a proportion of patients with ASD can benefit from bumetanide treatment. The present inventors find that cytokines can be used to evaluate and predict a response of a subject to bumetanide, i.e., evaluate and predict a therapeutic effect of bumetanide in subject with ASD. The age of the subject may be in the range of about 1 to about 45 years old, about 2 to about 40 years old, about 3 to about 30 years old, about 3 to about 20 years old, about 3 to about 12 years old, about 3 to about 10 years old. The subject may be male or female. In some embodiments, the subject may be a child with ASD. In some embodiments, the child is 3-12 years old. In some embodiments, the child is 3-10 years old.

The term “bumetanide” , as used herein, refers to 3-butylamino-4-phenoxy-5- sulfamoylbenzoic acid, or a pharmaceutically acceptable salt, solvate or tautomer thereof.

According to the present invention, characteristic information of a subject with ASD can be used to predict the response of said subject to bumetanide treatment. The characteristic information may comprise a variety of features of a subject, and can be interpreted as a set of multiple features of a subject. The characteristic information may include: (i) baseline expression levels of a set of cytokines in the subject; (ii) baseline behavioral performance of the subject; and (iii) clinical information of the subject, including gender, age, and BMI, etc.

The term “baseline” , when used in conjunction with characteristic information, refers to the characteristic information of the subject before the subject is treated with bumetanide. The baseline characteristic information can be measured or assessed at any appropriate time prior to treatment of subject with bumetanide, for example, half a day to one week prior to treatment of subject with bumetanide. It should be understood that different characteristic information may be acquired at different time points in the appropriate time period. The baseline expression level refers to the expression level of a cytokine of the subject before the subject is treated with bumetanide. The baseline behavioral performance refers to the behavioral performance of the subject before the subject is treated with bumetanide.

The set of cytokines may include 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48 or more cytokines. In some embodiments, the set of cytokines may be selected from the cytokines listed in Table 1, i.e., IL1β, IL6, IL8, IFNγ, TNFα, MCP1, , Eotaxin, IL17, IL4, IL2Rα, MIG, MIP1β, IFNα2, SDF1α, IL16, LIF, TNFβ, MIF, RANTES, IL18, PDGF-BB, IP10, IL13, MIP1α, GCSF, GROα, HGF, IL1α, SCF, TRAIL, MCSF, CTACK, IL7, IL9 and SCGFβ. In some embodiments, the set of cytokines may include any 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34 or all of 35 cytokines selected from those listed in Table 1.

It should be understood that the cytokines listed in Table 1 are merely illustrative and are not intended to limit the scope of the present invention. The present inventors have found that baseline expression levels of cytokines correlate with response to bumetanide in subjects with ASD, and the cytokines described herein can include any type and any number of cytokines and are not limited to the cytokines listed in Table 1. The skilled person in the art can identify specific cytokines for predicting response to bumetanide in subjects with ASD by different algorithms or using different prediction models, such as those described below. The set of cytokines may include other cytokines that are not listed in Table 1, that is, the set of cytokines may include may include 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47, 48 or more cytokines, some of which are cytokines selected from the cytokines listed in Table 1 and others are cytokines not listed in Table 1. In some embodiments, the set of cytokines may include 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45, 46, 47 or 48 cytokines selected from Basic FGF (bFGF) , β-NGF, CTACK, Eotaxin, G-CSF, GM-CSF, GRO-α, HGF, IFN-α2, IFN-γ, IL-10, IL-12p40, IL-12p70, IL-13, IL-15, IL-16, IL-17, IL-18, IL-1α, IL-1β, IL-1Ra, IL-2, IL-2Rα, IL-3, IL-4, IL-5, IL-6, IL-7, IL-8, IL-9, IP-10, LIF, M-CSF, MCP-1, MCP-3, MIF, MIG, MIP-1α, MIP-1β, PDGF-BB, RANTES, SCF, SCGF-β, SDF-1α, TNF-α, TNF-β, TRAIL and VEGF

In some embodiments, the set of cytokines includes a group selected from the following:

(i) IL16, GROα;

(ii) IL16, GROα and IL7;

(iii) IL16, GROα and TNFβ;

(iv) IL16, GROα, IL7, TNFβ and CTACK;

(v) IL16, GROα, IL7, TNFβ, LIF and MIF.

The expression level of a cytokine can be obtained by measuring the concentration of the cytokine in a sample derived from the subject. In some embodiments, the sample may be blood (e.g., peripheral blood) , plasma or serum.

The expression level of a cytokine can be measured by a variety of methods known in the art. These include, for example, an immunoassay, a radioimmunoassay (RIA) , an immunoradiometric assay (IRMA) , an enzyme-linked immunosorbent assay (ELISA) , western blot analysis, an ELISpot, CELISA (cellular enzyme-linked immunosorbent assay) , RHPA (reverse hemolytic plaque assay) , a kinase receptor activation assay (KIRA) , a cytokine immunotrapping assay (CITA) , or a radioreceptor assay (RRA) . Such determination may be performed in a multiplex or matrix -based format such as a multiplexed immunoassay. In some embodiments, the level of a cytokine can be determined by a bead-based multiplex assay, such as Bio-Plex multiplex immunoassay system.

Behavioral performance can be assessed or further quantified by any method known in the art. Behavioral performance can be measured as scores of any above-mentioned screening tools or diagnostic tools, such as CARS, ADOS, SRS, ADI-R or ABC, or any subscale of them, or any combination of them and their subscales, as shown in Table 1.

In some embodiments, the behavioral performance includes CARS score, i.e., CAR_total, CAR_S, CAR_N, CAR_D or any combination thereof. In some embodiments, the behavioral performance includes CAR_total, CAR_S, CAR_N and CAR_D. In some embodiments, the behavioral performance includes any 1, 2, 3 or 4 selected from the group consisting of CAR_total, CAR_S, CAR_N and CAR_D.

In some embodiments, the behavioral performance includes ADOS score, i.e., ADOS_S, ADOS_C, ADOS_P, ADOS_I or any combination thereof. In some embodiments, the behavioral performance includes ADOS_S, ADOS_C, ADOS_P and ADOS_I. In some embodiments, the behavioral performance includes any 1, 2, 3 or 4 selected from the group consisting of ADOS_S, ADOS_C, ADOS_P and ADOS_I.

In some embodiments, the behavioral performance includes SRS score, i.e., SRS_total, SRS_AWA, SRS_COG, SRS_COM, SRS_MOT, SRS_MANN or any combination thereof. In some embodiments, the behavioral performance includes SRS_total, SRS_AWA, SRS_COG, SRS_COM, SRS_MOT and SRS_MANN. In some embodiments, the behavioral performance includes any 1, 2, 3, 4 or 5 selected from the group consisting of SRS_total, SRS_AWA, SRS_COG, SRS_COM, SRS_MOT and SRS_MANN.

In some embodiments, the behavioral performance includes CAR_total, CAR_S, CAR_N, CAR_D, ADOS_S, ADOS_C, ADOS_P, ADOS_I, SRS_total, SRS_AWA, SRS_COG, SRS_COM, SRS_MOT and SRS_MANN. In some embodiments, the behavioral performance includes any 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12 or 13 selected from the group consisting of CAR_total, CAR_S, CAR_N, CAR_D, ADOS_S, ADOS_C, ADOS_P, ADOS_I, SRS_total, SRS_AWA, SRS_COG, SRS_COM, SRS_MOT and SRS_MANN.

In some embodiments, the behavioral performance includes a group selected from the following:

(i) ADOS_S, ADOS_C, ADOS_P, SRS_COM and CARS_D;

(ii) ADOS_S, ADOS_C, ADOS_P and CARS_total;

(iii) ADOS_S, ADOS_C, ADOS_P, SRS_COM, CARS_D, CARS_total and SRS_MOT;

(iv) ADOS_S, ADOS_C, ADOS_P, SRS_COM, CARS_D and CARS_total; and

(v) ADOS_S, ADOS_C, ADOS_P and CARS_total.

The clinical information of the subject includes gender, age, and BMI, etc. In some embodiments, the clinical information includes gender and age.

In some embodiments, the characteristic information used to predict the response of the subject to bumetanide treatment may include: (i) baseline expression levels of cytokines of IL1β, IL6, IL8, IFNγ, TNFα, MCP1, Eotaxin, IL17, IL4, IL2Rα, MIG, MIP1β, IFNα2, SDF1α, IL16, LIF, TNFβ, MIF, RANTES, IL18, PDGFβ, IP10, IL13, MIP1α, GCSF, GROα, HGF, IL1α, SCF, TRAIL, MCSF, CTACK, IL7, IL9 and SCGFβ; (ii) baseline scores of CAR_total, CAR_S, CAR_N, CAR_D, ADOS_S, ADOS_C, ADOS_P, ADOS_I, SRS_total, SRS_AWA, SRS_COG, SRS_COM, SRS_MOT and SRS_MANN; and (iii) gender and age

In some embodiments, the characteristic information used to predict the response of the subject to bumetanide treatment may include: (i) baseline expression levels of IL16, GROα and IL7; (ii) baseline scores of ADOS_S, ADOS_C, ADOS_P, SRS_COM and CARS_D; and (iii) gender and age.

In some embodiments, the characteristic information used to predict the response of the subject to bumetanide treatment may include: (i) baseline expression levels of IL16, GROα and TNFβ; (ii) baseline scores of ADOS_S, ADOS_C, ADOS_P and CARS_total; and (iii) gender and age.

In some embodiments, the characteristic information used to predict the response of the subject to bumetanide treatment may include: (i) baseline expression levels of IL16, GROα, IL7, TNFβ and CTACK; (ii) baseline scores of ADOS_S, ADOS_C, ADOS_P, SRS_COM, CARS_D, CARS_total and SRS_MOT; and (iii) gender and age.

In some embodiments, the characteristic information used to predict the response of the subject to bumetanide treatment may include: (i) baseline expression levels of IL16, GROα, IL7, TNFβ, LIF and MIF; (ii) baseline scores of ADOS_S, ADOS_C, ADOS_P, SRS_COM, CARS_D and CARS_total; and (iii) gender and age.

The response of the subject to bumetanide can be interpreted as the therapeutical effect of bumetanide in the subject. The response or the therapeutical effect may refer to, in the present invention, is the response or the therapeutic effect exhibited in the subject after a period of treatment with bumetanide. Said period of treatment is sufficient for bumetanide to demonstrate its therapeutic effect in the subject. The duration of treatment may be ranged from about one month to six months, e.g., two months to three months, two months to four months.

The response of the subject to bumetanide may be categorized to have response or no response, high response or low response, better response or least response, positive response or negative response, or to be responder or non-responder, etc. according to known criteria or a criterion determined through statistical analysis of a population of patients with ASD who are treated with bumetanide. Generally, having response, high response, better response, positive response or responder means better therapeutical effect.

In some embodiments, the response may be determined by the change of the scores of screening or diagnostic tools (such as the above-mentioned tools) after treatment with bumetanide relative to that before treatment. Typically, a decrease in the score (such as CARS score, ADOS score or SRS score) above a specific threshold means that there is improvement in ASD, and a decrease in the score (such as CARS score, ADOS score or SRS score) below a specific threshold or an increase or no change in the score means that there is no improvement in ASD. The threshold is known to a person skilled in the art for the common screening or diagnostic tools (such as the above-mentioned tools) . “Having response” , “high response” , “better response” , “positive response” or “responder” to bumetanide may refer to that the amount of the decrease of a screening or diagnostic tool score (for example, CARS score, e.g., CARS_total) after treatment with bumetanide relative to that before treatment is not less than (e.g., above or greater than) a specific threshold. The specific threshold for “having response” , “high response” , “better response” , “positive response” or “responder” may be about 2, 2.5 or 3. “No response” , “low response” , “least response” , “negative response” or “non-responder” to bumetanide may refer to that the amount of the decrease of a screening or diagnostic tool score (for example, CARS score, e.g., CARS_total) after treatment with bumetanide relative to that before treatment is below (or less than) a specific threshold or the screening or diagnostic tool score (for example, CARS score, e.g., CARS_total) after treatment with bumetanide is higher than or substantially equal to that before treatment, for example, that CARS_total after treatment with bumetanide is higher than or substantially equal to that before treatment, or is lower than that before treatment by a value less than or equal to approximately 1 or 2. The specific threshold for “no response” , “low response” , “least response” , “negative response” or “non-responder” may be 1 or 2.

In some embodiments, “having response” , “high response” , “better response” , “positive response” or “responder” to bumetanide may be defined as a therapeutical effect that is determined to be the better or the best in a population of patients with ASD who are treated with bumetanide using a statistical analysis. “No response” , “low response” , “least response” , “negative response” or “non-responder” to bumetanide may be defined as a therapeutical effect that is determined to be the least in a population of patients with ASD who are treated with bumetanide using a statistical analysis.

“Better” , “best” or “least” means that the therapeutical effect a portion of the population has a statistically significant difference from other portion of the population. The therapeutical effect can be quantified using the change of the score of any screening tool or diagnostic tool and higher score decrease generally means better therapeutical effect. A person skilled in the art has the knowledge of how to judge the therapeutical effect according a screening tool or diagnostic tool.

In some embodiments, the statistical analysis may be a clustering analysis, including but not limited to, centroid-based (e.g., k-means) clustering, hierarchical clustering (e.g., mean-shift or Agglomerative Hierarchy) , distribution-based clustering, density-based (e.g., density-based spatial clustering of applications with noise, DBSCAN) clustering, grid-based clustering. “Having response” , “high response” , “better response” , “positive response” or “responder” to bumetanide may be defined as the response of the cluster that is determined to has the best therapeutical effect using a clustering analysis. “No response” , “low response” , “least response” , “negative response” or “non-responder” to bumetanide may be defined as the response of the cluster that is determined to has the least therapeutical effect using a clustering analysis. In some embodiments, the clustering analysis is performed based on the change of a screening or diagnostic tool score (for example, CARS score, e.g., CARS_total) and the change of expression levels of cytokines (for example, MIG, IFN-α2, IFN-γ) of individuals in the population after treatment with bumetanide relative to those before treatment.

In some embodiments, “having response” , “high response” , “better response” , “positive response” or “responder” to bumetanide may be defined as a therapeutical effect that is better than the third quartile in a population of patients with ASD who are treated with bumetanide. “No response” , “low response” , “least response” , “negative response” or “non-responder” to bumetanide may be defined as a therapeutical effect that is worse than the first of the second quartile in a population of patients with ASD who are treated with bumetanide. The quartile can be determined in terms of therapeutical effect of bumetanide on the population of patients with ASD.

There is no mandatory requirement for the number of patients included in the population used to determine the criterion for the response to bumetanide, statistical results can be obtained. For example, the number of patients included in the population may be 50-200, e.g., 50-100.

The prediction of the response of a subject with ASD to bumetanide is essentially a classification method, and the prediction result is the classification result, i.e., subjects are classified as having different responses to bumetanide. The prediction of the response of a subject with ASD to bumetanide may be performed by using a prediction model, which may also be referred to as a classifier, that can be used to determine the response of a subject to bumetanide. A classifier can be a machine learning system and can characterize the response of ASD to bumetanide based on the characteristic information of a subject. In the present invention, use of the classifier means that the characteristic information of a subject may be used as input for a classifier, and the output is the response of the subject to bumetanide.

Typically, the model is built using characteristic information of subjects for which the classification (i.e., the response of the subject with ASD to bumetanide) has already been ascertained. Once the model (classifier) is built, it may be applied to characteristic information obtained from a subject with ASD in order to determine the response of the subject to bumetanide.

A variety of prediction models known in the art may be used as the classifier of the present invention to determine the response of the subject to bumetanide. For example, the classifier may comprise an algorithm selected from support vector machine (SVM) , partial least squares (PLS) , neural networks (NN) , sparse linear discriminant analysis (sLDA) , oblique random forest (ORF) , logistic regression, quadratic discriminant analysis (QDA) , Bayes, C4.5 decision tree, or k-nearest neighbor (KNN) .

The classifier may be a trained classifier. In some embodiments, the method of the present invention further includes the step of training the classifier before prediction. The classifier may have been trained using a training data set to choose an optimal algorithm for classification and build the prediction model. The training data set may comprise the characteristic information of a plurality of individual with ASD and the response of the individuals to bumetanide which have been determined, e.g., after said individuals are treated with bumetanide for a period of time (such as one month to six months, e.g., two months to three months) . The training data set may also comprise control individuals that have been identified as not having ASD or have been identified as have ASD but are not treated with bumetanide (e.g., are treated with placebo) .

The range of ages of a population in the training data set may be from about 1 years old to about 45 years old, about 2 years old to about 40 years old, about 3 years old to about 30 years old, about 3 years old to about 20 years old, about 3 years old to about 12 years old, about 3 years old to about 10 years old. The median age of a population in the training data set may be about 3 years old, 4 years old, 5 years old, 6 years old, 7 years old, 8 years old, 9 years old, or 10 years old, or more. The population may consist of all males or all females, or may consist of males and females.

Optionally, once a model is built, the validity of the model can be validated using methods known in the art. In some embodiments, the method of the present invention further includes the step of validating the classifier after training and before prediction. One way to validate the validity of the model is to apply the model to an independent data set, such as a validation data set. The validation data set may comprise the characteristic information of a plurality of individuals with ASD and the response of the individuals to bumetanide which have been determined and the individuals of the validation data set are different from the individuals of the training data set. Another way to validate the model is by cross validation of the dataset. To perform cross-validation, one, or a subset, of the individuals is eliminated and the model is built, as described above, without the eliminated individual, forming a “cross-validation model. ” The eliminated individual is then predicted according to the model, as described herein. This process is done with all the individuals, or subsets, of the initial dataset and an error rate is determined. The accuracy the model is then assessed.

The range of ages of a population in the validation data set may be from about 1 years old to about 45 years old, about 2 years old to about 40 years old, about 3 years old to about 30 years old, about 3 years old to about 20 years old, about 3 years old to about 12 years old, about 3 years old to about 10 years old. The median age of a population in the validation data set may be about 3 years old, 4 years old, 5 years old, 6 years old, 7 years old, 8 years old, 9 years old, or 10 years old, or more. The population may consist of all males or all females, or may consist of males and females.

In some embodiments, the characteristic information of the individuals in the training data set the validation data set may independently include any one of the following groups:

(i) baseline expression levels of cytokines of IL1β, IL6, IL8, IFNγ, TNFα, MCP1, Eotaxin, IL17, IL4, IL2Rα, MIG, MIP1β, IFNα2, SDF1α, IL16, LIF, TNFβ, MIF, RANTES, IL18, PDGFβ, IP10, IL13, MIP1α, GCSF, GROα, HGF, IL1α, SCF, TRAIL, MCSF, CTACK, IL7, IL9 and SCGFβ; baseline scores of CAR_total, CAR_S, CAR_N, CAR_D, ADOS_S, ADOS_C, ADOS_P, ADOS_I, SRS_total, SRS_AWA, SRS_COG, SRS_COM, SRS_MOT and SRS_MANN; and gender and age;

(ii) baseline expression levels of IL16, GROα and IL7; baseline scores of ADOS_S, ADOS_C, ADOS_P, SRS_COM and CARS_D; and gender and age;

(iii) baseline expression levels of IL16, GROα and TNFβ; baseline scores of ADOS_S, ADOS_C, ADOS_P and CARS_total; and gender and age;

(iv) baseline expression levels of IL16, GROα, IL7, TNFβ and CTACK; baseline scores of ADOS_S, ADOS_C, ADOS_P, SRS_COM, CARS_D, CARS_total and SRS_MOT; and gender and age;

(v) baseline expression levels of IL16, GROα, IL7, TNFβ, LIF and MIF; baseline scores of ADOS_S, ADOS_C, ADOS_P, SRS_COM, CARS_D and CARS_total; and gender and age; or

(vi) baseline expression levels of IL16, GROα and TNFβ; baseline scores of ADOS_S, ADOS_C, ADOS_P and CARS_total; and gender and age. As will be appreciated by the skilled artisan, the strength of the model may be assessed by a variety of parameters including, but not limited to, the accuracy, and AUC. Methods for computing accuracy, AUC are known in the art and described herein (See, e.g., the Examples) . The predicting classifier may have an accuracy of at least 60%, at least 65%, at least 70%, at least 75%, at least 80%, at least 85%, at least 90%, at least 95%, or more. The predicting classifier may have an accuracy in a range of about 60%to 70%, 70%to 80%, 80%to 90%, or 90%to 100%. The predicting classifier may have an AUC of at least 0.60, at least 0.65, at least 0.70, at least 0.75, at least 0.80, at least 0.85, at least 0.80, at least 0.95, or more. The predicting classifier may have an AUC in a range of about 0.60 to 0.70, 0.70 to 0.80, 0.80 to 0.90, or 0.90 to 1.00.

In some embodiments, the characteristic information may contain a large number of features, and some of these features may have a higher correlation with, or contribute more to, the predicted outcome, and some may have a lower or no correlation with the predicted outcome. Thus, a subset of relevant feature information may be selected from the characteristic information by a feature selection technique to improve the computational efficiency and accuracy of the classifier, and then the features in said subset may then be used to make predictions.

Feature selection technique that commonly used in the art mainly includefilter techniques which assess the relevance of features by looking at the intrinsic properties of the data, wrapper methods which embed the model hypothesis within a feature subset search, and embedded techniques in which the search for an optimal set of features is built into a classifier algorithm, which is well known to a person skilled in the art.

In the present invention, the prediction model (classifier) may also be used to select the subset of relevant characteristic information.

In some embodiments, the classifier is trained with a training data set containing characteristic information of a plurality of individuals to select a subset of characteristic information features. The features in the subset are highly correlated with, or contribute more to, the prediction results. Subsequently prediction can be made by said classifier using the features in said subset as input. As known to those skilled in the art, the selection of the subset may be done automatically by the classifier.

A person skilled in the art can understand that the subset of relevant characteristic information selected using different classifiers may vary, depending on the algorithm used by the classifier. For a particular classifier, using the most appropriate subset for that classifier can result in a higher accuracy.

In some embodiments, the training data set used to select the subset may include the characteristic information of a plurality of individuals with ASD and the response of the individuals to bumetanide that has been determined, wherein the characteristic information includes: (i) baseline expression levels of a set of cytokines in the individual; (ii) baseline behavioral performance of the individual; and (iii) clinical information of the individual;

wherein, the set of cytokines comprise the cytokines listed in Table 1, i.e., IL1β, IL6, IL8, IFNγ, TNFα, MCP1, Eotaxin, IL17, IL4, IL2Rα, MIG, MIP1β, IFNα2, SDF1α, IL16, LIF, TNFβ, MIF, RANTES, IL18, PDGFβ, IP10, IL13, MIP1α, GCSF, GROα, HGF, IL1α, SCF, TRAIL, MCSF, CTACK, IL7, IL9 and SCGFβ;

the behavioral performance comprise the scores listed in Table 1, i.e., CAR_total, CAR_S, CAR_N, CAR_D, ADOS_S, ADOS_C, ADOS_P, ADOS_I, SRS_total, SRS_AWA, SRS_COG, SRS_COM, SRS_MOT and SRS_MANN; and

the clinical information comprises gender and age.

In some embodiments, the subset of relevant characteristic information includes: (i) baseline expression levels of IL16, GROα and IL7; (ii) scores of ADOS_S, ADOS_C, ADOS_P, SRS_COM and CARS_D; (iii) gender and age. The classifier is support vector machine.

In some embodiments, the subset of relevant characteristic information includes: (i) baseline expression levels of IL16, GROα and TNFβ; (ii) scores of ADOS_S, ADOS_C, ADOS_P and CARS_total; (iii) gender and age. The classifier is partial Least Squares.

In some embodiments, the subset of relevant characteristic information includes: (i) baseline expression levels of IL16, GROα, IL7, TNFβ and CTACK; (ii) scores of ADOS_S, ADOS_C, ADOS_P, SRS_COM, CARS_D, CARS_total and SRS_MOT; (iii) gender and age. The classifier is Neural Networks.

In some embodiments, the subset of relevant characteristic information includes: (i) baseline expression levels of IL16, GROα, IL7, TNFβ, LIF and MIF; (ii) scores of ADOS_S, ADOS_C, ADOS_P, SRS_COM, CARS_D and CARS_total; (iii) gender and age. The classifier is sparse Linear Discriminant Analysis.

In some embodiments, the subset of relevant characteristic information includes: (i) baseline expression levels of IL16, GROα and TNFβ; (ii) scores of ADOS_S, ADOS_C, ADOS_P and CARS_total; (iii) gender and age. The classifier is Oblique Random Forest.

It should be understood that variant forms of the prediction method of the present invention are also included within the scope of protection of the present invention. For example, what is provided is the characteristic information or the subset of relevant characteristic information described herein for use in predicting a response to bumetanide in a subject with ASD. For another example, what is provided is use of the characteristic information or the subset of relevant characteristic information described herein in the preparation of a set of features for predicting the response to bumetanide in a subject with ASD. For another example, what is provided is use of agents for measuring the expression levels of the set of cytokines described in the present invention (such as antibodies that can specifically bind to the cytokines) in preparation of an agent or a kit for predicting a response to bumetanide in a subject with ASD, wherein the prediction can be performed, for example, through the prediction method described in the present invention. For another example, what is provided is use of the characteristic information or the subset of relevant characteristic information described herein in built a prediction model for predicting the response to bumetanide in a subject with ASD.

The prediction method of the present invention may be performed by a computer.

The present invention also provides a prediction device, which may be a computer device, the prediction device comprising:

an input module which is configured to receive characteristic information of a subject with ASD, wherein the characteristic information includes: (i) baseline expression levels of a set of cytokines in the subject; (ii) baseline behavioral performance of the subject; and (iii) clinical information of the subject, including gender, age, and BMI, etc.;

a classifying module which is configured to comprise a classifier, wherein the classifier can predict the response of the subject to bumetanide using a classifier based on the characteristic information; and

an output module which is configured to output the predicted result.

In some embodiments, the prediction device may further comprise a training module which is configured to train the classifier using a training data set.

In some embodiments, the prediction device may further comprise a validating module which is configured to validate the classifier.

The present invention also provides a computer readable medium comprising computer executable instructions recorded thereon for performing the operation comprising:

receiving characteristic information of a subject with ASD, wherein the characteristic information includes: (i) baseline expression levels of a set of cytokines in the subject; (ii) baseline behavioral performance of the subject; and (iii) clinical information of the subject, including gender, age, and BMI, etc.;

In some embodiments, the operation further comprises validating the classifier.

In the prediction device and the computer readable medium, the definitions related to characteristic information and the classifier are the same as in the previous section.

According to the present invention, based on the predicted subject’s response to bumetanide, it can be determined whether the subject will benefit from bumetanide treatment, or whether the patient is a candidate for bumetanide treatment. For example, “having response” , “high response” , “better response” , “positive response” or “responder” means that the subject is likely to benefit from bumetanide treatment, “no response” , “low response” , “least response” , “negative response” or “non-responder” means that the patient is unlikely to benefit from bumetanide treatment to bumetanide. From there, treatment can be selected for the patient based on that prediction result. For example, bumetanide may be administered to a subject that is likely to benefit from bumetanide treatment, while bumetanide may not be administered to a subject that is unlikely to benefit from bumetanide treatment, or the subject that is unlikely to benefit from bumetanide treatment may be administered other treatments that do not include bumetanide.

In some embodiments, the present invention provides a method for treating ASD in a subject, the method comprising:

predicting a response of the subject afflicted with ASD to bumetanide treatment based on characteristic information of the subject, wherein the characteristic information includes: (i) baseline expression levels of a set of cytokines in the subject; (ii) baseline behavioral performance of the subject; and (iii) clinical information of the subject, including gender, age, and BMI, etc.;

selecting a treatment for the subject based on the prediction result.

In the method for treating ASD, the definitions related to the characteristic information and the implementation of the prediction are the same as described in the previous section.

In the present invention, for the treatment of a subject with ASD using bumetanide, it should be understood that bumetanide can be administered to the subject in an effective amount and in an appropriate way, which can be determined by a skilled clinician.

For example, bumetanide may be administered parenterally or non-parenterally, e.g., orally, intravenously, intramuscularly or by any other suitable route. Bumetanide may be formulated in a dosage form suitable for the above routes of administration. For example, dosage forms include those adapted for oral administration such as tablet, capsule, caplet, pill, troche, powder, syrup, elixir, suspension, solution, emulsion, sachet, and cachet; or parenteral administration such as sterile solution, suspension, and powder for reconstitution.

For example, bumetanide may be administered to the subject at a daily total dosage ranging from about 0.5 to 10 mg, preferably from 1 to 6 mg, and more preferably from 2 mg to 4 mg, divided into one, two, or three doses. It may be administered orally once, twice, or thrice daily to the patient using a dosage form that comprises 0.5, 1, 2 mg bumetanide, or a pharmaceutically acceptable salt thereof. Administration of a single dose may enhance patient compliance, while administration of several smaller doses ensures constant serum levels.

The period of treatment of bumetanide should be sufficient to demonstrate its therapeutic effect in the subject. The duration of treatment may be ranged from about one month to six months, e.g., two months to three months, two months to four months.

The present invention also provides a kit to predict a response to bumetanide in a subject with ASD. The kit may comprise agents for measuring the expression levels of the set of cytokines described in the present invention, such as antibodies that can specifically bind to the cytokines. The kit may also comprise instructions for the prediction method in the form of a label or separate insert. For example, the instructions may inform how to obtain the characteristic information of a subject and how to predict a response of the subject to bumetanide, for example, as described in the present invention. In some embodiment the kit can comprise one or more containers used to accommodate the agents.

In some embodiments, the kit may further comprise bumetanide and such a kit can be used to treat ASD in subject. The kit can also comprise one or more containers used to accommodate bumetanide. The instructions may further inform that how to treat a subject based on the prediction result, for example, as described in the present invention.

In the kit, the definitions related to the characteristic information and the implementation of the prediction are the same as described in the previous section.

EXAMPLES

Materials and Methods

Participants

The ASD participants were recruited from the Shanghai Xinhua ASD registry at Shanghai Jiaotong University Medical School Affiliated Xinhua Hospital in Shanghai, China, including the participants from two previous registered clinical studies, i.e., CHICtr-OPC-16008336 and NCT03156153. The patients were diagnosed with ASD according to the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5) . Diagnoses were confirmed with the Autism Diagnostic Observation Schedule (ADOS) , and a Children Autism Rating Scale (CARS) total score of no less than 30. Exclusion criteria include liver and kidney dysfunction; a history of allergy to sulfa drugs; abnormal electrocardiography; genetic or chromosomal abnormalities; suffering from nervous system diseases (e.g., epilepsy, etc. ) . Comprehensive behavioral assessments and collections of clinical samples were performed for all patients. Between May 1^st, 2018, to April 30^th, 2019, a total of 90 ASD children, aged 3-10 years old, under a three-month stable treatment of bumetanide without behavioural interventions and any concomitant psychoactive medications had both blood draws and behavioral assessments. Among these patients, 11 of them were further excluded due to the lack of the follow-up data at month 3. Therefore, the current analysis used a subsample of 79 young children with ASD, whose blood samples were available both before and after the treatment. The blood samples were sent in 2 batches (Discovery Set: n=37 on December 4, 2019; and Validation Set: n=42 on May 22, 2019) to measure the plasma levels of 48 cytokines for the immune response, and the clinical symptoms were assessed using CARS, ADOS and the Social Responsiveness Scale (SRS) . The 48 cytokines include Basic FGF (bFGF) , β-NGF, CTACK, Eotaxin, G-CSF, GM-CSF, GRO-α, HGF, IFN-α2, IFN-γ, IL-10, IL-12p40, IL-12p70, IL-13, IL-15, IL-16, IL-17, IL-18, IL-1α, IL-1β, IL-1Ra, IL-2, IL-2Rα, IL-3, IL-4, IL-5, IL-6, IL-7, IL-8, IL-9, IP-10, LIF, M-CSF, MCP-1, MCP-3, MIF, MIG, MIP-1α, MIP-1β, PDGF-BB, RANTES, SCF, SCGF-β, SDF-1α, TNF-α, TNF-β, TRAIL and VEGF, 35 of which are used in the subsequent procedures (as shown below) . The 35 cytokines and the clinical assessments are listed in Table 1.

Table 1. Measured immune factors, clinical assessments and demographic parameters

Following the protocols of previous studies ⁸, bumetanide treatment consisted of two 0.5 mg tablets per day for three months, given at 8: 00 am and 4: 00 pm. The tablet size is 8mm diameter x 2mm thickness, which is quite small. Each time, the patient took half of a tablet, which was not difficult for most of the patients. However, the careers were recommended to grind the half-tablet into powder and give the powder in water, if necessary. Possible side effects were closely monitored during the treatment. Blood parameters (serum potassium and uric acid) were monitored via laboratory tests (Table 2) and symptoms (thirst, diuresis, nausea, vomiting, diarrhea, constipation, rash, palpitation, headache, dizziness, shortness of breath, and any other self-reported symptoms) were telephone interviewed (Table 3) , and both of them were reported to the research team by telephone at 1 week and 1 month after the initiation of treatment and at the end of the treatment period. The cytokine levels of the children with gastrointestinal problems were compared with those without such problems (Table 4) . Behavioural assessments of CARS and ADOS and measurements of cytokine levels were performed at the baseline before the treatment and after the 3-month treatment. The behavioural assessment of SRS was used at the baseline only. The study was conducted in accordance with the provisions of the Declaration of Helsinki and Good Clinical Practice guidelines and was approved by the Ethics Committee of Xinhua Hospital affiliated to Shanghai Jiao Tong University School of Medicine. Written informed consent was obtained from the parent or legal guardian of each participant before sample collection.

Table 2. Side effects measured by blood parameters reported during the treatment*.

*Score = 0 means none symptoms; Score = 1 in K means hypokalemia (< 3.5 mmol/L) ; Score = 1 in U means
increased urine elimination (> 417 umol/L) .

before, at the baseline; 1-week, after 1-week treatment of bumetanide; 1-month, after 1-month treatment of bumetanide; 3-month, after 3-month treatment of bumetanide.

Table 3. Side effects measured by symptoms reported during the treatment*.

*Score = 0, 1, 2 and 3 means none, mild, moderate and severe symptoms.

Table 4. The cytokine levels of the children who have gastrointestinal problems were compared with who don’t have.

¹Data [i.e., mean (SD) ] were first normalized and second corrected for batch effect.
²Mann-Whitney U test with multiple group comparisons.
³FDR adjustment for multiple testing.

Measures

Clinical assessments

The CARS was used to diagnose and evaluate the severity of clinical symptoms of ASD patients. The CARS consisted of 15 items rated on a 7-point scale from one to four; higher scores are associated with a higher level of impairment. Total scores can range from a low of 15 to a high of 60; scores below 30 indicate that the individual is in the non-autistic range, scores between 30 and 36.5 indicate mild to moderate autism, and scores from 37 to 60 indicate severe autism. We further categorized these items into three subscales ²⁸: Social impairment, Negative emotionality, and Distorted sensory response. ADOS was used as a supplement to gauge disease severity, and it contained total score items and 4 modules for assessment of Social interaction, Communication, Play, and Imaginative use of materials for individuals suspected of having ASD ²⁹. SRS identified a wide spectrum of deficits in reciprocal social behavior, ranging from absent to severe, based on observations of a child’s behavior in naturalistic social settings, focused on the behavior of a child or adolescent between the ages of 4 and 18 years. It was a 65-item questionnaire that is completed by teacher, a parent, and/or another adult caregiver. Scoring is on a four-point Likert Scale. Five subscales are also provided: Social Awareness (AWA) , Social cognition (COG) , Social Communication (COM) , Social Motivation (MOT) , and Autistic Mannerism (MANN) ³⁰.

Cytokine levels

For each subject peripheral blood was collected, centrifuged at 2300 rpm for 10 min, and the plasma harvested. Plasma was aliquoted and stored at -70 ℃ until cytokine analysis. A plane of 48 cytokines, chemokines, and growth factors were measured using the Bio-Plex multiple immunoassays (BIO‐RAD Laboratories, Inc. ) . Prior to setting up our assays, the Bio-Plex 200 System plane Reader instrument was calibrated according to manufacturer instructions. To prepared experimental samples, frozen plasma aliquots were passively thawed to room temperature and diluted four-fold in assay buffer (15 μL sample +45 μL sample diluent HB) . After preparation of capture bead mixture, and standards, the immunoassay was carried out on a 96-well plane. The experimental steps were in accordance with the instructions. Data acquisition was set to a 50-bead count minimum per analyte per well. Unknown sample cytokine concentrations were processed and presented with Bio-plex Manager software using a standard curve derived from the known reference cytokine concentrations supplied by the manufacturer. A five-parameter model was used to calculate final concentrations and values were expressed in pg/ml. The sensitivity of this assay allowed the detection of cytokine concentrations within the following ranges: IFN-α2 3.6-3992.4 pg/ml; IFN-γ 0.9-14556.8 pg/ml; IL-1β 0.3-5375.0 pg/ml; IL-4 0.2-3455.0 pg/ml; IL-6 0.4-5961.8 pg/ml; IL-8 0.2-15570.4 pg/ml; MIG 5.7-30955.2 pg-ml; TNF-α 3.3-52256.0 pg/ml, etc. Concentrations obtained below the sensitivity limit of detection (LOD) of the method were excluded from all subsequent analyses.

Statistical analysis

Data preprocessing

After excluding the cytokines whose values were less than the limit of detection, 35 cytokines were included in our analysis. Data from two batches, containing both the baseline and the change (the difference value from the baseline and the follow-up data) of the cytokine levels, were min-max normalized and log transformed separately. To adjust for the batch effect, we applied an empirical Bayes approach to the baseline of cytokine levels using the ‘ComBat’ parametric algorithm provided in the R package ‘sva’ ³¹. Principle component analysis (PCA) was used to visualize the non-biological variation due to the batch effect and was repeated to confirm its adjustment (Figure 4) . Before and after treatment, we compared the demographic parameters (i.e., sex proportion, age, body mass index [BMI] ) and symptom severity (i.e., ADOS and CARS) between these two data sets using t test, Mann-Whitney U test, or Pearson’s chi-squared test where applicable.

Multivariate association analysis to characterize the immuno-behavioural covariation

First, the pairwise correlation between the CARS_total score and each of the 35 cytokine levels were assessed by the Spearman-rank correlation. The correlation between the change in the CARS_total score and the change in each of the 35 cytokine levels after the treatment was also tested. The false discovery rate (FDR) was used to correct for the multiple comparisons.

Second, to uncover the multivariate association between the behavioural assessments and the cytokine levels, we employed the sparse canonical correlation analysis (sCCA) provided in a R-package “sRDA” (version 1.0.0) ³². Canonical correlation analysis (CCA) is a classical method for determining the relationship between two sets of variables. Given two data sets X₁ and X₂ of dimensions n×p₁ and n×p₂ respectively from n observations, CCA seeks the pairs of linear combinations (i.e., the canonical variables) , one from the variables in X₁ and the other from the variables in X₂, that are maximally correlated with each other. However, some variables may make negligible but non-zero contributions to the canonical variables. sCCA was developed to address this issue. sCCA applies an L₁ penalty to the canonical weights, which forces some of them to take a value of exactly zero. Mathematically,

Here, c₁ and c₂ are assumed to fall within the boundsandwhere p₁ and p₂ are the numbers of features in X₁ and X₂ respectively. We refer to w₁ and w₂ as the canonical weights, and X₁w₁ and X₂w₂ as the canonical scores. Therefore, this algorithm could identify a linear combination of three CARS subscales (i.e., the behavioural-component) that was significantly associated with another linear combination of a few cytokine levels (i.e., the cytokine-component) . Meanwhile, the sparsity of this algorithm ensured only the key cytokines driving the behavioural association were selected in the immune component.

In the Discovery Set, we explored the sCCA between CARS subscales and cytokine levels using the baseline data or using the changes between baseline and follow-up. The significance of an identified canonical correlation was assessed by 5000 permutations³². Only the significant canonical components were retained. Sensitivity analysis was conducted using the data from the Validation Set by re-evaluating the canonical correlation after controlling for the potential confounders including age, sex, BMI and the canonical variables at the baseline. Using the cytokine-component score (x-axis) and the behavioural-component score (y-axis) established by sCCA, each patient could be mapped onto a 2-dimensional, called the immuno-behavioural covariation plane, characterizing the immuno-behavioural covariation in the response patterns to the bumetanide treatment among young children with ASD.

Clustering analysis to identify the immuno-behavioural groups

We applied k-means, an unsupervised clustering algorithm, to identify the clusters of patients according to the immuno-behavioural covariation. The patients in each cluster (i.e., an immuno-behavioural group within ASD) had the similar canonical scores, suggesting they shared the similar patterns of response to bumetanide in both immune system and clinical behaviour. The cluster structures were first identified using the Discovery Set and then validated using the Validation Set. The optimal number of clusters was selected based on the elbow (maximum change) of the scree plot using the Hubert statistic implemented in the R package ‘NBclust’ ³³.

To demonstrate the distinct patterns of response to bumetanide in the immuno-behavioural groups, we applied the one-sample t test to the after-treatment changes of both the CARS subscales and the cytokines selected by the sCCA algorithm as described above. We also compared these changes among the above identified immuno-behavioural groups using the Kruskal-Wallis rank sum test with the FDR correction for multiple comparisons.

Prediction of the treatment response to bumetanide using the baseline information

To predict the response to bumetanide of a patient with ASD, we trained the classifiers for the immuno-behavioural groups identified above using the baseline information before the treatment. The baseline information included the 35 cytokine levels, 3 types of clinical assessment (CARS, SRS, ADOS) and 2 demographic parameters (sex and age) . The classifiers included the Oblique Random Forest (ORF) , Partial Least Squares (PLS) , sparse Linear Discriminant Analysis (sLDA) , Neural Networks (NN) and Support Vector Machine (SVM) as implemented in the R package ‘caret’ with both feature selection and oversampling ³⁴. The models were first trained using the Discovery Set, and their performances were compared using the Validation Set. The 95%confidence interval of the difference between the areas under the curves of a pair of models was constructed by 100 bootstraps.

First, we tested whether including the cytokine levels at the baseline could improve the prediction of the immuno-behaviourally defined responders. The averaged performances of these five classifiers (i.e., the averaged area under the curve, ) were reported and compared.

Second, we tested whether the behaviourally defined responders were more difficult to be predicted at the baseline compared with the immuno-behaviourally defined ones. In the previous clinical trials of bumetanide for ASD ^5-7, the proportion of patients who responded positively to the treatment was between 30%and 40%. Therefore, we divided the patients with ASD into two groups according to the ΔCARS_total (= the baseline CARS_total –the follow-up CARS_total) with a cut-off of 2.5 (n=22, 27.85%of the patients had ΔCARS_total > 2.5) or 2 (n=32, 40.51%of the patients had ΔCARS_total > 2) points.

Results

Participants

Two data sets of children with ASD (n = 79) were used in the current study. Data Set 1 (n = 37) had a mean age of 47 months (±17.35 months) , 18.92%of whom were girls; Data Set 2 (n = 42) had a mean age of 54 months (±20.19 months) , 23.81%of whom were girls. No significant difference in clinical characteristics or cytokine levels was identified between these two data sets (Table 5; Table 6) .

Table 5. The demographic and clinical (mean (SD) ) characteristics of two data sets

¹T-test statistic for normal features and Mann-Whitney U test for non-normal features, while chi-
square test for sex.
²Sample size for ADOS data in Discovery Set and Validation Set are 36 and 41.
³Sample size for SRS data in Discovery Set and Validation Set are 21 and 39.

BMI, body mass index; CARS, Childhood Autism Rating Scale; ADOS, the Autism Diagnostic Observation Schedule; SRS, the Social Responsiveness Scale; CARS_total, CARS total score; CARS_S, CARS score on social impairment domain; CARS_N, CARS score on negative emotionality domain; CARS_D, CARS score on distorted sensory response domain; ADOS_S, ADOS score on social interaction; ADOS_C, ADOS score on communication; ADOS_P, ADOS score on play; ADOS_I, ADOS score on imaginative use of materials; SRS_AWA, SRS score on social awareness; SRS_COG, SRS score on social cognition; SRS_COM, SRS score on social communication; SRS_MOT, SRS score on social motivation; SRS_MANN, SRS score on autistic mannerism; SRS_total, SRS total score;

Table 6. The baseline levels of cytokines in two data sets.

¹Data [i.e., mean (SD) ] were first normalized and second corrected for batch effect.
²Mann-Whitney U test.
³FDR adjustment for multiple testing.

Changes after the administration of bumetanide

All patients were treated with bumetanide for 3 months, and the CARS total score decreased after the treatment (effect size Cohen’s d = 1.26, t₇₈ = 11.21, p < 0.001) . The treatment effect showed no difference between two data sets (ΔCARS_total: mean (±SD) : 1.54 (±1.40) vs. 1.90 (±1.34) ) . Consistent to the previous studies of the low-dose bumetanide for ASD, the side effects were rarely reported (Table 2-3) . No significant difference in the cytokine levels between the children with and without the gastrointestinal problems at the baseline (Table 4) . A number of cytokine levels were changed significantly after the treatment (Table 7) . No significant pairwise association could be both identified in the Discovery Set and validated using the Validation Set among four groups of variables, including the baseline CARS total score, the baseline cytokine levels, the change of CARS total score, and the changes of cytokine levels (Figure 5) .

Table 7. The change levels of cytokines in two data sets.

¹The degree of freedom for the One sample t-test statistic is 36.
²The degree of freedom for the One sample t-test statistic is 41.
³Mann-Whitney U test.
⁴FDR adjustment for multiple testing.

Covariation between symptom improvement and cytokine changes

Using the Discovery Set, we found a canonical correlation (r = 0.459; p < 0.001 by permutation; Figure 1A) between 2 canonical components (i.e., cytokine-component and behavioural-component) by sCCA. The cytokine-component was a combination of the changes of the 3 cytokine levels, including the MIG, IFN-α2, IFN-γ. The behavioural-component was a combination of the changes of 3 subscales of the CARS scores, including the social impairment score, negative emotionality score, distorted sensory response score. Applying these canonical weights to an independent data set (i.e., the Validation Set) , we confirmed the correlation between the cytokine-component and the behavioural-component (r = 0.316; p =0.012 by permutation; Figure 1B) .

Sensitivity analysis

Using the Validation Set, we found that the correlation between the canonical components identified above remained significant (Table 8) . At the baseline, we also found that the cytokines canonical scores were associated with other clinical assessments, including the SRS_total (r = -0.269; t₆₃ = -2.21, p = 0.031) , ADOS_S (r = 0.296; t₇₆ = 2.70, p = 0.009) , ADOS_P (r = -0.244; t₇₆ = -2.19, p = 0.032) , and ADOS_I (r = -0.251; t₇₆ = -2.26, p = 0.027) .

Table 8. Correlation between the identified canonical components after controlling for baseline variables.

Sensitivity analysis was conducted by re-evaluating this correlation after controlling for potential confounders at the baseline.

¹The degree of freedom for the Student's t-test statistic is 37.
^*P value is less than 0.05.

CARS_S, CARS score on social impairment domain; CARS_N, CARS score on negative emotionality domain; CARS_D, CARS score on distorted sensory response domain; IFN-γ, Interferon gamma; IFN-α2, Interferon alpha 2; MIG, Monokine induced by gamma interferon; BMI, Body Mass Index.

Three distinct response patterns revealed by the immuno-behavioural covariation

Using both the CARS-and the cytokine-component scores, we found that the patients fell into 3 clusters (Figure 2A-B; Figure 6) . No significant difference in the distribution of patients among these 3 clusters between the Discovery Set and the Validation Set. Besides, no significant difference in the baseline of CARS scores and cytokine levels among these 3 clusters. Comparing among these 3 groups of patients in both data sets (Figure 2C-D; Table 9; Figure 7) , we found that the best responding group (n = 17) had the greatest reduction in the CARS total score (ΔCARS_total: 3.32 (±1.47) , Hedge’s g = 2.16, p < 0.001) , which was most prominent in the social impairment score (ΔCARS_S: 2.15 (±1.13) , g = 1.81, p < 0.001) and increased cytokine levels (ΔMIG: g = 0.72, p = 0.012; ΔIFN-α2: g = 0.84, p = 0.006) . The least responding group (n = 18) had the least reduction in the CARS total score (ΔCARS_total: 1.03 (±0.96) , g = 1.02, p < 0.001) , which was most significant in the negative emotionality score (ΔCARS_N: 0.69 (±0.55) , g = 1.21, p < 0.001) , and decreased in the least responding group (ΔMIG: g = -1.07, p < 0.001; ΔIFN-γ: g = -1.32, p < 0.001) . The medium responding group had a significant decrease in both the CARS_total score and all subscales with a small effect size each, while the IFN-γ level decreased and the IFN-α2 level increased in this group (Table 9) .

Baseline cytokine levels helped in identifying the treatment-responders

Training 5 different types of classifiers with Discovery Set and testing the performance using Validation Set, we found that including the cytokine levels at the baseline significantly improved the prediction accuracy for both the best responding group (with/without the baseline cytokine levels = 0.768/0.646; improvement in95%confidence interval (CI) : (0.103, 0.130) ; Figure 3A-B) and the least responding group (with/without the baseline cytokine levels = 0.698/0.618; improvement in95%CI: (0.064, 0.097) ; Figure 3C-D) . Each of the five models showed a better performance after including the cytokine levels into the model (Table 10) .

Wherein,

35 cytokines introduced in the model: IL1beta, IL6, IL8, IFNgamma, TNFalpha, MCP1/MCAF, Eotaxin, IL17, IL4, IL2Rα, MIG, MIP1β, IFNα2, SDF1α, IL16, LIF, TNFβ, MIF, RANTES, IL18, PDGF-BB, IP10, IL13, MIP1α, GCSF, GROα, HGF, IL1α, SCF, TRAIL, MCSF, CTACK, IL7, IL9, SCGFβ;

other variables introduced by the model: age, sex, CARS_total, CARS_S, CARS_N, CARS_D, ADOS_S, ADOS_C, ADOS_P, ADOS_I, SRS_AWA, SRS_COG, SRS_COM, SRS_MOT, SRS_MANN, SRS_total;

all variables were at baseline level.

The output was whether the subject belongs to the best responding group as defined above.

In Table 10, “included” meant that the 35 cytokines and other variables mentioned above were included, and “not included” means that the other variables mentioned above were included and the 35 cytokines were not included.

We also trained 5 different types of classifiers with Discovery Set, tested the performance using Validation Set and used the following variables to predict behaviourally defined responders and immuno-behaviourally desfined responders.

35 cytokines introduced in the model: IL1β, IL6, IL8, IFNγ, TNFα, MCP1/MCAF, Eotaxin, IL17, IL4, IL2Rα, MIG, MIP1β, IFNα2, SDF1α, IL16, LIF, TNFβ, MIF, RANTES, IL18, PDGF-BB, IP10, IL13, MIP1α, GCSF, GROα, HGF, IL1α, SCF, TRAIL, MCSF, CTACK, IL7, IL9, SCGFβ; other variables introduced by the model: age, sex, CARS_total, CARS_S, CARS_N, CARS_D, ADOS_S, ADOS_C, ADOS_P, ADOS_I, SRS_AWA, SRS_COG, SRS_COM, SRS_MOT, SRS_MANN, SRS_total; all variables were at baseline level.

To define behaviourally defined responders, we divided the patients with ASD into two groups according to the ΔCARS_total (= the baseline CARS_total –the follow-up CARS_total) with a cut-off of 2.5 (n=22, 27.85%of the patients had ΔCARS_total > 2.5) points or 2 (n=32, 40.51%of the patients had ΔCARS_total > 2) points. These two groups were defined as behaviourally defined responders.

The immuno-behaviourally defined responders were defined as described in “Three distinct response patterns revealed by the immuno-behavioural covariation” above. Specifically, to define immuno-behaviourally defined responders, we applied k-means, an unsupervised clustering algorithm, to identify the clusters of patients according to the immuno-behavioural covariation. The patients in each cluster (i.e., an immuno-behavioural group within ASD) had the similar canonical scores, suggesting they shared the similar patterns of response to bumetanide in both immune system and clinical behaviour. The cluster structures were first identified using the Discovery Set and then validated using the Validation Set. The optimal number of clusters was selected based on the elbow (maximum change) of the scree plot using the Hubert statistic implemented in the R package ‘NBclust’ ³³.

To demonstrate the distinct patterns of response to bumetanide in the immuno-behavioural groups, we applied the one-sample t test to the after-treatment changes of both the CARS subscales and the cytokines selected by the sCCA algorithm as described above. We also compared these changes among the above identified immuno-behavioural groups using the Kruskal-Wallis rank sum test with the FDR correction for multiple comparisons. The immuno-behaviourally defined responders were classified to the best responding group, the medium responding group and the least responding group.

We found that the five models could predict the behaviourally defined responder group with a high accuracy (Table 11; Figure 8A) . Similar results were found when using the threshold of 2.5 of CARS score change to behaviourally define the responders (Figure 8B) .

Furthermore, we found that the immuno-behaviourally defined responders were of higher accuancy to be predicted at the baselinecompared with the behaviourally-defined responders.

Table 10. Comparison of AUC between the models including and without including the cytokines to predict the immuno-behaviourally defined best responding group and the least responding group.

ORF –Oblique Random Forest; SVM –support vector machine; PLS –Partial Least Squares, sLDA –sparse
Linear Discriminant Analysis, NN –Neural Networks

Table 11. Comparison of AUC between the models predicting the immuno-behaviourally defined responder group and the behaviourally defined responder group

¹Models was constructed by 100 bootstraps.

AUC, area under ROC; CI, confidence interval; SD, standard deviation; ORF, Oblique Random Forest; PLS, Partial Least Squares; SVM, Support Vector Machine; sLDA, sparse Linear Discriminant Analysis; NN, Neural Networks; ΔAUC, the differences between AUC of models with and without each cytokine.

Taking advantages of the five models established above, we make some examples to show the efficacy how these models work. For example:

We took advantage of Support-Vector Machine (SVM) and found that the predictive model with three cytokines (i.e., IL16, GROalpha and IL7) and seven other variables (i.e., age, sex, ADOS_S, ADOS_C, ADOS_P, SRS_COM and CARS_D) achieved a high accuracy of 0.833 in the best responding group prediction (area under curve, AUC = 74.0%) . (Figure 9)

We took advantage of Partial Least Squares (PLS) and found that the predictive model with three cytokines (i.e., IL16, GROalpha and TNFbeta) and six other variables (i.e., age, sex, ADOS_S, ADOS_C, ADOS_P and CARS_total) achieved a high accuracy of 0.833 in the best responding group prediction (area under curve, AUC = 74.0%) . (Figure 9)

We took advantage of Neural Networks (NN) and found that the predictive model with five cytokines (i.e., IL16, GROalpha IL7, CTACK and TNFbeta) and nine other variables (i.e., age, sex, ADOS_S, ADOS_C, ADOS_P, SRS_COM, CARS_D, CARS_total and SRS_MOT) achieved a high accuracy of 0.810 in the best responding group prediction (area under curve, AUC = 74.0%) . (Figure 9)

We took advantage of sparse Linear Discriminant Analysis (sLDA) and found that the predictive model with six cytokines (i.e., IL16, GROalpha, IL7, LIF, MIF and TNFbeta) and eight other variables (i.e., age, sex, ADOS_S, ADOS_C, ADOS_P, SRS_COM, CARS_D and CARS_total) achieved a high accuracy of 0.786 in the best responding group prediction (area under curve, AUC = 76.2%) . (Figure 9)

We took advantage of Oblique Random Forest (ORF) and found that the predictive model with three cytokines (i.e., IL16, GROalpha and TNFbeta) and six other variables (i.e., age, sex, ADOS_S, ADOS_C, ADOS_P, and CARS_total) achieved a high accuracy of 0.810 in the best responding group prediction (area under curve, AUC = 68.2%) . (Figure 9)

Discussion

In this study, we observed a significant improvement of clinical symptoms with bumetanide treatment in children with ASD, and such improvement was associated with a pattern of changes in three cytokine levels, namely the IFN-γ, MIG and IFN-α2 (r=0.459 in the Discovery Set and r=0.316 in the Validation Set) . These cytokine levels at the baseline could improve the prediction of the bumetanide responders compared with using the behavioural assessments alone, and the best predictor achieved an AUC of 0.83 in the independent test data set (Table 10) . The implications of these findings may be twofold: 1) a significant part of the clinical heterogeneity in the treatment effect of bumetanide for ASD is associated with the differences in the immune system of patients, and 2) the component score of cytokines had a potential to construct a blood signature for predicting and monitoring the bumetanide treatment in young children with ASD.

Our finding of an immuno-behavioural covariation highlights the role of the immune system in the clinical effect of bumetanide in young children with ASD. IFN-γ, as a T helper cell 1 (Th1) cytokine with pro-inflammatory effects, was selected by the sCCA algorithm to be one of the three cytokines to form the canonical score that was associated with the improvement in CARS. Compared to controls, higher level of IFN-γ has been reported in the brain tissue²², cerebrospinal fluid (CSF) ²³, plasma³⁵ and peripheral blood mononuclear cell (PBMC) ³⁶ in ASD patients, and lower level has been observed in neonatal dried blood samples (n-DBSS) of ASD children³⁷. Accumulating evidences support that IFN-γ can inhibit chloride secretion³⁸ and down-regulate both the NKCC1 expression^16, 38 and the Na⁺-K⁺-ATPase expression¹⁶, which had been implicated in the GABAergic dysfunction in ASD^10, 39. Indirectly, an animal study also showed that stimulation with high concentration of IFN-γ could increase the expression of IL-1β⁴⁰, which is an inflammatory cytokine that can affect the expression of chloride transporters and delay the developmental switch of GABA signaling¹⁷. Therefore, the immune system may interact with the mechanism of action of bumetanide to restore the GABA function in ASD.

The cytokine-symptom association was identified in the changes after the treatment of bumetanide but not before the treatment, suggesting that bumetanide might interact with the cytokines and the changes of which contributed to the treatment effect of bumetanide. Animal studies showed a rapid brain efflux of bumetanide, but a number of clinical trials have shown a significant treatment effect for neuropsychiatric disorders, including ASD, epilepsy and depression ^41, 42. These findings may suggest the possible systemic effects of bumetanide as a neuromodulator for these neuropsychiatric disorders. Considering its molecular structure, bumetanide has been recently identified by an in vitro screen of small molecules that can act as an anti-proinflammatory drug via interleukin inhibition ⁴³. This anti-proinflammatory activity of bumetanide might alter the blood levels of cytokines outside the brain-blood-barrier (BBB) . In fact, it has already been reported that bumetanide reduced the Lipopolysaccharide-induced production of proinflammatory cytokines following a direct pulmonary administration in RAW264.7 cells and in lung-injured mice ⁴⁴. These inflammatory signaling messengers may pass the BBB ⁴⁵ and influence the neuronal chloride homeostasis via, for example, altering the KCC2 expression ¹⁸. The plausibility of reducing inflammation to enhance the KCC2 expression has recently been discussed in a 2020 review ¹⁷. Indeed, we found that the best responding group (Hedge’s g = 2.16 for the reduction of CARS total score) had the greatest increase in the cytokine-component score. In the contrary, the least responding group (g = 1.02) had the greatest decrease in the cytokine-component score. Taken together, these findings suggest that bumetanide may be a drug to inhibit NKCC1 and enhance KCC2 through its interplay with the cytokines inside and/or outside of the brain.

Our findings may suggest that the identified canonical score of cytokines had a potential to construct a blood signature for predicting and monitoring the bumetanide treatment in young children with ASD. Accurately identifying patients who are likely to respond positively to bumetanide can facilitate the precision medicine for ASD. Our prediction model based on the cytokine levels before the treatment may provide a potentially new tool for the precision medicine of ASD. Given the inherit heterogeneity of ASD, it is of great clinical value to accurately identify the subgroup of ASD patients that are likely to respond positively to its medical treatments⁴⁶. Multiple factors, including both genetic and environmental factors, could contribute to the heterogeneity of ASD and its response to treatment ^12, 19, 47. For example, prenatal insults including maternal infection and subsequent immunological activation during gestation may increase the risk of autism in the child¹⁹. Increased exposure to air pollution during gestation was also associated with abnormalities in mitochondrial metabolism during childhood, which may also increase the risk for ASD⁴⁷. But our findings suggested that using the cytokine levels improved the prediction of response to the bumetanide treatment for ASD in three folds: 1) The immuno-behavioural covariation enabled the identification of more homogenous subgroup of ASD in terms of response to bumetanide. Using 5 models and an independent test data set, we demonstrated consistent evidence that the responding group identified by the immuno-behavioural covariation could be significantly better predicted by the baseline information compared with the responder group defined by the CARS alone. 2) Combining the cytokine levels with clinical assessments of CARS, ADOS and SRS before treatment, we achieved a higher accuracy of 84.3%in identifying the ASD children that were likely to respond positively to the bumetanide treatment. 3) The blood cytokine levels are more easily accessible in clinical practice.

There are several limitations of our study. Although we had two separate data sets to validate our findings, the sample sizes were limited. Previously a sex difference in the cytokine-symptom association had been reported, but we could not test such sex difference owing to the small numbers of girls with ASD in our sample⁴⁸. Hence, future multi-center, prospective studies with lager sample sizes are necessary to confirm the current findings. Second, the hypothesized molecular mechanism underlying the bumetanide treatment effect for ASD requires causal confirmation in animal studies.

In summary, we identified an association between the changes of the cytokine levels and the improvements in symptoms after the bumetanide treatment in young children with ASD, and found that the treatment effect of bumetanide can be better characterized by an immuno-behavioural covariation. This finding may provide new clinically important evidence supporting the hypothesis that immune responses may interact with the mechanism of bumetanide to restore the GABAergic function in ASD. This finding may also have relevance for determining enriched samples of ASD children to participate in novel drug treatment studies of drugs with a similar mode of action to bumetanide, but with potentially greater efficacy and fewer side effects.

DATA AND CODE AVAILABILITY

The data of this study are available under reasonable and ethically approved request to the corresponding authors. The R code of this study is available at https: //github. com/qluo2018/ImmunoASD4BTN.

REFERENCES

1. Lord C, Elsabbagh M, Baird G, Veenstra-Vanderweele J. Autism spectrum disorder. Lancet 2018; 392 (10146) : 508-520.

2. Hirvikoski T, Mittendorfer-Rutz E, Boman M, Larsson H, Lichtenstein P, S. Premature mortality in autism spectrum disorder. Br J Psychiatry 2016; 208 (3) : 232-238.

3. Phillips AG. Placing old wine into new bottles: Successful repurposing of bumetanide for treatment of autism spectrum disorder. Science Bulletin 2021.

4. James BJ, Gales MA, Gales BJ. Bumetanide for Autism Spectrum Disorder in Children: A Review of Randomized Controlled Trials. Ann Pharmacother 2019; 53 (5) : 537-544.

5. Zhang L, Huang C-C, Dai Y, Luo Q, Ji Y, Wang K et al. Symptom improvement in children with autism spectrum disorder following bumetanide administration is associated with decreased GABA/glutamate ratios. Transl Psychiatry 2020; 10 (1) : 9.

6. Lemonnier E, Degrez C, Phelep M, Tyzio R, Josse F, Grandgeorge M et al. A randomised controlled trial of bumetanide in the treatment of autism in children. Transl Psychiatry 2012; 2: e202.

7. Lemonnier E, Villeneuve N, Sonie S, Serret S, Rosier A, Roue M et al. Effects of bumetanide on neurobehavioral function in children and adolescents with autism spectrum disorders. Transl Psychiatry 2017; 7 (3) : e1056.

8. Dai Y, Zhang L, Yu J, Zhou X, He H, Ji Y et al. Improved symptoms following bumetanide treatment in children aged 3 to 6 years with autism spectrum disorder: a randomized, double-blind, placebo-controlled trial. Science Bulletin 2021.

9. Sprengers JJ, van Andel DM, Zuithoff NPA, Keijzer-Veen MG, Schulp AJA, Scheepers FE et al. Bumetanide for Core Symptoms of Autism Spectrum Disorder (BAMBI) : A Single Center, Double-Blinded, Participant-Randomized, Placebo-Controlled, Phase-2 Superiority Trial. Journal of the American Academy of Child & Adolescent Psychiatry 2021; 60 (7) : 865-876.

10. Coghlan S, Horder J, Inkster B, Mendez MA, Murphy DG, Nutt DJ. GABA system dysfunction in autism and related disorders: from synapse to symptoms. Neurosci Biobehav Rev 2012; 36 (9) : 2044-2055.

11. Chao H-T, Chen H, Samaco RC, Xue M, Chahrour M, Yoo J et al. Dysfunction in GABA signalling mediates autism-like stereotypies and Rett syndrome phenotypes. Nature 2010; 468 (7321) : 263-269.

12. Tyzio R, Nardou R, Ferrari DC, Tsintsadze T, Shahrokhi A, Eftekhari S et al. Oxytocin-mediated GABA inhibition during delivery attenuates autism pathogenesis in rodent offspring. Science 2014; 343 (6171) : 675-679.

13. Lemonnier E, Ben-Ari Y. The diuretic bumetanide decreases autistic behaviour in five infants treated during 3 months with no side effects. Acta Paediatr 2010; 99 (12) : 1885-1888.

14. Du L, Shan L, Wang B, Li H, Xu Z, Staal WG et al. A Pilot Study on the Combination of Applied Behavior Analysis and Bumetanide Treatment for Children with Autism. J Child Adolesc Psychopharmacol 2015; 25 (7) : 585-588.

15. Jones KA, Thomsen C. The role of the innate immune system in psychiatric disorders. Mol Cell Neurosci 2013; 53: 52-62.

16. Bertelsen LS, Eckmann L, Barrett KE. Prolonged interferon-gamma exposure decreases ion transport, NKCC1, and Na+-K+-ATPase expression in human intestinal xenografts in vivo. Am J Physiol Gastrointest Liver Physiol 2004; 286 (1) : G157-G165.

17. Pozzi D, Rasile M, Corradini I, Matteoli M. Environmental regulation of the chloride transporter KCC2: switching inflammation off to switch the GABA on? Transl Psychiatry 2020; 10 (1) : 349.

18. Corradini I, Focchi E, Rasile M, Morini R, Desiato G, Tomasoni R et al. Maternal Immune Activation Delays Excitatory-to-Inhibitory Gamma-Aminobutyric Acid Switch in Offspring. Biol Psychiatry 2018; 83 (8) : 680-691.

19. Meltzer A, Van de Water J. The Role of the Immune System in Autism Spectrum Disorder. Neuropsychopharmacology 2017; 42 (1) : 284-298.

20. Saghazadeh A, Ataeinia B, Keynejad K, Abdolalizadeh A, Hirbod-Mobarakeh A, Rezaei N. Anti-inflammatory cytokines in autism spectrum disorders: A systematic review and meta-analysis. Cytokine 2019; 123: 154740.

21. Kordulewska NK, Kostyra E, Piskorz-Ogórek K, Moszyńska M, A, Fiedorowicz E et al. Serum cytokine levels in children with spectrum autism disorder: Differences in pro-and anti-inflammatory balance. J Neuroimmunol 2019; 337: 577066.

22. Li X, Chauhan A, Sheikh AM, Patil S, Chauhan V, Li X-M et al. Elevated immune response in the brain of autistic patients. J Neuroimmunol 2009; 207 (1-2) : 111-116.

23. Vargas DL, Nascimbene C, Krishnan C, Zimmerman AW, Pardo CA. Neuroglial activation and neuroinflammation in the brain of patients with autism. Ann Neurol 2005; 57 (1) : 67-81.

24. Ashwood P, Krakowiak P, Hertz-Picciotto I, Hansen R, Pessah IN, Van de Water J. Associations of impaired behaviors with elevated plasma chemokines in autism spectrum disorders. J Neuroimmunol 2011; 232 (1-2) : 196-199.

25. Ashwood P, Enstrom A, Krakowiak P, Hertz-Picciotto I, Hansen RL, Croen LA et al. Decreased transforming growth factor beta1 in autism: a potential link between immune dysregulation and impairment in clinical behavioral outcomes. J Neuroimmunol 2008; 204 (1-2):149-153.

26. Ashwood P, Krakowiak P, Hertz-Picciotto I, Hansen R, Pessah I, Van de Water J. Elevated plasma cytokines in autism spectrum disorders provide evidence of immune dysfunction and are associated with impaired behavioral outcome. Brain Behav Immun 2011; 25 (1) : 40-45.

27. Masi A, Glozier N, Dale R, Guastella AJ. The Immune System, Cytokines, and Biomarkers in Autism Spectrum Disorder. Neurosci Bull 2017; 33 (2) : 194-204.

28. DiLalla DL, Rogers SJ. Domains of the childhood autism rating scale: Relevance for diagnosis and treatment. Journal of Autism and Developmental Disorders 1994; 24 (2) : 115-128.

29. Lord C, Risi S, Lambrecht L, Cook EH, Leventhal BL, DiLavore PC et al. The autism diagnostic observation schedule-generic: a standard measure of social and communication deficits associated with the spectrum of autism. Journal of autism and developmental disorders 2000; 30 (3) : 205-223.

30. Constantino JN, Gruber CP. Social responsiveness scale: SRS-2. Western Psychological Services Torrance, CA2012.

31. Johnson WE, Li C, Rabinovic A. Adjusting batch effects in microarray expression data using empirical Bayes methods. Biostatistics 2007; 8 (1) : 118-127.

32. Csala A, Voorbraak FPJM, Zwinderman AH, Hof MH. Sparse redundancy analysis of high-dimensional genetic and genomic data. Bioinformatics 2017; 33 (20) : 3228-3234.

33. Charrad M, Ghazzali N, Boiteau V, Niknafs A. NbClust: An R Package for Determining the Relevant Number of Clusters in a Data Set. Journal of Statistical Software 2014; 61: 1-36.

34. Kuhn M. Building Predictive Models in R Using the caret Package. Journal of Statistical Software; Vol 1, Issue 5 (2008) 2008.

35. Saghazadeh A, Ataeinia B, Keynejad K, Abdolalizadeh A, Hirbod-Mobarakeh A, Rezaei N. A meta-analysis of pro-inflammatory cytokines in autism spectrum disorders: Effects of age, gender, and latitude. J Psychiatr Res 2019; 115.

36. Molloy CA, Morrow AL, Meinzen-Derr J, Schleifer K, Dienger K, Manning-Courtney P et al. Elevated cytokine levels in children with autism spectrum disorder. J Neuroimmunol 2006; 172 (1-2) : 198-205.

37. Abdallah MW, Larsen N, Mortensen EL, AtladóttirB, Bonefeld-EC et al. Neonatal levels of cytokines and risk of autism spectrum disorders: an exploratory register-based historic birth cohort study utilizing the Danish Newborn Screening Biobank. J Neuroimmunol 2012; 252 (1-2) : 75-82.

38. Resta-Lenert S, Barrett KE. Probiotics and commensals reverse TNF-alpha-and IFN-gamma-induced dysfunction in human intestinal epithelial cells. Gastroenterology 2006; 130 (3) : 731-746.

39. Rivera C, Voipio J, Payne JA, Ruusuvuori E, Lahtinen H, Lamsa K et al. The K+/Cl-co-transporter KCC2 renders GABA hyperpolarizing during neuronal maturation. Nature 1999; 397(6716) : 251-255.

40. Schmidt J, Barthel K, Wrede A, Salajegheh M, M, Dalakas MC. Interrelation of inflammation and APP in sIBM: IL-1 beta induces accumulation of beta-amyloid in skeletal muscle. Brain 2008; 131 (Pt 5) : 1228-1240.

41. Kharod SC, Kang SK, Kadam SD. Off-Label Use of Bumetanide for Brain Disorders: An Overview. Front Neurosci 2019; 13: 310.

42. Hu D, Yu Z-L, Zhang Y, Han Y, Zhang W, Lu L et al. Bumetanide treatment during early development rescues maternal separation-induced susceptibility to stress. Scientific Reports 2017; 7 (1) : 11878.

43. Wang Z, Wang Y, Vilekar P, Yang S-P, Gupta M, Oh MI et al. Small molecule therapeutics for COVID-19: repurposing of inhaled furosemide. PeerJ 2020; 8: e9533-e9533.

44. Hung C-M, Peng C-K, Wu C-P, Huang K-L. Bumetanide attenuates acute lung injury by suppressing macrophage activation. Biochemical Pharmacology 2018; 156: 60-67.

45. Banks WA, Kastin AJ, Broadwell RD. Passage of cytokines across the blood-brain barrier. Neuroimmunomodulation 1995; 2 (4) : 241-248.

46. Loth E, Murphy DG, Spooren W. Defining Precision Medicine Approaches to Autism Spectrum Disorders: Concepts and Challenges. Front Psychiatry 2016; 7: 188-188.

47. Frye RE, Cakir J, Rose S, Delhey L, Bennuri SC, Tippett M et al. Prenatal air pollution influences neurodevelopment and behavior in autism spectrum disorder by modulating mitochondrial physiology. Mol Psychiatry 2021; 26 (5) : 1561-1577.

48. Masi A, Breen EJ, Alvares GA, Glozier N, Hickie IB, Hunt A et al. Cytokine levels and associations with symptom severity in male and female children with autism spectrum disorder. Mol Autism 2017; 8: 63-63.

Claims

A method of predicting a response of a subject afflicted with an autism spectrum disorder (ASD) to bumetanide, the method comprising:

obtaining characteristic information of the subject, wherein the characteristic information includes: (i) baseline expression levels of a set of cytokines in the subject; (ii) baseline behavioral performance of the subject; and (iii) clinical information of the subject, including gender and age;

predicting the response of the subject to bumetanide based on the characteristic information.
The method according to claim 1, wherein the set of cytokines include three or more cytokines selected from the group consisting of IL1β, IL6, IL8, IFNγ, TNFα, MCP1, Eotaxin, IL17, IL4, IL2Rα, MIG, MIP1β, IFNα2, SDF1α, IL16, LIF, TNFβ, MIF, RANTES, IL18, PDGFβ, IP10, IL13, MIP1α, GCSF, GROα, HGF, IL1α, SCF, TRAIL, MCSF, CTACK, IL7, IL9 and SCGFβ.
The method according to claim 2, wherein the set of cytokines include a group selected from the following:

(i) IL16, GROα and IL7;

(ii) IL16, GROα and TNFβ;

(iii) IL16, GROα, IL7, TNFβ and CTACK; and

(iv) IL16, GROα, IL7, TNFβ, LIF and MIF.
The method according to claim 2, wherein the set of cytokines include IL1β, IL6, IL8, IFNγ, TNFα, MCP1, Eotaxin, IL17, IL4, IL2Rα, MIG, MIP1β, IFNα2, SDF1α, IL16, LIF, TNFβ, MIF, RANTES, IL18, PDGFβ, IP10, IL13, MIP1α, GCSF, GROα, HGF, IL1α, SCF, TRAIL, MCSF, CTACK, IL7, IL9 and SCGFβ.
The method according to any one of claims 1-4, wherein the behavioral performance includes one or more scores of screening tools or diagnostic tools for ASD.
The method according to claim 5, wherein the behavioral performance includes 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12 or 13 scores selected from the group consisting of CAR_total, CAR_S, CAR_N, CAR_D, ADOS_S, ADOS_C, ADOS_P, ADOS_I, SRS_total, SRS_AWA, SRS_COG, SRS_COM, SRS_MOT and SRS_MANN.
The method according to claim 6, wherein the behavioral performance includes a group of scores selected from the following:

(i) ADOS_S, ADOS_C, ADOS_P, SRS_COM and CARS_D;

(ii) ADOS_S, ADOS_C, ADOS_P and CARS_total;

(iii) ADOS_S, ADOS_C, ADOS_P, SRS_COM, CARS_D, CARS_total and SRS_MOT;

(iv) ADOS_S, ADOS_C, ADOS_P, SRS_COM, CARS_D and CARS_total; and

(v) ADOS_S, ADOS_C, ADOS_P and CARS_total.
The method according to any one of claims 1-7, wherein the method comprises predicting the response of the subject to bumetanide based on the characteristic information by using a classifier.
The method according to claim 8, wherein the classifier is selected from the group consisting of Oblique Random Forest (ORF) , Partial Least Squares (PLS) , sparse Linear Discriminant Analysis (sLDA) , Neural Networks (NN) and Support Vector Machine (SVM) .
The method according to claim 8 or 9, wherein the classifier has been trained.
The method according to any one of claims 8-10, wherein:

the characteristic information includes: (i) baseline expression levels of IL16, GROαand IL7; (ii) scores of ADOS_S, ADOS_C, ADOS_P, SRS_COM and CARS_D; (iii) gender and age; and the classifier is support vector machine;

the characteristic information includes: (i) baseline expression levels of IL16, GROαand TNFβ; (ii) scores of ADOS_S, ADOS_C, ADOS_P and CARS_total; (iii) gender and age; and the classifier is partial Least Squares;

the characteristic information includes: (i) baseline expression levels of IL16, GROα, IL7, TNFβ and CTACK; (ii) scores of ADOS_S, ADOS_C, ADOS_P, SRS_COM, CARS_D, CARS_total and SRS_MOT; (iii) gender and age; and the classifier is Neural Networks;

the characteristic information includes: (i) baseline expression levels of IL16, GROα, IL7, TNFβ, LIF and MIF; (ii) scores of ADOS_S, ADOS_C, ADOS_P, SRS_COM, CARS_D and CARS_total; (iii) gender and age; and the classifier is sparse Linear Discriminant Analysis; or

the characteristic information includes: (i) baseline expression levels of IL16, GROαand TNFβ; (ii) scores of ADOS_S, ADOS_C, ADOS_P and CARS_total; (iii) gender and age; and the classifier is Oblique Random Forest.
The method according to any one of claims 1-11, wherein the method comprises obtaining characteristic information of the subject by measuring the expression levels of the cytokines in a sample from the subject.
The method according to claim 12, wherein the sample is plasma.
A method of predicting a response of a subject afflicted with an autism spectrum disorder (ASD) to bumetanide, the method comprising:

training a classifier with a training data set containing characteristic information of a plurality of individuals with ASD and the response of the individuals to bumetanide that has been determined select a subset of relevant characteristic information, wherein the characteristic information includes: (i) baseline expression levels of cytokines of IL1β, IL6, IL8, IFNγ, TNFα, MCP1, Eotaxin, IL17, IL4, IL2Rα, MIG, MIP1β, IFNα2, SDF1α, IL16, LIF, TNFβ, MIF, RANTES, IL18, PDGFβ, IP10, IL13, MIP1α, GCSF, GROα, HGF, IL1α, SCF, TRAIL, MCSF, CTACK, IL7, IL9 and SCGFβ; (ii) baseline scores of CAR_total, CAR_S, CAR_N, CAR_D, ADOS_S, ADOS_C, ADOS_P, ADOS_I, SRS_total, SRS_AWA, SRS_COG, SRS_COM, SRS_MOT and SRS_MANN; and (iii) gender and age;

predicting the response of the subject to bumetanide based on the subset of relevant characteristic information of the subject.
The method according to claim 14, wherein the classifier is selected from the group consisting of Oblique Random Forest (ORF) , Partial Least Squares (PLS) , sparse Linear Discriminant Analysis (sLDA) , Neural Networks (NN) and Support Vector Machine (SVM) .
The method according to claim 14 or 15, wherein the expression levels of the cytokines are obtained by measuring expression level of the cytokines in a sample from the subject.
The method according to claim 16, wherein the sample is plasma.
A method of treating autism spectrum disorder (ASD) in a subject, the method comprising:

predicting a response of a subject afflicted with ASD to bumetanide using the method according to any one of claims 1-13;

administering an effective amount of bumetanide to the subject that is identified to have response to bumetanide.
A prediction device comprising:

an input module which is configured to receive characteristic information of a subject with autism spectrum disorder (ASD) , wherein the characteristic information includes: (i) baseline expression levels of a set of cytokines in the subject; (ii) baseline behavioral performance of the subject; and (iii) clinical information of the subject, including gender and age;

a classifying module which is configured to comprise a classifier, wherein the classifier can predict the response of the subject to bumetanide using a classifier based on the characteristic information.
The prediction device according to claim 19, further comprises a training module which is configured to train the classifier using a training data set.
A computer readable medium comprising computer executable instructions recorded thereon for performing the operation comprising:

receiving characteristic information of a subject with autism spectrum disorder (ASD) , wherein the characteristic information includes: (i) baseline expression levels of a set of cytokines in the subject; (ii) baseline behavioral performance of the subject; and (iii) clinical information of the subject, including gender and age;

predicting the response of the subject to bumetanide using a classifier algorithm based on the characteristic information.
The computer readable medium according to claim 21, wherein the operation further comprises training the classifier using a training data set.
The prediction device or the computer readable medium according to any one of claims 19-22, wherein the set of cytokines include three or more cytokines selected from the group consisting of IL1β, IL6, IL8, IFNγ, TNFα, MCP1, Eotaxin, IL17, IL4, IL2Rα, MIG, MIP1β, IFNα2, SDF1α, IL16, LIF, TNFβ, MIF, RANTES, IL18, PDGFβ, IP10, IL13, MIP1α, GCSF, GROα, HGF, IL1α, SCF, TRAIL, MCSF, CTACK, IL7, IL9 and SCGFβ.
The prediction device or the computer readable medium according to any one of claims 19-23, wherein the set of cytokines include a group selected from the following:

(i) IL16, GROα and IL7;

(ii) IL16, GROα and TNFβ;

(iii) IL16, GROα, IL7, TNFβ and CTACK; and

(iv) IL16, GROα, IL7, TNFβ, LIF and MIF.
The prediction device or the computer readable medium according to any one of claims 19-24, wherein the set of cytokines include IL1β, IL6, IL8, IFNγ, TNFα, MCP1, Eotaxin, IL17, IL4, IL2Rα, MIG, MIP1β, IFNα2, SDF1α, IL16, LIF, TNFβ, MIF, RANTES, IL18, PDGFβ, IP10, IL13, MIP1α, GCSF, GROα, HGF, IL1α, SCF, TRAIL, MCSF, CTACK, IL7, IL9 and SCGFβ.
The prediction device or the computer readable medium according to any one of claims 19-25, wherein the behavioral performance includes 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12 or 13 scores selected from the group consisting of CAR_total, CAR_S, CAR_N, CAR_D, ADOS_S, ADOS_C, ADOS_P, ADOS_I, SRS_total, SRS_AWA, SRS_COG, SRS_COM, SRS_MOT and SRS_MANN.
The prediction device or the computer readable medium according to any one of claims 19-26, wherein the behavioral performance includes a group of scores selected from the following:

(i) ADOS_S, ADOS_C, ADOS_P, SRS_COM and CARS_D;

(ii) ADOS_S, ADOS_C, ADOS_P and CARS_total;

(iii) ADOS_S, ADOS_C, ADOS_P, SRS_COM, CARS_D, CARS_total and SRS_MOT;

(iv) ADOS_S, ADOS_C, ADOS_P, SRS_COM, CARS_D and CARS_total; and

(v) ADOS_S, ADOS_C, ADOS_P and CARS_total.
The prediction device or the computer readable medium according to any one of claims 19-27, wherein the classifier is selected from the group consisting of Oblique Random Forest (ORF) , Partial Least Squares (PLS) , sparse Linear Discriminant Analysis (sLDA) , Neural Networks (NN) and Support Vector Machine (SVM) .
The prediction device or the computer readable medium according to any one of claims 19-28, wherein the classifier has been trained.
The prediction device or the computer readable medium according to any one of claims 19-29, wherein:

the characteristic information includes: (i) baseline expression levels of IL16, GROαand IL7; (ii) scores of ADOS_S, ADOS_C, ADOS_P, SRS_COM and CARS_D; (iii) gender and age; and the classifier is support vector machine;

the characteristic information includes: (i) baseline expression levels of IL16, GROαand TNFβ; (ii) scores of ADOS_S, ADOS_C, ADOS_P and CARS_total; (iii) gender and age; and the classifier is partial Least Squares;

the characteristic information includes: (i) baseline expression levels of IL16, GROα, IL7, TNFβ and CTACK; (ii) scores of ADOS_S, ADOS_C, ADOS_P, SRS_COM, CARS_D, CARS_total and SRS_MOT; (iii) gender and age; and the classifier is Neural Networks;

the characteristic information includes: (i) baseline expression levels of IL16, GROα, IL7, TNFβ, LIF and MIF; (ii) scores of ADOS_S, ADOS_C, ADOS_P, SRS_COM, CARS_D and CARS_total; (iii) gender and age; and the classifier is sparse Linear Discriminant Analysis; or

the characteristic information includes: (i) baseline expression levels of IL16, GROαand TNFβ; (ii) scores of ADOS_S, ADOS_C, ADOS_P and CARS_total; (iii) gender and age; and the classifier is Oblique Random Forest.
A kit for predicting a response to bumetanide in a subject with autism spectrum disorder (ASD) , the kit comprising agents for measuring expression levels of a set of cytokines defined in any one of claims 1-13 and instructions describing the predicting method according to any one of claims 1-13.
A kit for treating autism spectrum disorder (ASD) in a subject, the kit comprising:

agents for measuring expression levels of a set of cytokines defined in any one of claims 1-13;

bumetanide; and

instructions describing the predicting method according to any one of claims 1-13 and that an effective amount of bumetanide is administered to the subject that is identified to have response to bumetanide.
The characteristic information defined in any one of claims 1-13 for use in predicting a response to bumetanide in a subject with autism spectrum disorder (ASD) .
Use of the characteristic information defined in any one of claims 1-13 in the preparation of a set of features for predicting the response to bumetanide in a subject with autism spectrum disorder (ASD) .
Use of agents for measuring the expression levels of the set of cytokines defined in any one of claims 1-13 in preparation of an agent or a kit for predicting a response to bumetanide in a subject with autism spectrum disorder (ASD) , wherein the prediction is performed through the prediction method according to any one of claims 1-13.
Use of the characteristic information defined in any one of claims 1-13 in built a prediction model for predicting the response to bumetanide in a subject with autism spectrum disorder (ASD) .