Leveraging Gene Expression Data for Drug Repurposing in Schizophrenia: A Signature Reversion Approach

Abstract

Background/Objectives: Despite continuous pharmacological advances, the treatment of schizophrenia remains challenging, and suboptimal outcomes are still too frequent. There are currently limited new approved drugs without resistance. Methods: For this reason, drug repurposing presents a promising solution for identifying existing drugs with therapeutic effects for schizophrenia. In this study, we provide a workflow of signature-based drug repurposing methodology. We initially utilized a dataset from Gene Expression Omnibus which consists of RNA sequence data from blood-derived leukocyte samples from individuals with schizophrenia and control subjects, and conducted an analysis. Results: This analysis identified 1205 statistically significant differentially expressed genes, of which 150 upregulated and 150 downregulated genes were used in the CMap and L1000CDS2 tools. Then, each database generated a list of potential compounds that could reverse the disease’s signature and potentially have therapeutic effects for schizophrenia. Subsequently, the compounds associated with the disease, as identified in the research, were chemically clustered, and then their modes of action were predicted. In the last stage, we conducted a literature review to evaluate the relationship of these modes of action with the disease. Conclusions: This systematic analysis provided a list of potential drugs for schizophrenia treatment so that their efficacy can be evaluated in the wet-lab experiments, which is the next stage of drug repurposing.

1. Introduction

Schizophrenia is a chronic, complex and polygenic neuropsychiatric disease that typically manifests in adulthood. It is characterized by a variety of symptoms, and it affects a person’s thoughts, feelings and behavior [1]. The disorder has different impact on the patient’s life; some can function normally while others exhibit severe disability [2]. Despite affecting approximately 1% of the global population, it causes a significant health burden, accounting for 1.9–2.8% of total years lived with disability and reducing life expectancy by 10–20 years. However, the primary medical treatment remains unchanged, involving antipsychotic drugs that typically achieve partial symptom remission. Only 15–25% of patients become completely symptom-free, and the treatment often causes adverse side effects such as weight gain, metabolic disturbances, oversedation, and extrapyramidal symptoms [3].

In recent years, there have been some promising advances in pharmacotherapy for schizophrenia. In 2021, the Food and Drug Administration (FDA) confirmed the combination of olanzapine/samidorphan (OLZ/SAM) as an improved and better-tolerated therapy for schizophrenia and bipolar disease. Olanzapine is a well-known atypical antipsychotic commonly used for these disorders, while samidorphan is a newer medication that blocks the μ-opioid receptor, a key regulator of mood and behavior [4]. The FDA also recently approved Cobenfy, marking a significant advancement in schizophrenia therapy. This state-of-the-art antipsychotic drug functions through a novel mechanism, targeting cholinergic receptors instead of the typical dopamine receptors. It consists of xanomeline, an oral muscarinic cholinergic receptor agonist combined with tropsium chloride, a peripheral muscarinic receptor antagonist. Xanomeline, initially studied for Alzheimer’s disease, effectively modulates brain muscarinic receptors, particularly the M1 and M4 subtypes, which are crucial in regulating cognitive and psychotic symptoms. Tropsium, firstly used for overactive bladder, acts as a peripheral antagonist to minimize the xanomeline’s adverse effects, while preserving its central effects. This synthesis fulfills the growing need for alternative treatments that can overcome the limitations of current medications [5].

The development of new treatments remains challenging. The heterogeneity of schizophrenia has resulted in a limited understanding of its pathophysiology and insufficient identification of reliable biomarkers and molecular targets of current drugs. These challenges, combined with the high failure rate of late-stage clinical trials and the prolonged development timelines for central nervous system drugs, have contributed to the decline in the discovery of new treatments for this disorder [3,6]. Research and development in the field of psychiatric medication usually progresses more slowly compared to other areas of drug development. In fact, according to the Pharmaceutical Research and Manufacturers of America only 240 drugs are in development for mental health as of 2011, whereas more than 3000 were being developed for cancer and 750 for infectious diseases [4].

Conventional drug discovery has been a time-consuming and expensive process, while yearly 90% of drugs do not receive FDA approval because of issues such as low efficacy, toxicity and adverse reactions [7,8]. Consequently, there is urgent need for innovative strategies that can speed up the successful creation of new medications [8]. Drug repurposing, also known as drug repositioning or reprofiling, offers a promising solution to these difficulties. This approach identifies new therapeutic uses for existing drugs, including those that are already approved for other conditions, or those still undergoing clinical trials [8]. It can be broadly classified into two main strategies: the serendipity and hypothesis driven approach. Drug discovery through serendipity is a well-established concept, where treatments are found by chance or accident. The other approach involves both computational and experimental methods, with the latter focusing on screening existing drugs to discover new pharmacological uses through experimental assays [9,10].

Computational drug repurposing enables the discovery of potential drugs with use of molecular interactions between compounds and protein receptors [10]. This approach is classified into four subcategories: target-based approach, drug-centric approach, disease/therapy-based approach and signature-based approach [9]. The signature-based methodology discovers new uses for existing drugs through genes or protein expression patterns (signatures) from omics data. It identifies new therapeutic targets by analyzing these patterns across various diseases and treatments. The main principle is that a drug may be effective if it can revert the expression levels of the genes to their normal state. The advancements in sequencing technology and other biological methods have made this approach more precise and targeted [11]. In this study, we developed a computational drug repurposing pipeline using a signature-based approach. This method uses gene expression patterns (signatures) derived from disease-related omics data to discover new therapeutic uses for existing drugs. A compound is considered potentially effective against schizophrenia if it can reverse the disease-associated gene expression signature. The outcome of this analysis is a curated list of candidate drugs, generated based on their predicted modes of action and established associations with schizophrenia.

2. Results

2.1. Schizophrenia’s Signature

In our investigation, differential gene expression analysis identified 1205 DEGs, including 623 upregulated and 582 downregulated genes. To meet the input requirements of the tools utilized in our study, we selected the top 150 upregulated and 150 downregulated genes from the disease signature, as presented in

Table A1. List of upregulated genes in schizophrenia from GEO dataset.

Upregulated Genes
Symbol	logFC	p.Value	adj.P.Val	Symbol	logFC	p.Value	adj.P.Val
MGST1	1.196	0	0	DUSP1	1.023	0.001	0.004
HEBP1	1.271	0	0.001	ETF1	1.045	0	0
SLC38A5	1.326	0.001	0.003	EGR1	2.009	0.002	0.009
EHD2	1.517	0.004	0.014	GJB6	3.558	0	0
HMGB3	1.044	0.001	0.006	SPINK4	1.216	0.012	0.036
TG	1.003	0.001	0.003	PLAU	2.572	0	0.001
HSPA5	1.404	0	0	G0S2	3.038	0.002	0.009
DKK3	1.19	0.002	0.008	SLPI	1.598	0	0
RRP12	1.178	0	0	SNAI1	1.529	0.003	0.01
PSD	1.041	0	0.001	STX16	1.375	0	0
SPAG4	1.021	0.001	0.004	SEMG1	1.789	0	0.001
GLTSCR1	1.099	0	0	TREM1	1.091	0.002	0.008
GSTO2	1.365	0	0	SOX4	1.206	0	0
KLF6	1.131	0	0	FOSB	1.712	0.004	0.014
LAPTM4A	1.389	0	0	TGM3	1.494	0.001	0.005
BCL3	1.1	0	0	GZF1	1.022	0	0.001
MAOB	1.643	0.002	0.009	ID1	4.194	0	0
JMJD6	1.249	0	0	SBDS	1.047	0	0
PABPC1	1.249	0	0	HSPA2	1.15	0.013	0.037
P4HA2	1.431	0	0	HRASLS	1.094	0.003	0.012
RAB7A	1.142	0	0	EMC6	1.025	0	0.001
RAP1GAP	1.243	0.016	0.043	LRFN1	1.207	0	0
EDN1	1.068	0.01	0.03	ZFP36	1.254	0	0
TP53INP2	1.376	0	0	ADORA2A	1.632	0	0.002
CD82	1.038	0	0	DNAJB9	1.3	0	0
IGSF9	1.04	0.005	0.016	ALDH1A2	1.217	0	0
B4GALT1	1.027	0	0	LOXL1	1.727	0	0.001
PPP1R15A	1.207	0	0	FGF13	1.149	0.009	0.029
ADAMTS2	3.645	0	0	KLF16	1.025	0	0
RFX2	1.04	0	0.001	LPPR3	1.933	0	0
RPH3A	1.724	0	0.001	CBFA2T3	1.337	0	0
FUS	1.503	0	0	PVRL2	1.123	0.001	0.004
CCNK	1.313	0	0	GADD45G	1.489	0	0.002
TREM2	1.572	0	0.001	TNNT3	1.443	0	0
HNRNPH3	1.023	0	0	COL5A1	1.279	0.008	0.026
GADD45B	1.009	0	0	PDLIM4	2.101	0	0
DDT	1.373	0	0.001	RPL27	1.288	0	0
SEC14L3	1.07	0.007	0.022	LGALS3	1.05	0	0.001
UPB1	1.051	0	0	EIF5A	1.12	0	0
NFKBIA	1.626	0	0	KDM6B	1.562	0	0
SIRPB1	1.021	0	0.001	MTSS1L	1.06	0	0.001
CHMP4B	1.158	0	0	NXT1	1.136	0	0.001
EMD	1.159	0	0	CCDC59	1.091	0	0
USB1	1.191	0	0	PTGFRN	1.242	0	0
TULP2	2.028	0	0	ECHDC3	1.934	0	0
RPS16	1.658	0	0	AKIRIN2	1.175	0	0
CCDC9	1.194	0	0	CD63	1.228	0	0
SCN1B	1.03	0	0	CKAP4	1.104	0	0
ERF	1.085	0	0	NACAD	1.056	0.002	0.009
NAMPT	1.251	0	0	SCN2A	1.327	0.005	0.018
UBE2R2	1.098	0	0	ANP32B	1.273	0	0
UBE2S	1.02	0	0	HIST1H2AB	1.442	0	0.001
RPL28	1.252	0	0	GCM1	1.347	0	0
AREG	2.226	0.006	0.019	SLCO5A1	1.134	0	0.001
MAPK10	1.294	0.001	0.004	SDCBP	1.031	0	0
SH3D19	1.612	0	0.001	RPLP1	1.066	0	0
SLC1A2	3.11	0	0	SMAD6	1.561	0.016	0.043
POU2AF1	1.137	0	0.001	HNRNPD	1.079	0	0
MGP	1.468	0.001	0.006	CCNG2	1.042	0	0
PHACTR1	1.488	0	0	ANXA3	1.072	0.001	0.006
CD83	1.465	0.001	0.004	GPR84	1.414	0.013	0.037
PTP4A1	1.409	0	0	RAB20	1.17	0.002	0.008
VNN1	1.018	0	0.001	REM2	1.083	0	0
CCR6	1	0.007	0.023	BCL2A1	1.039	0.005	0.018
HBEGF	1.507	0.005	0.018	RLBP1	1.572	0	0.001
BCL6	1.059	0	0.001	MAP1LC3B	1.041	0	0
EIF1B	1.51	0	0	OSGIN1	1.031	0.003	0.012
IL1R2	1.354	0	0	TOB1	1.633	0	0
IL1RL2	2.118	0.002	0.007	ZMYND15	1.707	0	0
PAPPA2	1.611	0.007	0.022	CSNK1D	1.674	0	0
TNFAIP3	1.21	0.005	0.018	NFIC	1.07	0	0
UBE2B	1.039	0	0	CTD-3222D19.2	1.07	0.006	0.021
KLF9	1.095	0	0	DMRTC2	1.92	0	0.001
NR4A3	2.491	0.001	0.003	SIK1	1.304	0.006	0.02
YPEL5	1.041	0	0	NR1I3	1.226	0	0

and

Table A2. List of downregulated genes in schizophrenia from GEO dataset.

Downregulated Genes
Symbol	logFC	p.Value	adj.P.Val	Symbol	logFC	p.Value	adj.P.Val
PDK4	−1.099	0.002	0.007	WNT3	−1.191	0.001	0.003
ZMYND10	−1.673	0	0	CCL2	−2.109	0.017	0.045
MEOX1	−1.301	0.006	0.021	CACNG1	−1.88	0.002	0.009
TTC22	−1.172	0	0	GNRHR	−1.511	0.001	0.005
CACNA2D2	−1.031	0	0.001	B3GAT1	−1.682	0	0
OSBPL5	−1.213	0	0	SNX15	−2.006	0	0
CPS1	−1.322	0.007	0.024	UPK2	−1.117	0.007	0.023
HSD17B6	−1.319	0.017	0.046	CALCA	−1.538	0.01	0.03
INSRR	−1.333	0.015	0.041	ENDOU	−1.119	0.006	0.02
SLC18A1	−2.203	0.002	0.007	FANCE	−1.257	0	0
RTN4R	−1.629	0	0.002	ENPP5	−1.74	0	0
USP28	−1.127	0	0	NUDT12	−1.076	0.003	0.011
EPN3	−1.135	0.007	0.022	KIF20A	−1.144	0.001	0.003
SLC4A8	−1.014	0	0	PDGFRB	−2.167	0	0
LAMC3	−2.011	0.005	0.017	SERPINI2	−1.274	0.001	0.003
TNIP3	−1.042	0.016	0.044	HHLA2	−1.723	0.001	0.004
MCOLN3	−1.568	0	0	C3orf52	−1.319	0.001	0.004
PCGF2	−1.287	0.003	0.013	RTKN	−1.459	0	0
ZNF112	−2.017	0.004	0.014	ZAP70	−1.003	0	0
TSPAN32	−1.138	0	0	PCSK4	−1.039	0.001	0.006
NGFR	−1.833	0.002	0.007	GNLY	−1.433	0	0
CDON	−1.2	0.007	0.023	LCT	−1.212	0.007	0.022
FGFR2	−1.078	0.018	0.05	GRIN3B	−1.005	0.006	0.02
LLGL2	−1.343	0	0	DHCR24	−1.176	0	0
TBX21	−1.247	0	0	C1orf21	−1.194	0	0
TTC38	−1.139	0	0	OLFML3	−1.526	0.001	0.003
UBE2T	−1.676	0	0	CD160	−1.149	0	0.001
AMPH	−1.657	0	0	TSPAN1	−1.697	0	0.001
DDX43	−1.211	0.008	0.024	FASLG	−1.941	0	0
JMJD4	−1.006	0	0	ATP10B	−1.61	0	0.001
ATP8B1	−1.103	0.017	0.046	FILIP1	−1.379	0.01	0.031
FAT1	−1.056	0.013	0.038	RARRES1	−1.353	0	0.001
CETP	−1.042	0.001	0.005	PROX2	−1.202	0	0.001
C3orf18	−1.205	0	0	C10orf95	−1.431	0.016	0.044
SIRT4	−2.144	0	0	HOXB3	−1.367	0	0
FLT3LG	−1.098	0	0.002	SMAD9	−2.201	0.001	0.006
SLC26A3	−1.95	0.002	0.008	EPX	−1.061	0.007	0.023
IL5RA	−1.305	0.001	0.003	TAS2R10	−1.203	0.017	0.045
TF	−1.533	0.016	0.044	PRB2	−1.329	0.01	0.03
RGS17	−1.456	0.002	0.009	FABP3	−1.111	0	0
ANGPT2	−1.032	0.001	0.005	ZSCAN20	−1.293	0	0
TGM1	−1.081	0	0	WIPF3	−1.792	0.002	0.007
SEMA4G	−1.509	0	0	SLC12A5	−1.556	0	0.001
ARVCF	−1.791	0	0	BTN1A1	−1.719	0.006	0.019
MMP11	−1.461	0.009	0.028	BFSP1	−2.255	0	0
P2RX6	−1.23	0.007	0.024	LRRN4	−1.638	0.001	0.006
SOX10	−1.171	0.008	0.026	TMEM74B	−1.452	0.001	0.005
APOL4	−1.35	0.005	0.018	PROZ	−2.243	0	0.002
MLC1	−1.435	0	0	TAS2R3	−2.058	0	0.001
GZMH	−1.4	0	0	TAS2R4	−2.196	0.002	0.009
GZMB	−1.507	0	0	TAS2R5	−1.047	0.007	0.024
ASB2	−1.102	0.001	0.004	CRYGN	−2.107	0.001	0.005
NINL	−1.862	0.001	0.004	ZNF835	−1.669	0	0.002
TSNAXIP1	−1.596	0.001	0.005	CDHR3	−1.231	0	0
SMPD3	−1.116	0.001	0.004	USP6	−1.127	0.004	0.016
METRN	−1.041	0	0	MRM1	−1.213	0	0
SYT17	−1.054	0.012	0.034	ZSCAN10	−1.528	0.016	0.043
SCG3	−1.555	0.005	0.017	USHBP1	−1.135	0.004	0.015
TRPA1	−1.639	0	0.002	ACSBG2	−1.941	0	0
CALB1	−1.14	0.018	0.049	PKDREJ	−1.142	0.001	0.006
POP1	−1.077	0	0	ULBP3	−1.701	0.002	0.007
PYCRL	−1.046	0	0	IDO1	−2.434	0	0.001
LHB	−1.933	0	0.001	CA6	−1.099	0.01	0.03
SARS2	−1.027	0.003	0.012	CKMT2	−1.889	0	0.001
OLFM2	−1.003	0.007	0.023	ZNF132	−1.328	0	0
FSD1	−1.76	0	0.002	SLC52A1	−1.598	0.007	0.022
SIGLEC8	−1.726	0.001	0.004	GUCY2D	−1.66	0	0.002
LIM2	−1.644	0.001	0.005	ACY3	−1.629	0.003	0.012
NKG7	−1.106	0	0	MMACHC	−1.222	0	0
ZNF175	−1.023	0	0	ZSWIM3	−1.261	0	0.001
DFNA5	−1.091	0.001	0.003	LGR6	−1.323	0	0.001
AGFG2	−1.047	0	0	SFTPD	−1.996	0	0.002
SFRP4	−1.687	0.002	0.007	ADAM20	−1.173	0.003	0.01
PTGDS	−1.377	0	0	KLRD1	−1.178	0	0
RGS9	−1.186	0	0.002	HNF1A	−1.419	0	0

which can be found in the Appendix A.

2.2. Query Results

2.2.1. CMap Tool

The outcome of querying the previously mentioned schizophrenia signature in this tool was a list of 115 unique compounds with association score below −60 (

Table A3. CMap Results: List of potential compounds.

Drug Name	Connectivity Score	Drug Name	Connectivity Score
PIK-75	−96.86	PD-166793	−72.03
triptolide	−94.35	AT-7519	−71.73
ZG-10	−94	staurosporine	−71.51
ascorbic-acid	−93.85	myriocin	−71.48
PI-103	−93.67	bisindolylmaleimide-ix	−71.44
chromomycin-a3	−93.52	PPT	−70.78
dactinomycin	−93.15	flufenamic-acid	−70.32
KIN001-242	−92.87	carbetocin	−70.27
idarubicin	−92.68	azithromycin	−70.23
CS-110266	−92.62	MEK1-2-inhibitor	−70.17
TWS-119	−92.57	chloramphenicol	−69.88
pirarubicin	−91.86	parbendazole	−69.79
daunorubicin	−90.64	selumetinib	−69.46
ER-27319	−90.61	L-655240	−69.31
tivozanib	−90.52	5-iodotubercidin	−69.13
pazopanib	−90.29	minoxidil	−69.12
pidorubicine	−89.47	hexylcaine	−69.03
9-methyl-5H-6-thia-4,5-diaza-chrysene-6,6-dioxide	−88.83	byssochlamic-acid	−68.83
saracatinib	−88.72	methoxsalen	−68.82
H-7	−88.6	rifampicin	−68.07
alvocidib	−87.49	metformin	−67.58
CGP-60474	−85.21	BRL-37344	−67.57
veliparib	−84.65	kavain	−67.5
GDC-0941	−83.94	ampicillin	−66.96
PP-2	−82.97	FR-122047	−66.61
AZD-7762	−82.58	mestranol	−66.44
tofacitinib	−82.49	ochratoxin-a	−66.43
nornicotine	−82.4	fluoxetine	−66.41
tianeptine	−82.13	KU-0063794	−66.29
tramadol	−81.78	cobalt(II)-chloride	−65.87
erismodegib	−81.69	maackiain	−65.43
fasudil	−80.93	amitriptyline	−65.12
fostamatinib	−80.81	doxorubicin	−64.92
benproperine	−80.22	aminogenistein	−64.54
WH-4023	−79.94	tenofovir	−64.24
acamprosate	−79.27	PP-30	−64.09
etilefrine	−78.89	mofezolac	−63.55
cosmosiin	−78.71	dictamnine	−63.5
DCPIB	−78.71	ML-7	−63.4
fluspirilene	−77.9	AS-601245	−63.39
W-12	−77.66	AM-580	−63.24
RO-90-7501	−77.6	bifemelane	−62.92
cefoxitin	−77.59	OSI-027	−62.89
lestaurtinib	−77.35	benzatropine	−62.73
MLN-8054	−77.14	n-formylmethionylalanine	−62.41
dasatinib	−76.83	caffeine	−62.15
solanine	−76.15	LY-364947	−61.91
erythrosine	−75.66	canrenoic-acid	−61.84
clopidogrel	−75.19	SB-205607	−61.71
lamivudine	−75.11	tipifarnib	−61.56
AZ-628	−73.85	amoxicillin	−61.28
bongkrek-acid	−73.81	nitrofural	−61.08
didanosine	−73.69	JNJ-38877605	−60.83
carbidopa	−73.57	olopatadine	−60.81
EI-231	−73.47	mitoxantrone	−60.73
MK-2206	−73.33	piretanide	−60.61
TAK-715	−73.06	SB-202190	−60.01
ENMD-2076	−72.41

).

2.2.2. L1000CD²

The outcome of the L1000CDS² tool was a list of 50 compounds with information about their signature. The overlap score of each drug refers to the percentage of the input gene signature genes that get reversed by said drug. After deduplication, 20 unique drugs were retained (

Table A4. L1000CD² Results: List of potential compounds.

Drug Name	Overlap Score
CGP-60474	0.0729
alvocidib	0.0648
BMS-387032	0.0567
A443654	0.0526
radicicol	0.0486
PHA-793887	0.0486
daunorubicin	0.0445
epirubicin	0.0445
triptolide	0.0445
geldanamycin	0.0445
ER 27319 maleate	0.0405
KN-93	0.0405
16-hydroxytriptolide	0.0405
evodiamine	0.0405
parbendazole	0.0405
656402-250MG	0.0405
linifanib	0.0405
AT-7519	0.0405
mitoxantrone	0.0405
JNK-9L	0.0405

).

2.3. Prescreening of Results

The initial queries yielded a total of 135 compounds (115 + 20). To refine this collection and focus on candidates with greater potential relevance, we performed a bibliographic search targeting compounds linked—either directly or indirectly—to mechanisms associated with schizophrenia. From this screening, 22 compounds were selected for further analysis.

H7 is broad-spectrum protein kinase inhibitor, and can be connected to schizophrenia because reduced kinase activity, including pathways regulating NMDA receptor function, has been implicated in the disorder’s glutamatergic deficits [12].

Fasudil, the results from [13] suggest that fasudil has antipsychotic-like effects on the MK-801-treated pharmacological mouse schizophrenia model.

Saracatinib and PP-2 (Src kinase inhibitor PP2), according to [14], they inhibit Src kinase, whose reduced activity has been implicated in NMDA receptor hypoactivity observed in the disorder.

Fostamatinib may become a drug choice for schizophrenic patients that develop immune thrombocytopenic purpura (ITP) after treatment [15].

Tofacitinib is a JAK inhibitor, may be relevant for schizophrenia by modulating the JAK/STAT pathway, which influences neurogenesis, synaptic plasticity, and inflammation-processes disrupted in the disorder [16].

Dasatinib, as part of senolytic treatment with quercetin, may help schizophrenia by targeting cellular senescence pathways linked to accelerated aging and reduced health span in the disorder [17].

Caffeine has been shown to modulate some schizophrenia-induced symptoms, particularly negative and cognitive ones, potentially through effects on the gut–brain axis, immune and endocrine systems, and individual metabolic variability [18].

TWS-119, a GSK-3β inhibitor, addresses pathways implicated in schizophrenia by regulating neuronal differentiation and synaptic function, which are disrupted in the disorder and contribute to cognitive deficits [19].

Tianeptine has been shown as a potentially useful as a treatment option for schizophrenia [20].

KN-93, a CaMKII inhibitor, influences schizophrenia-related pathways by reducing D2 high receptor states in the striatum, a mechanism that may counteract dysregulated dopamine signaling implicated in the disorder [21].

PIK-75, a PI3K inhibitor, targets dysregulated PI3K signaling implicated in schizophrenia, potentially addressing synaptic dysfunction that contributes to cognitive and behavioral symptoms of the disorder [22].

Kavain, a bioactive component of kava, may influence schizophrenia by modulating neurotransmission through multiple mechanisms, including enhancing GABA_A receptor activity, inhibiting glutamate release, and affecting norepinephrine signaling, which together could help alleviate symptoms of the disorder [23].

Amitriptyline use in individuals with schizophrenia is associated with a lower risk of mortality compared to non-use, suggesting potential benefits in this population [24].

Alvocidib: in multiple sclerosis and schizophrenia, CDK inhibitors such as this drug have been shown to reduce working memory problems by promoting remyelination by inhibiting microglia activation [25].

Fluoxetine may improve schizophrenia-related deficits by enhancing brain-derived neurotrophic factor (BDNF) levels, which helps attenuate sensorimotor gating impairments and provides neuroprotective effects in the disorder [26].

Ampicillin potentially influences schizophrenia-related pathways by altering gut microbiota, which can affect immune signaling, neurotransmitter balance, and behavior, potentially contributing to psychiatric symptoms [27].

Nornicotine could be relevant to schizophrenia by activating α4β2 and α7 nicotinic acetylcholine receptors, which are involved in cognitive processes and sensory gating often disrupted in the disorder [28].

Fluspirilene is one of the already approved drugs for the treatment of schizophrenia [29].

Metformin has shown potential in improving cognitive and other symptom dimensions in patients with schizophrenia, as suggested by current evidence [30].

Ascorbic acid has the potential to improve cognitive and other symptom dimensions in patients with schizophrenia, according to [31].

Chromomycin A3 may be relevant to schizophrenia research by affecting Sp1-dependent transcription, a pathway that can influence neuronal survival and function [32].

2.4. Clustering Drug List

The 22 candidate compounds identified in the prescreening step were used as input for chemical clustering analysis with ChemBioServer 2.0. Clustering was performed in an unsupervised manner based on pairwise Tanimoto distances, applying hierarchical clustering with the Ward linkage method and a threshold of 1. This process grouped the compounds according to structural similarities and differences, offering insights into their molecular relationships and potential mechanisms of action.

The clustering analysis offers practical utility in several ways. For example, within the clusters, fluspirilene, an approved antipsychotic [29], is in the same group with other compounds that share molecular similarities. If a patient experiences adverse effects from this drug, alternative drugs from the same cluster could be considered. Similarly, if a drug previously used for a particular disease proves ineffective in a patient, compounds from a different cluster may be explored to target alternative mechanisms. Finally, in the context of drug repurposing studies, selecting at least one compound from each cluster for experimental testing can help ensure comprehensive coverage of diverse chemical space, maximizing the chances of identifying effective candidates.

The columns in

Table 1. List of drugs clustered based on chemical similarity.

Group 1	Group 2	Group 3	Group 4	Group 5	Group 6
H-7	saracatinib	tianeptine	kavain	metformin	ascorbic acid
fasudil	fostamatinib	KN-93	amitriptyline		chromomycin-a3
	tofacitinib	PIK-75	alvocidib
	dasatinib		fluoxetine
	caffeine		ampicillin
	TWS-119		nornicotine
	PP-2		fluspirilene

correspond to different chemical groups, allowing for a visual representation of these structural clusters.

2.5. Predicted Mechanisms of Actions

The final phase of the study involved using the tool L1000FWD to predict the mode of action of our results and assess their association with the disease mechanism through current research. The final list of candidate compounds with the potential therapeutic effects for the disease is presented in

Table 2. List of drugs with their predicted Mode of Action.

Drug Name	Predicted MOA	Probability
H-7	Dopamine receptor antagonist	0.6717
fasudil	Dopamine receptor antagonist	0.7290
saracatinib	EGFR inhibitor	0.9684
fostamatinib	EGFR inhibitor	0.3228
dasatinib	Aurora kinase inhibitor	0.3754
TWS-119	Dopamine receptor antagonist	0.4940
PP-2	Cyclooxygenase inhibitor	0.2409
KN-93	EGFR inhibitor	0.2617
PIK-75	Cyclooxygenase inhibitor	0.2910
kavain	Cyclooxygenase inhibitor	0.4005
amitriptyline	Histamine receptor antagonist	0.9347
alvocidib	Topoisomerase inhibitor	0.4321
fluoxetine	Histamine receptor antagonist	0.5377
fluspirilene	Dopamine receptor antagonist	0.9649

. The probabilities in this table represent confidence scores generated by L1000FWD through statistical comparison of gene expression signatures using gene set enrichment analysis (comparing the ranked list of genes from our input signature against reference signatures of drugs). The resulting enrichment scores are transformed into probability-like values that indicate the confidence that the drug acts via the predicted mode of action. These scores are computational estimates based on transcriptional similarity, not direct experimental probabilities, and are intended to guide hypothesis generation and further validation.

The drugs tofacitinib, caffeine, tianeptine, ampicillin, nornicotine, metformin, ascorbic acid and chromomycin A3 are not available in the tool, and thus their mechanisms of action were not predicted. Although these eight drugs could not be examined due to tool limitations, such gaps are expected when using computational resources with finite reference databases and do not affect the broader conclusions of the study. MOA predictions for the remaining 14 compounds were included to illustrate potential mechanistic links and generate hypotheses, and they do not serve as definitive evidence but rather complement the literature-based assessment. The focus of the study is the identification and prioritization of compounds, with MOA predictions serving as an additional layer of mechanistic insight where available.

3. Discussion

The current study presents a methodology for conducting drug repurposing in schizophrenia by utilizing gene signature reversion to identify potential treatments. The drugs H-7, fasudil, TWS-119, and fluspirilene are predicted to be dopamine receptor antagonists. The dopamine hypothesis of schizophrenia, one of the most widely accepted models, proposes that the disorder is caused by the overactivity of D2 dopamine receptors [33,34,35]. This perspective has shaped much of the pharmacological research in schizophrenia. Increased striatal dopamine activity has been noticed in patients with psychotic prodromal symptoms, suggesting that dopamine dysregulation precedes the onset of the disease. Dopamine receptors are categorized into the D1 and D2 classes and the latter one includes D2, D3 and D4 receptors. Current antipsychotic drugs have mainly targeted D2 receptors, but recent research highlights the therapeutic potential of D3 receptors antagonists, particularly in treating the negative and cognitive symptoms. Moreover, reduced D1 receptor activity has been associated with the development of negative symptoms, while D4 receptor genes have been linked to an increased genetic predisposition to the disease [36].

Saracatinib, fostamatinib and KN-93 are predicted to be inhibitors of epidermal growth factor (EGF) receptor. Research has demonstrated a strong connection between EGF signaling and the regulation of dopaminergic neurons, as well as its association with schizophrenia. Individuals with schizophrenia exhibit reduced EGF protein levels in the prefrontal cortex, striatum, and serum. In addition, selective knockout of HB-EGF in the ventral forebrain resulted in schizophrenia-like symptoms, which are improved by both typical and atypical antipsychotic medications. Moreover, EGF gene polymorphisms have been associated with this disorder [37]. Several studies have further reported a link between members of EGF and the disease. One study examined neuropathological changes in EGF ligands and ErbB1-4 receptors in patients with schizophrenia, revealing elevated ErbB1 receptor levels in the forebrain and reduced EGF concentrations in the blood. Concurrently, Decode Genetics Inc. conducting an analysis of Iceland’s national genome bank and identified a genetic association between a neuregulin-1 haplotype and schizophrenia. Moreover, Groenestege and colleagues reported that a family lineage with EGFR-mutation-driven renal disease also showed a co-occurrence of the disease. Furthermore, elevated levels of neuregulin-1 mRNA and its receptor protein ErbB4 have been observed in the brain of patients, along with increased neuregulin protein levels in the blood [38].

The compounds PP-2, PIK-75, and Kavain were predicted to be cyclooxygenase (COX) inhibitors. Extensive evidence highlights the significant role in enhancing the effectiveness of therapies for various diseases, particularly schizophrenia. Although they do not directly treat the disorder, these inhibitors provide complementary benefits when combined with standard treatments, particularly in the disease’s early phase [39]. For instance, Müller [40], reported that patients with acute schizophrenia who were treated with both risperidone and COX-2 inhibitor (celecoxib) showed better clinical outcomes than those who received only risperidone. However, a large study involving a broader population of schizophrenia patients found no significant benefit from the COX-2 inhibition in chronic schizophrenia (duration up to 10 years). These findings suggest that the therapeutic effects of COX-2 inhibitors are limited to acute phases of the disease and are not in long-term schizophrenia [40].

Alvocidib was predicted to be a topoisomerase inhibitor targeting functions associated with neurological conditions, including schizophrenia. Specifically, aberrant activity of Top1 and Top2 results from mutation of enzymes responsible for their regulation. Direct inactivation of Top3β, another topoisomerase, has been linked to schizophrenia. Research shows that individuals at increased risk of schizophrenia and intellectual disability often have a single copy loss of this enzyme [41]. Notably, Top3β interacts with FMRP, a protein silenced in Fragile X syndrome, the leading cause of autism. Since autism and schizophrenia share common etiological factors, independent genome-wide sequencing studies have examined whether Top3β mutations are linked to mental disorders. These studies found two de novo single nucleotide variants (SNVs) of Top3β in individuals diagnosed with either autism or schizophrenia [8]. Additionally, Top3β binds multiple mRNAs encoded by genes implicated in both conditions [41].

The histamine receptor antagonists’ properties of the amitriptyline and fluoxetine as predicted by the L1000FWD tool, suggest their significant role in the pathophysiology of several neuropsychiatric diseases such as schizophrenia. Post-mortem studies reveal elevated tele-methylhistamine (t-MH) levels in the cerebrospinal fluid (CSF) of schizophrenia patients, suggesting increased central histaminergic activity. In the dorsolateral prefrontal cortex, H3 receptor expression was significantly elevated in the patients treated with atypical antipsychotics, while PET studies indicated reduced H1 receptor binding in the frontal, prefrontal, and cingulate cortices. Although second-generation antipsychotics such as clozapine and olanzapine strongly block H1 receptors, their contribution to therapeutic effects or side effects in schizophrenia remains unclear, with the clinical findings offering mixed results [42].

As a final note to this bibliographic investigation, an absence of literature connecting a drug repurposing candidate with the disease, does not exclude its possible therapeutic effect, it only validates the drug, making it more likely to be effective.

Beyond the drug targets identified in this analysis, several other studies have highlighted additional pharmacological targets with potential therapeutic benefits for individuals with schizophrenia. The glutamatergic system plays a key role in the disease’s pathogenesis, with evidence pointing to elevated glutaminase expression and abnormalities in NMDA1 glutamate receptor. Riluzole, a glutamate-modulating drug, has shown to decrease the synaptic glutamate levels and improve the negative symptoms by changing striatocortical connectivity. Memantine, an NMDA antagonist, appears effective in treating both positive and negative symptoms. Moreover, muscarinic acetylcholine receptors (mAChR), especially the CHM1 and CHM4, are implicated in the pathophysiology and treatment of the disease. Xanomeline, a CHM1/CHM4 agonist, has demonstrated antipsychotic and precognitive effects without the long-term adverse effects typically associated with dopamine-based antipsychotics. TAAR1 (trace amine-associated receptor 1) has emerged as a novel target. Ulotaront, a TAAR1 antagonist, has shown promising results in clinical trials, with efficacy in treating both positive and negative symptoms and a safety profile [43].

In the final stage of drug repurposing, the validation of the findings from the computational analysis is crucial. This involves conducting wet-lab experiments, such as in vivo and in vitro assays, as well as controlled populations studies [44]. To ensure comprehensive validation, selecting at least one drug from each cluster represented in the table columns is recommended. This study identifies candidate drugs with potential therapeutic effects for schizophrenia, offering valuable insights that could reduce experimental efforts, and associated costs, while supporting more efficient downstream validation [45].

4. Materials and Methods

This study presents a systematic method to identify potential drug candidates through a signature-based method for drug repurposing. The workflow of this method, which is based on two research articles [46,47], is depicted in Figure 1.

4.1. Dataset

The data used in this study were obtained from the Gene Expression Omnibus (GEO) platform [48]. The dataset (accession number GSE263180) includes RNA sequencing (RNA-seq) data profiling gene expression levels in peripheral blood leukocytes from 9 patients with schizophrenia and 20 healthy controls [49]. While peripheral blood leukocytes cannot fully capture central nervous system pathology, prior studies support their utility as accessible proxies for investigating disease-related transcriptional patterns. Indeed, there is evidence that peripheral blood mononuclear cells and brain tissue of patients with schizophrenia share common enriched pathways [50]. The RNA sequencing samples in the study were collected from schizophrenia patients experiencing acute episodes who had not been treated with systemic or effective antipsychotic medications. Healthy controls were recruited through a community survey. The participants of both groups were free from neuroimmunological and neurodegenerative disorders, cardiovascular and cerebral vascular diseases, serious infections, or recent surgery [49].

4.2. Differential Gene Expression Analysis

This phase of the study focused on identifying genes with differential expression between patients diagnosed with schizophrenia and healthy controls. The final list of differentially expressed genes (DEGs) produced in this study was initially derived by [49] but the process was replicated in this paper for two reasons. First, reproducing published results is useful because it helps confirm whether earlier findings remain consistent when repeated independently. Given the ongoing reproducibility crisis in research, this effort contributes to improving transparency and reliability. Second, we aimed to provide a clear step-by-step demonstration that could serve as a practical guide for other researchers interested in deriving DEGs using similar methods.

Differential expression analysis was conducted using DESeq2 [51], an R/Bioconductor package (3.18), within the R programming environment (version 4.3). The dataset was initially divided into two groups: schizophrenia and control, and a normalization process was applied to ensure comparability. DESeq2 fits a negative binomial generalized linear model to compare gene expression levels between control and disease groups, estimates fold changes, and tests for significance using the Wald test. It estimates gene-wise dispersion and shrinks them toward a trend fitted across all genes to stabilize variance estimates and improve statistical power, especially in small sample sizes.

To account for multiple testing, resulting p-values were adjusted using the Benjamini & Hochberg procedure to control the false discovery rate. Explicit batch effect correction could not be performed because batch information was not available in the original dataset. However, DESeq2’s size factor normalization was applied to account for differences in library size and sequencing depth, which reduces certain sources of technical variability. Genes with adjusted p-values below the significance threshold were considered differentially expressed. The analysis can be fully reproduced following the standard DESeq2 workflow, and no additional scripts were used.

Genes were classified as follows:

• Upregulated Genes: Genes with adjusted p-value < 0.05 and log2FC > 0 (FC refers to fold-change)
• Downregulated Genes: Genes with adjusted p-value < 0.05 and log2FC < 0

In addition, the following filters were applied for the selection of DEGs.

• Genes with adjusted p-value exceeding 0.05 and absolute log2FC below 1 were removed.
• Duplicate genes were filtered out.
• Genes classified as both upregulated and downregulated were excluded.
• Genes that are invalid or cannot be used by CMap were also removed.

The final output of this phase was a list of upregulated and downregulated genes, a genetic signature of schizophrenia.

4.3. Drug Repurposing Tools

For finding new potential therapeutic compounds, we exploited the schizophrenia’s signature (differentially expressed genes) in two online tools, CMap [52] and L1000CDS2 [53], which compare this signature against other drug specific gene expression profiles.

4.3.1. CMap Tool

The Connectivity Map (CMap) is a resource that supports data-driven research on drug mechanisms and drug repurposing. It identifies relationships between input gene signatures and a large database of gene expression profiles [54]. This database is generated by treating various cell lines with a range of chemical compounds and measuring the resulting alterations in gene expression. The output is a ranked list of perturbagens based on their connectivity scores (CS), which can be either positive or negative values [55]. The connectivity score (tau) is a rank-based metric derived from comparing the input gene expression signature to reference signatures of perturbagens in the Connectivity Map. It is calculated using a Kolmogorov–Smirnov–like enrichment statistic that measures the tendency of the input’s up- and down-regulated genes to appear at the extremes of a perturbagen’s ranked gene list. A tau score close to +100 indicates strong similarity, meaning the perturbagen induces a gene expression pattern similar to the input, while a score near –100 indicates strong reversal, meaning the perturbagen produces an opposing pattern. Scores with an absolute value greater than 90 are typically considered highly significant and may warrant further investigation, https://clue.io/connectopedia/connectivity_scores (accessed on 15 March 2025) [56].

In this study, we used the top 150 upregulated and top 150 downregulated genes from the signature obtained in 2.2 as our query. Since the goal was to identify compounds that could reverse this signature, we selected perturbagens with connectivity scores lower than −60 [57]. A subsequent literature review was conducted for all candidate drugs to assess their association with schizophrenia. The final list includes only those compounds for which research has demonstrated their potential therapeutic effect in treating the disease.

4.3.2. L1000CDS²

L1000CDS2 is a search-engine within LINCS L1000 framework, designed to analyze gene expression signature. It utilized the L1000 dataset analyzed through the characteristic direction (CD) method, which improves the signal-to-noise ratio more efficiently than the MODZ approach presently applied for generating L1000 signatures. This tool ranks signatures from numerous small-molecules and their pairings to determine whether they imitate or oppose an input gene expression pattern using two techniques. It also identifies therapeutic targets for all small compounds analyzed with the L1000 assay. These targets are predicted by comparing their signatures with a large set of single-gene perturbations signatures from the GEO using cosine similarity [53].

This tool operates similarly to CMap, where a disease signature is input, and the output is a ranked list of 50 candidate drugs based on their score. The score is calculated as the proportion of genes affected by each compound to the total genes included in the signature. Notably, the same drug may appear multiple times, indicating its transcriptional response being recorded across multiple cell lines [46]. In our analysis, we utilized the programming language R to retain only one instance of each compound and then we subsequently reviewed the research literature to investigate their potential relationship with schizophrenia.

4.4. Clustering with ChemBioServer 2.0

The ChemBioServer is a publicly available tool designed to support drug discovery and repurposing by enabling filtering, clustering, and constructing structural similarity networks of chemical compounds [57]. This software was used to perform structural grouping of the potential compounds, following the principle that chemical similarity typically aligns with similarity in mode of action. The analysis included only compounds identified in prior research as potentially associated with schizophrenia. Hierarchical clustering was performed using Soergel distance as the metric and Ward’s method for linkage, with a clustering threshold set at 1 [57].

4.5. Predicting MOAs Using L1000FWD

L1000 Fireworks display (L1000FWD) [58] is an online application that offers interactive visualizations of over 16,000 gene expression profiles triggered by drugs and small molecules. It allows users to color-code signatures according to various criteria, such as drug-related attributes like mode of action (MOA). The tool also helps identify signatures that either mimic or reverse a given set of upregulated and downregulated genes [58]. In the final stage of our study, we used this platform to predict the mechanism for each candidate compound as it is a useful tool for discovering new small molecule functions through unsupervised clustering and investigating drug MOAs. In addition, we conducted a literature review to assess the existing evidence connecting these potential drugs to schizophrenia-related pathways and mechanisms. This combined approach allowed us to better understand their potential therapeutic applications in the context of schizophrenia.