Multicenter Validation of lncRNA and Target mRNA Diagnostic and Prognostic Biomarkers of Acute Ischemic Stroke From Peripheral Blood Leukocytes

Abstract

Background

Long noncoding RNA (lncRNA) and mRNA profiles in leukocytes have shown potential as biomarkers for acute ischemic stroke (AIS). This study aimed to identify altered lncRNA and target mRNA profiles in peripheral blood leukocytes as biomarkers and to assess the diagnostic value and association with AIS prognosis.

Methods and Results

Differentially expressed lncRNAs (DElncRNAs) and differentially expressed target mRNAs (DEmRNAs) were screened by RNA sequencing in the discovery set, which consisted of 10 patients with AIS and 20 controls. Validation sets consisted of a multicenter (311 AIS versus 303 controls) and a nested case–control study (351 AIS versus 352 controls). The discriminative value of DElncRNAs and DEmRNAs added to the traditional risk factors was estimated with the area under the curve. NAMPT‐AS, FARP1‐AS1, FTH1, and NAMPT were identified in the multicenter case–control study (P<0.05). LncRNA NAMPT‐AS was associated with cis‐target mRNA NAMPT and trans‐target mRNA FTH1 in all validation sets (P<0.001). Similarly, AIS cases exhibited upregulated lncRNA FARP‐AS1 and FTH1 expression (P<0.001) in the nested case–control study (P<0.001). Furthermore, lncRNA FARP1‐AS1 expression was upregulated in AIS patients at discharge with an unfavorable outcome (P<0.001). Positive correlations were found between NAMPT expression level and NIHSS scores of AIS patients (P<0.05). Adding 2 lncRNAs and 2 target mRNAs to the traditional risk factor model improved area under the curve by 22.8% and 5.2% in the multicenter and the nested case–control studies, respectively.

Conclusions

lncRNA NAMPT‐AS and FARP1‐AS1 have potential as diagnostic biomarkers for AIS and exhibit good performance when combined with target mRNA NAMPT and FTH1.

Nonstandard Abbreviations and Acronyms

AIS

acute ischemic stroke

DElncRNA

differentially expressed long noncoding RNA

DEmRNA

differentially expressed target mRNA

FC

fold change

IS

ischemic stroke

LAA

large‐artery atherosclerosis

lncRNA

long noncoding RNA

mRS

modified Rankin Scale

NAMPT

nicotinamide phosphoribosyltransferase

NIHSS

National Institute of Health Stroke Scale

SVO

small‐vessel occlusion

TOAST

Trial of ORG 10172 in Acute Stroke Treatment

TRF

traditional risk factor

Research Perspective.

What Is New?

• Our multicenter and nested case–control studies provide the first essential evidence that lncRNA NAMPT‐AS and FARPI‐AS have the potential to be acute ischemic stroke diagnostic biomarkers, and the combination of biomarkers NAMPT‐AS, FARP1‐AS1, NAMPT, and FTH1 improves the discrimination of acute ischemic stroke.

What Question Should Be Addressed Next?

• Further applications of these biomarkers to the diagnosis and prognosis of acute ischemic stroke would be warranted, and the potential therapeutic value needs further exploration.

Stroke remains the second‐leading cause of death and disability in the world as per Global Burden of Disease reports, with 101.5 million prevalent strokes around the world in 2019. ¹ Annually, there are >7.6 million new ischemic strokes (ISs), which accounts for approximately 60% to 80% of strokes. ¹ , ² , ³ However, only 10% to 24% of patients with acute ischemic stroke (AIS) received recanalization therapies in the eligible therapeutic time window. ⁴ , ⁵ Additionally, stroke reduces mobility in more than half of stroke survivors aged ≥65 years, ⁶ and 3.3 million deaths are attributed to IS annually. ³ This highlights the urgency of finding more effective biomarkers that can improve AIS diagnosis and prognosis.

High‐throughput sequencing technology has facilitated the discovery of blood‐based transcriptome biomarkers that play a crucial role in the diagnosis and prognosis of IS. Long noncoding RNA (lncRNA) is a type of RNA longer than 200 nucleotides and has little protein‐coding potential, which is reported to be involved in stroke pathophysiology including inflammation, atherogenesis, and stress response. ⁷ A case–control study found that the combination of exo‐lnc_000048, exo‐lnc_001350, and exo‐lnc_016442 based on RNA sequencing exhibited diagnostic potential for the diagnosis and prognosis of large‐artery atherosclerosis (LAA) stroke. ⁸ LncRNA NR_120420 in whole blood was reported to have a high diagnostic value for AIS with an area under the curve (AUC) of 0.86. ⁹ Therefore, altered lncRNA profiles in circulation showed the potential of diagnostic, therapeutic, and prognostic biomarkers for IS. ¹⁰ , ¹¹ , ¹²

However, lncRNA expression differs among various leukocytes, ¹³ and the expression level of lncRNAs is relatively lower than mRNAs and proteins in the circulatory system. ¹⁴ Most studies focusing on single lncRNA lack evidence from other blood components. Although accumulative bioinformatics analyses ¹⁵ , ¹⁶ and experiments in vitro and in vivo ¹⁷ , ¹⁸ , ¹⁹ , ²⁰ have stressed the importance of using lncRNA‐related networks as biomarkers, there is still a lack of internal and external validation and model improvement reports in human studies.

Leukocytes, containing a nucleus and able to produce RNA, can synthesize protein and induce inflammatory processes. A study confirmed that lncRNA H19 promoted leukocyte inflammation by targeting the miR‐29b/C1QTNF6 axis in cerebral ischemic injury. ²¹ Additionally, a study performed on RNA sequencing focused on different blood components of IS cases observed that there were more transcriptome dynamic changes in neutrophils and leukocytes than monocytes over the first few days following stroke. ¹³ These studies showed the altered lncRNA profiles in leukocytes could be biomarkers of IS.

Thus, the primary objective of this study was to identify biomarkers from peripheral blood leukocytes for the diagnosis and prognosis of AIS. We tested differentially expressed lncRNAs (DElncRNAs) and target mRNAs (DEmRNAs) and validated positive results in a multicenter and nested case–control study. Then, we evaluated the discriminative and predictive values of DElncRNAs and DEmRNAs for AIS in both morbid and new‐onset statuses, as well as their values for AIS prognosis.

METHODS

Data Availability Statement

The data used in this work are subject to the following permissions and restrictions: “The data sets presented in this article are not available due to participants' confidentiality.” Requests for access to these data sets should be addressed to the corresponding author.

Subjects

This research consisted of a multicenter and a nested case–control study. The 321 AIS cases were recruited between May 2019 to June 2021 from 4 hospitals including the Affiliated Hospital of Xuzhou Medical University, Jurong People's Hospital, Nanjing Jiangning Hospital, and Yixing People's Hospital. All cases were diagnosed by professional neurologists according to computed tomography, magnetic resonance imaging, and typical clinical symptoms. The exclusion criteria were described clearly in Data S1. In addition, referenced the TOAST (Trial of ORG 10172 in Acute Stroke Treatment) ²² , ²³ subgroups, we focused on patients with LAA, cardioembolism, and small‐vessel occlusion (SVO). A total of 323 controls matched for AIS cases by age (±2 years) and sex came from prospective cohort studies conducted in Siyang County and Jurong City from 2018 to 2020. The excluded subjects were those who had severe heart, lung, brain, kidney, and tumor diseases, and those who were underweight due to malnutrition.

For the nested case–control study, there were 106 and 247 AIS cases from the Yixing cohort (n=7000) and Jurong cohort (n=8000) between May 2018 and October 2022, and 2 AIS participants were excluded since blood sample volume was under the limit of detection. AIS and death events during the follow‐up were collected from the local Centers for Disease Control and Prevention. The outcome confirmation was carried out by community doctors and trained investigators. A total of 352 age‐ and sex‐matched controls were selected from the corresponding cohort. The inclusion and exclusion criteria were the same as in the former case–control study.

Research Ethics and Informed Consent

Informed consent signatures were obtained from subjects or their relatives before the study, and the study was approved by the Research Ethics Committee of Nanjing Medical University (Nos. 2018676, 2019929, and 2015111), the Medical Ethics Committee of Xuzhou Medical University Affiliated Hospital (XYFY2018‐KL078), and the Medical Ethics Committee of Jurong People's Hospital (JRSRMYY‐2019‐02).

Data Collection and Laboratory Measurements

Basic data were collected by field survey, including age; sex; smoking and drinking status; chronic disease history including diabetes, hypertension, and dyslipidemia ²⁴ (described in Data S1); stroke subtype; clinical scores; blood pressure; fasting plasma glucose; and blood lipid indicators. We used the same methodology as the one employed in a previous study to detect indicators including glucose (GLU), total cholesterol, triglycerides, high‐density lipoprotein cholesterol, and low‐density lipoprotein cholesterol. ²⁴

National Institutes of Health Stroke Scale (NIHSS) scores and modified Rankin Scale (mRS) scores of AIS cases were evaluated by clinical experts at admission and discharge. The short‐term outcome of AIS cases was assessed by case severity at discharge and change of NIHSS or mRS (admission scores minus discharge scores). The severity of AIS cases was defined as follows: NIHSS scores (mild, 0–5; moderate, 6–13; severe, 14–40) ²⁵ , ²⁶ , ²⁷ or mRS scores (favorable, mRS ≤2; unfavorable, mRS>2). ²⁸ Long‐term outcomes of AIS cases including death and stroke death were recorded through the local chronic disease and death registration system, using the International Classification of Diseases, Tenth Revision (ICD‐10), which were verified by investigators to ensure data authenticity.

The detailed leukocyte samples collection and experimental procedures are described in Data S1, in which the 2^−ΔΔCT method was used to calculate relative expression levels of gene expression. Fold change (FC) was used to assess differential expressions between cases and controls.

RNA Sequencing and lncRNA and Target mRNA Biomarker Selection

The discovery set consisted of 5 LAA and 5 SVO AIS cases. In addition, we matched 10 hypertension controls and 10 healthy controls according to age and sex. All samples met the quality control criteria (as described in Data S1). Through bioinformatics analysis, we identified up‐ and downregulated DElncRNAs (P _adj<0.05 and |log₂FC|>1). After quantitative real‐time polymerase chain reaction melting curve analysis, a Sanger sequencing of quantitative real‐time polymerase chain reaction products, we included 8 DElncRNAs (RP11‐290D2.6, RP11‐196G18.24, RP13‐270P17.3, RP11‐554J4.1, RP11‐171I2.5, NAMPT‐AS, CYP17A1‐AS1, FARP1‐AS1) for further validation. Then we included 8 predicted and target DEmRNAs (FC>1) into the validation list on the basis of the definitions of cis‐ and trans‐target mRNAs (described in Data S1), which were supported by previous research evidence and successfully detected. Hence, we selected the 8 lncRNAs and 8 target mRNAs for further validation as potential biomarkers for the diagnosis and prognosis of AIS (See Table S1 for primer sequences).

Biomarker Validation in the Case–Control Study and Nested Case–Control Studies

The biomarker verification process consisted of an internal set (149 AIS cases versus 149 controls), an external validation set (162 AIS cases versus 154 controls), and a nested case–control set (351 AIS cases versus 352 controls), only those biomarkers that met certain criteria were selected for further validation. These criteria included a consistent FC direction between the discovery set and validation set and significant differences in gene expression (P<0.05) between total cases or any AIS subtype groups (LAA/SVO/cardioembolism) and controls (Figure 1).

Figure 1. Flowchart of study design.

The design of this study displayed the selection procedural of biomarkers, the validation process, and standards of biomarkers into the next validation, and the diagnosis and prognosis value assessment of the biomarker panel. For lncRNA selection, we designed primers for the top 30 upregulated and downregulated DElncRNAs, evaluated the results using quantitative real‐time polymerase chain reaction melting curve analysis, agarose electrophoresis, and Sanger sequencing of quantitative real‐time polymerase chain reaction products, and successfully designed primers for 13 up‐regulated and 4 down‐regulated DElncRNAs. We further validated the DElncRNAs in pre‐experiments and only 8 up‐regulated DElncRNAs were successfully detected. The internal set is from the same source as the sequenced cases, while the external validation set has another three centers. Both the internal and external validation sets were case–control studies. Only biomarkers that passed the two validation sets were tested in nested case–control samples. AIS indicates acute ischemic stroke; AUC, area under the curve; CE, cardioembolism; DElncRNA, differently expressed lncRNA; DEmRNA, differently expressed mRNA; IDI, integrated discrimination improvement; LAA, large‐artery atherosclerosis; mRS, modified Rankin scale; NIHSS, National Institute of Health Stroke Scale; NRI, Net Reclassification Index; RCS, restricted cubic spline; ROC, receiver operating characteristic; and SVO, small‐vessel occlusion.

Statistical Analysis

Descriptive data were generated for all variables, with continuous variables described using median and interquartile range, while categorical variables were described using frequencies and percentages. Mann–Whitney U test and χ² test were used to compare the differences between case and control groups for continuous and categorical variables, respectively. Mann–Whitney U test and Kruskal–Wallis test were employed to assess the difference in gene expression among disease severity groups, and the Bonferroni correction was used to adjust P values in subtype comparisons.

Spearman's rank correlation analysis was used to assess the strength of the expression of lncRNA and related target gene mRNA expression. Restricted cubic spline regression was conducted to assess the dose–response relationship between gene expression level and AIS risk and test for nonlinearity. The expression of genes was categorized into 4 groups by quartiles. Binary logistic regression was applied to calculate odds ratios (ORs) and corresponding 95% CIs for the association of gene expression and AIS with adjustment for covariates. In the multicenter case–control study, covariates adjusted in restricted cubic spline, logistic, and Cox regression analyses included age, sex, smoking, drinking, hypertension, diabetes, and dyslipidemia and also included body mass index in analyses of the nested case–control study.

The receiver operating characteristic curve was constructed to obtain the AUC in different models with adjustment for covariates. Delong test, net reclassification improvement, and integrated discrimination improvement were conducted to assess the improvement of gene expression in addition to the traditional risk factors (TRFs). Spearman's rank correlation analysis was used to assess the correlation between gene expression and NIHSS or mRS in admission and discharge and change of NIHSS or mRS. Cox regression analyses were performed to calculate the hazard ratio with 95% CIs to estimate the association between lncRNA or mRNA expression and death outcome for patients with AIS.

All analyses were performed using SAS software 9.4 (SAS Institute, Cary, NC) and R version 4.3.1 (R Foundation for Statistical Computing, Vienna, Austria). A 2‐tailed P value <0.05 was considered significant.

RESULTS

Participants Characteristics

In the case–control study (Table 1), age and sex were balanced (P>0.05) in these 2 sets. The median (interquartile range) age was 66.0 years in the internal set and 68.0 years in the external set. AIS cases of the 2 sets presented a higher prevalence of hypertension, the risk factors (smoking and drinking), and systolic blood pressure and high‐density lipoprotein‐cholesterol than controls (P<0.05).

Table 1.

Demographic and Clinical Characteristics of the Population in the Case–Control Study of Acute Ischemic Stroke

Characteristic	Groups	Internal validation set (n=298)				External validation set (n=316)
		AIS (n=149)	Control (n=149)	Z/χ 2	P	AIS (n=162)	Control (n=154)	Z/χ 2	P value
Age, y		66.0 (56.0–74.0)	66.0 (56.0–74.0)	0.245*	0.807	69.0 (58.0–73.0)	68.0 (61.0–74.0)	0.064*	0.949
Sex, %	Male	91 (61.1)	90 (60.4)	<0.001†	1.000	91 (53.8)	93 (55.0)	0.012†	0.913
	Female	58 (38.9)	59 (39.6)			78 (46.2)	76 (45.0)
Smoking, %	No	136 (91.3)	109 (73.2)	15.514†	<0.001	134 (79.3)	102 (60.4)	13.494†	<0.001
	Yes	13 (8.7)	40 (26.8)			35 (20.7)	67 (39.6)
Drinking, %	No	140 (94.0)	113 (75.8)	19.081†	<0.001	147 (87.0)	109 (64.5)	22.043†	<0.001
	Yes	9 (6.0)	36 (24.2)			22 (13.0)	60 (35.5)
Hypertension, %	No	66 (44.3)	90 (60.4)	7.116†	0.008	54 (32.0)	81 (48.0)	8.337†	0.004
	Yes	83 (55.7)	59 (39.6)			115 (68.0)	88 (52.0)
Diabetes, %	No	116 (77.9)	135 (90.6)	8.184†	0.004	122 (72.2)	134 (79.3)	1.948†	0.163
	Yes	33 (22.1)	14 (9.4)			47 (27.8)	35 (20.7)
Dyslipidemia, %	No	84 (56.4)	95 (63.8)	1.399†	0.237	84 (49.7)	121 (71.6)	16.066†	<0.001
	Yes	65 (43.6)	54 (36.2)			85 (50.3)	48 (28.4)
Systolic blood pressure, mm Hg		151 (135–165)	140 (129–156)	3.253*	0.001	156 (138–170)	149 (135–160)	2.186*	0.029
Diastolic blood pressure, mm Hg		85 (78–95)	84 (77–93)	0.779*	0.436	84 (75–93)	80 (73–88)	2.761*	0.006
Fasting blood glucose, mmol/L		5.70 (4.97–7.08)	5.71 (5.27–6.36)	0.221*	0.825	5.35 (4.75–6.32)	6.09 (5.43–7.47)	6.252*	<0.001
TC, mmol/L		4.47 (3.59–5.18)	4.66 (3.94–5.21)	1.814*	0.070	4.55 (3.96–5.03)	4.84 (4.29–5.31)	3.225*	0.001
Triglyceride, mmol/L		1.24 (0.97–1.68)	1.26 (0.87–1.85)	0.481*	0.631	1.43 (1.03–2.01)	1.31 (0.97–1.83)	1.487*	0.137
High‐density lipoprotein cholesterol, mmol/L		1.12 (0.92–1.31)	1.20 (1.01–1.44)	2.348*	0.019	1.11 (0.97–1.30)	1.43 (1.20–1.78)	7.925*	<0.001
Low‐density lipoprotein cholesterol, mmol/L		2.63 (1.99–3.11)	2.73 (2.14–3.17)	1.217*	0.224	2.79 (2.38–3.27)	2.64 (2.15–3.02)	2.794*	0.005

AIS indicates acute ischemic stroke; SBP, systolic blood pressure; TC, total cholesterol.

^*Mann–Whitney U test.

^† χ ² test.

For the nested case–control study (Table S2), there was a median follow‐up of 2.16 years. The median age of nested case–control study participants was 70.0 years. There was a higher prevalence of hypertension, dyslipidemia, and other indicators (systolic blood pressure, diastolic blood pressure, GLU) of AIS cases than in controls (P<0.05).

Bioinformatics Analysis of Discovery Set

The differential gene expression and pathway analysis results of the discovery set show that differential gene expression analysis showed a total of 257 upregulated and 407 downregulated DElncRNAs, which were displayed by a heatmap and a volcano plot (Figure S1). Pathway analysis of cis and trans genes was performed concerning the Kyoto Encyclopedia of Genes and Genomes and Gene Ontology. As for the Kyoto Encyclopedia of Genes and Genomes analysis of cis target genes, toxoplasmosis, leishmaniasis, and herpes simplex virus 1 infection were identified (Figure S1).

Comparison of lncRNAs and target mRNAs expression between AIS cases and controls.

In the internal set, lncRNA RP11‐290D2.6, RP11‐554J4.1, NAMPT‐AS, CYP17A1‐AS1, FARP1‐AS1, and mRNA FTH1, NAMPT, PCBP1, and PINK1 expression were higher in AIS (Table 2) and LAA and SVO cases (P<0.05) compared with controls (Table S3).

Table 2.

Expression of lncRNA and Related Target mRNAs in the Case–Control Study and the Nested Case–Control Study of AIS

Set	Type	lncRNA/target mRNA	AIS	Control	FC	Z	P value*	Regulation direction		Consistency
								Validation	Sequencing
Internal validation set			n=149	n=149
	lncRNA	RP11‐290D2.6	2.036 (1.100–5.221)	0.946 (0.476–2.013)	2.160	5.490	<0.001	Up	Up	True
	lncRNA	RP11‐196G18.24	1.214 (0.612–2.464)	0.921 (0.440–2.023)	1.315	1.457	0.145	Up	Up	True
	lncRNA	RP13‐270P17.3	1.171 (0.498–3.683)	0.986 (0.365–2.191)	1.194	1.809	0.070	Up	Up	True
	lncRNA	RP11‐554J4.1	1.143 (0.399–2.264)	0.944 (0.442–1.919)	1.213	0.330	0.741	Up	Up	True
	lncRNA	RP11‐171I2.5	0.988 (0.392–2.523)	1.016 (0.435–2.022)	0.970	0.221	0.825	Down	Up	False
	lncRNA	NAMPT‐AS	2.298 (1.157–3.957)	0.958 (0.552–1.735)	2.411	6.348	<0.001	Up	Up	True
	lncRNA	CYP17A1‐AS1	1.585 (0.666–5.183)	0.872 (0.398–2.168)	1.816	3.773	<0.001	Up	Up	True
	lncRNA	FARP1‐AS1	2.263 (0.825–7.315)	0.895 (0.216–3.036)	2.539	3.804	<0.001	Up	Up	True
	mRNA	FEMIA	1.017 (0.336–2.559)	0.862 (0.452–1.665)	1.174	0.115	0.909	Up	Up	True
	mRNA	ICAM4	0.897 (0.447–2.561)	0.886 (0.559–1.682)	1.011	0.061	0.952	Up	Up	True
	mRNA	ACAD11	1.173 (0.566–2.499)	0.891 (0.614–1.430)	1.315	1.735	0.083	Up	Up	True
	mRNA	FTH1	1.553 (0.956–2.831)	0.944 (0.679–1.496)	1.649	5.333	<0.001	Up	Up	True
	mRNA	NAMPT	1.719 (0.979–2.860)	1.033 (0.591–1.623)	1.660	5.545	<0.001	Up	Up	True
	mRNA	PCBP1	1.210 (0.702–1.719)	0.872 (0.596–1.479)	1.391	2.602	0.009	Up	Up	True
	mRNA	PINK1	1.237 (0.659–2.087)	0.936 (0.616–1.556)	1.323	2.135	0.033	Up	Up	True
	mRNA	RHOF	0.866 (0.390–1.421)	0.978 (0.707–1.290)	0.887	2.121	0.034	Down	Up	False
External validation set			n=162	n=154
	lncRNA	RP11‐290D2.6	0.643 (0.268–1.795)	0.708 (0.251–3.324)	0.914	1.078	0.281	Down	Up	False
	lncRNA	NAMPT‐AS	1.498 (0.725–3.561)	1.020 (0.425–2.501)	1.461	2.955	0.003	Up	Up	True
	lncRNA	CYP17A1‐AS1	0.763 (0.356–1.414)	0.978 (0.494–1.863)	0.784	2.578	0.01	Down	Up	False
	lncRNA	FARP1‐AS1	1.617 (0.705–3.083)	0.963 (0.376–2.265)	1.677	3.007	0.003	Up	Up	True
	mRNA	ACAD11	0.769 (0.400–1.535)	0.869 (0.471–1.979)	0.884	1.903	0.057	Down	Up	False
	mRNA	FTH1	1.940 (1.013–2.925)	0.953 (0.701–1.386)	2.042	6.835	<0.001	Up	Up	True
	mRNA	NAMPT	1.434 (0.725–4.079)	0.897 (0.403–2.199)	1.607	4.157	<0.001	Up	Up	True
	mRNA	PCBP1	0.849 (0.529–1.584)	0.740 (0.418–1.860)	1.135	0.631	0.528	Up	Up	True
	mRNA	PINK1	0.715 (0.416–1.227)	0.862 (0.444–1.977)	0.826	2.081	0.037	Down	Up	False
Combined set			n=311	n=303
	lncRNA	RP11‐290D2.6	1.339 (0.438–2.964)	0.819 (0.332–2.402)	1.642	2.458	0.014	Up	Up	True
	lncRNA	NAMPT‐AS	1.876 (0.858–3.880)	0.996 (0.482–2.064)	1.889	6.436	<0.001	Up	Up	True
	lncRNA	CYP17A1‐AS1	1.009 (0.506–2.317)	0.920 (0.456–1.909)	1.087	1.142	0.254	Up	Up	True
	lncRNA	FARP1‐AS1	1.783 (0.739–4.030)	0.899 (0.310–2.557)	2.000	4.908	<0.001	Up	Up	True
	mRNA	ACAD11	0.955 (0.473–1.896)	0.891 (0.565–1.666)	1.067	0.379	0.704	Up	Up	True
	mRNA	FTH1	1.721 (0.967–2.898)	0.946 (0.687–1.455)	1.830	8.162	<0.001	Up	Up	True
	mRNA	NAMPT	1.589 (0.844–3.293)	0.966 (0.507–1.755)	1.646	6.498	<0.001	Up	Up	True
	mRNA	PCBP1	0.975 (0.604–1.670)	0.822 (0.506–1.603)	1.183	1.875	0.061	Up	Up	True
	mRNA	PINK1	0.843 (0.507–1.682)	0.915 (0.545–1.745)	0.923	0.593	0.553	Down	Up	False
Nested case–control study			n=351	n=352
	lncRNA	NAMPT‐AS	1.001 (0.579–1.838)	0.887 (0.542–1.761)	1.250	0.381	0.406	Up	Up	True
	lncRNA	FARP1‐AS1	1.221 (0.594–2.702)	0.900 (0.530–1.815)	1.356	3.216	0.001	Up	Up	True
	mRNA	FTH1	1.163 (0.897–1.598)	1.001 (0.805–1.298)	1.160	5.103	<0.001	Up	Up	True
	mRNA	NAMPT	1.051 (0.789–1.572)	1.040 (0.743–1.387)	1.050	1.866	0.062	Up	Up	True

AIS indicates acute ischemic stroke; and FC, fold change.

^*Mann–Whitney U‐test.

Hence, in the external validation set, we selected the 4 lncRNAs and 5 target mRNAs for verification. Among them, NAMPT‐AS, FARP1‐AS1, FTH1, and NAMPT showed consistency with sequencing results and a significant difference between cases and controls for the combined set (NAMPT‐AS: FC=1.889, P<0.001; FARP1‐AS1 FC=2.000, P<0.001; FTH1: FC=1.830, P<0.001; NAMPT: FC=1.646, P<0.001; Table 2). Besides, NAMPT‐AS, FARP1‐AS1, and NAMPT expression levels were higher in LAA and SVO cases than in controls, and FTH1 expression level was higher in all 3 subgroups of cases (LAA, SVO, cardioembolism) compared with controls (P<0.05; Table S3).

In the nested case–control study, the change direction of NAMPT‐AS, FARP1‐AS1, NAMPT, and FTH1 was consistent with the sequencing results (all FC>1). LncRNA FARP1‐AS1 (FC=1.356, P=0.001) and target mRNA FTH1 expression levels (FC=1.160, P<0.001) showed significant differences between cases and controls (Table 2). Furthermore, the analysis of DElncRNAs and DEmRNAs expression in hypertensive and nonhypertensive populations revealed no significant differences (Table S4).

Correlation Between lncRNA NAMPT ‐AS and Target mRNAs ( FTH1 and NAMPT )

In the case–control study, NAMPT‐AS expression showed a significant correlation with NAMPT (ρ=0.398, P<0.001) and FTH1 (ρ=0.231, P<0.001). In the nested case–control study, NAMPT‐AS expression showed a significant correlation with NAMPT (ρ=0.517, P<0.001) and FTH1 (ρ=0.285, P<0.001; Figure S2).

Association of Expression Level of lncRNAs and Target mRNAs With AIS Risk

In the multicenter case–control study (Figure 2), increased expression level of NAMPT‐AS, FARP1‐AS1, FTH1, and NAMPT was associated with a higher risk of AIS (All P _overall<0.001, P _nonlinear<0.001). Moreover, the restricted cubic spline regression models demonstrated a significant dose–response association between gene expression level and AIS risk in a nested case–control study (Figure 3) (All P _overall<0.001, P _nonlinear>0.001).

Figure 2. LncRNAs and target mRNAs expression with the ORs of AIS in multicenter case–control validation set.

A through D show in the case–control study the ORs with 95% CIs for ischemic stroke correlated to lncRNAs and target mRNAs expression in 614 participants (P<0.05), adjusted for age, sex, smoking, drinking, hypertension, diabetes, and dyslipidemia. AIS indicates acute ischemic stroke; lncRNA, long noncoding RNA; and OR, odds ratio.

Figure 3. LncRNAs and target mRNAs expression with ORs for AIS in the nested case–control validation set.

A through D show that in a nested case–control study, the ORs with 95% CIs for AIS correlated with 4 gene expressions in the 704 participants (P _overall<0.05), adjusted for age, sex, body mass index, smoking, drinking, hypertension, diabetes, and dyslipidemia. AIS indicates acute ischemic stroke; and OR, odds ratio.

Compared to individuals in the referenced group, other groups of NAMPT‐AS, FARP1‐AS1, FTH1, and NAMPT expression levels displayed gradually increased risk of AIS with adjusted ORs of 1.179 to 8.548 (Figure S3).

In the nested case–control study, the increased expression level of lncRNA FARP1‐AS1 was associated with the risk of AIS (P _nonlinear=0.002; Figure 3), and the adjusted ORs of Q2 to Q4 were gradually increased by 1.172 to 2.474, after adjustment for age, sex, body mass index, smoking, drinking, hypertension, diabetes, and dyslipidemia (Figure S4).

Predictive Value of lncRNAs and Target mRNAs for AIS, LAA, and SVO

In the case–control study, the TRF model including age, sex, smoking, drinking, hypertension, diabetes, and dyslipidemia had an AUC of 0.727 (Figure 4). Compared with the TRF model, the TRF+expression of 4 genes model (AUC, 0.806; specificity, 0.851; sensitivity, 0.631; Table S5) improved the predictive power by 22.8% (net reclassification improvement, 0.228 [95% CI, 0.105–0.340]; integrated discrimination improvement, 0.131 [95% CI, 0.104–0.158]; Table S6). Moreover, there were 13.7% and 20.1% improvements for the LAA and SVO subtype diagnosis (Tables S7 through S9).

Figure 4. The receiver operating characteristic (ROC) curve of genes and traditional risk factors for AIS.

The differences in the AUC between TRF and any model added to any gene expression are shown in (A) multi‐center case–control study (P<0.001) and (B) nested case–control study (P>0.05). Moreover, in (A), compared with the TRF model, the TRF+expression of 4 genes (AUC, 0.806) model improved 22.8% predictive power (NRI, 0.228 [95% CI, 0.105–0.340]). In (B), compared with the TRF model, the TRF+expression of 4 genes (AUC, 0.710) model positively improved 5.2% predictive power (NRI, −0.076 [95% CI, −0.192 to 0.069]). TRF in (A) includes age, sex, smoking, drinking, hypertension, diabetes, and dyslipidemia; and TRF in (B) added body mass index on the basis of these variables. AUC indicates the area under the curve; IDI, integrated discrimination improvement; NRI, Net Reclassification Index; and ROC, receiver operating characteristic.

In the nested case–control study (Table S10), The TRF+expression of the 4‐gene model did improve the probability of accurately identifying participants by 5.2% (integrated discrimination improvement, 0.052 [95% CI, 0.035–0.069]; Table S11).

Analysis of Gene Expression Level Relevant to Clinical Score, Onset Time, and Prognosis

Positive correlations were found between NAMPT expression and NIHSS scores at admission (ρ=0.212, P<0.001) and discharge (ρ=0.199, P=0.001) in patients with AIS (Figure 5). Additionally, a correlation was found between the change in NIHSS scores and NAMPT expression (ρ=0.128, P<0.05; Table S12).

Figure 5. The association between NIHSS scores and expression of lncRNA and target mRNA among AIS cases.

A and B show the correlation between the expression of target mRNA NAMPT and the NIHSS scores of admissions and discharge in the case–control study of AIS. NAMPT target mRNA expression has significant relevance to NIHSS scores of admissions (ρ=0.212, P<0.05) and discharge (ρ=0.199, P=0.001) for patients with AIS from the case–control study. C and D showed that the expression of FARP1‐AS1 had significant differences between severe (NIHSS score ≧14) AIS cases and mild (0 ≧NIHSS score ≧5) AIS cases both in admission (P=0.045) and discharge (P=0.046). NIHSS indicates National Institutes of Health Stroke Scale.

In the nested case–control study, negative correlations were found between the expression levels of NAMPT‐AS (ρ=−0.114, P=0.003) and FTH1 (ρ=−0.104, P=0.006) and the time to onset of AIS incidence (Table S13 and Figure S5).

Furthermore, the expression level of FARP1‐AS1 was significantly higher in severe cases than in mild cases at both admission (FC=1.707, P=0.045) and discharge (FC=2.182, P=0.046) (Table S14). FARP1‐AS1 expression level at the discharge of those with unfavorable outcomes was higher than those without unfavorable outcomes (FC=1.497, P=0.021) (Table S15).

As shown in restricted cubic spline curve plots, there are no associations between gene expression level and outcomes including death and stroke death (P _nonlinear>0.05) of AIS patients in the case–control study (Figure S6).

DISCUSSION

In this study, NAMPT‐AS, FARP1‐AS1, FTH1, and NAMPT showed the potential of being diagnosis biomarkers for AIS. The risk of AIS generally exhibited an upward trend with elevated NAMPT‐AS, FARP1‐AS1, FTH1, and NAMPT expression levels in a case–control study. In contrast, only FARP1‐AS1 had significant associations in a nested case–control study set. Furthermore, the TRF plus expression of the 4‐gene model presented better performance than the TRF model in the multicenter and nested case–control study.

NAMPT‐AS (also known as RP11‐22N19.2) was found locating in the promoter region of mRNA NAMPT and transcribed from the antisense strand of NAMPT. Since NAMPT‐AS is a novel transcript, there is scarce research observing its association with any diseases. NAMPT‐AS was highly expressed in LAA and SVO cases compared with controls with a 1.889 and a 1.250 FC in the multicenter and nested case–control study set, respectively. NAMPT‐AS was significantly associated with cis‐target mRNA NAMPT and trans‐target mRNA FTH1, which were consistent with the prediction based on the RNA sequencing results of the discovery set. These findings suggest that lncRNA NAMPT‐AS may be coexpressed with NAMPT and affected by FTH1, presenting an upward trend for LAA and SVO stroke. NAMPT‐AS may also be a key module in leukocyte activation involved in inflammatory responses and ferroptosis, which are the main functions and molecular mechanisms of NAMPT and FTH1 in AIS as the discovery set predicted. Therefore, NAMPT‐AS may have the potential as a diagnosis biomarker for AIS.

Previous research has shown that FARP1‐AS1 is differentially expressed in various N stages (N0, N1, N2, and N3) in breast cancer, exhibiting upward dysregulation. ²⁹ In our study, we observed for the first time that FARP1‐AS1 was upregulated in patients with AIS, and the FARP1‐AS1 expression in severe cases was higher than mild ones both at admission and discharge. Moreover, elevated FARP1‐AS1 expression was associated with an unfavorable outcome at discharge with a 1.497 FC compared with AIS cases with a favorable outcome. These results suggested that FARP1‐AS1 may have potential as a diagnosis and short‐term prognostic biomarker for AIS.

Among these genes, both NAMPT and FTH1 have been identified as potential targets and biomarkers for IS. Nicotinamide phosphoribosyltransferase (NAMPT) is coded by the mRNA NAMPT, which is a specific diagnostic and therapeutic biomarker of IS. ³⁰ NAMPT is also known as visfatin or pre–B‐cell colony–enhancing factor ³¹ , ³² , ³³ and is a rate‐limiting enzyme for biosynthesizing nicotinamide adenine dinucleotide in mammals. ³⁴ Circulating NAMPT protein is dominantly secreted from adipocytes ³⁵ and leukocytes, ³⁶ and the serum or plasma NAMPT protein level in IS cases is 2 to 8 times higher than in controls. ³⁷ , ³⁸ , ³⁹ , ⁴⁰ In the case–control study validation, we found that the NAMPT expression level in peripheral blood leukocytes in patients with AIS was 1.646 times higher than that in controls. In addition, the upregulated expression of NAMPT showed a correlation with severity and short‐term poor outcomes of AIS. For the low correlation coefficients observed between NAMPT and the NIHSS score, the explanation may be the expression of NAMPT showed a skewed distribution in the low expression level. Most NIHSS scores (0–14) of AIS were distributed in the mild group (0–5), which accounted for 77.6% and 80.0% at admission and discharge, respectively. Therefore, a gentle upward trend was observed in the correlation curve with low coefficients. The upregulated mRNA expression of NAMPT was consistent with the alteration of the coded protein in circulation, and the Gene Ontology analysis of the discovery set in our study found that NAMPT was associated with leukocyte activation involved in inflammatory responses. Other studies have demonstrated that NAMPT protein plays a crucial role in IS, both in cerebroprotection in the acute phase ⁴¹ , ⁴² and vascular repair and neurogenesis in the chronic phase ⁴³ , ⁴⁴ through the NAMPT–nicotinamide adenine dinucleotide –sirtuin signaling pathway. In conclusion, our study again observed the vital value of mRNA NAMPT in the diagnosis and prognosis of AIS.

FTH1 mRNA codes for ferritin heavy chain 1 protein, which plays a crucial role in iron metabolism and storage. Evidence suggests that inflammation and ferroptosis, an iron‐dependent form of oxidative cell death, are significant contributors to the development of IS pathology. ¹³ Ferroptosis has been identified as a treatment and drug target for IS. ⁴⁵ , ⁴⁶ For instance, electroacupuncture can inhibit ferroptosis by increasing GPX4 and FTH1 iron‐related mRNA and proteins, which helps protect against middle cerebral artery occlusion in rat models. ⁴⁵ Additionally, FTH1–bone marrow stromal cells transplanted for treating focal cerebral infarction were safe, reliable, and traceable by magnetic resonance imaging. ⁴⁶ A new ferroptosis inhibitor named Herein, which targets nuclear receptor coactivator 4–ferritin heavy chain interaction, could provide a new strategy to treat and combat IS. ⁴⁷ The elevated alteration of FTH1 mRNA expression and FTH1 protein in vitro and vivo experiments in other studies were consistent with our validation direction in the circulation of humans. First, FTH1 was upregulated in AIS cases compared with controls in both the multicenter and nested case–control studies. Furthermore, there was a negative correlation between FTH1 and time to onset in the nested case–control study, indicating that elevated expression of FTH1 was associated with a shorter time to incident AIS. The consistent upregulation of FTH1 expression in both the case–control study and nested case–control study suggests that FTH1 may serve as a potential biomarker for AIS, not only for clinical prognosis but also for early screening and prevention before the occurrence of AIS events.

Overall, the identified biomarkers including NAMPT‐AS, FARP1‐AS1, FTH1, and NAMPT, were consistent in 3 validation sets. Particularly, lncRNA NAMPT‐AS and FARP1‐AS1 were first demonstrated that associated with LAA and SVO stroke, these 2 biomarkers all showed a positive correlation with severity and short‐term prognosis. Moreover, these biomarkers presented a great performance for AIS, LAA, and SVO prediction. Therefore, these biomarkers were suited for predicting prevalence risk for AIS and showed association with short‐term poor outcomes for AIS.

Strengths and Limitations

One of the key strengths of this study is its use of a multicenter and nested case–control study with a sufficient sample size, which assessed the altered lncRNA and mRNA profile characteristics in diseased‐ versus natural‐onset states. In addition, lncRNA from peripheral blood leukocytes as biomarkers brings benefits of the convenience of blood collection compared with brain cell samples, as well as lower costs and lower risks during the collection process. Furthermore, the study's analysis framework is comprehensive, including an evaluation of the predictive power improvement of the biomarkers, as well as the association between the biomarkers and subgroups, severity, discharge, and long‐term outcomes of AIS cases.

Since there is no gold standard for screening in the validation of DElncRNAs, we selected only the top 30 lncRNAs with the most significant up‐ and downregulation levels among 10 AIS cases and 20 controls, which was a relatively small sample size for RNA sequencing but could be overcome by large validation sets. In addition, the verification effect of other screening strategies such as pathway analysis needs to be further explored. Moreover, the AIS diagnosis biomarkers verified in this study have not yet verified their specificity in other cardiovascular diseases and need to be further validated. Finally, in the prognostic analysis of this study, the median follow‐up time of cases was 2.16 years; however, follow‐up is continuing, ideally removing restrictions on concluding the association between biomarkers and the long‐term mortality rate.

CONCLUSIONS

In summary, lncRNA FARP1‐AS1 and NAMPT‐AS in peripheral blood leukocytes were first identified to have the potential of diagnostic biomarkers for AIS with upregulation. lncRNA FARP1‐AS1 could be an early screening biomarker and is also associated with poor prognosis for patients with AIS at discharge. lncRNA NAMPT‐AS was positively correlated with the cis‐acting target mRNA NAMPT and the trans‐acting target mRNA FTH1. Finally, compared with traditional risk factor models, the biomarkers combination of lncRNA NAMPT‐AS, FARP1‐AS1, and mRNA NAMPT and FTH1 significantly improved the discriminative value of AIS.

Sources of Funding

This work was supported by the National Natural Science Foundation of China (Grant Nos. 8217121613 and 81872686); the National Key Research and Development Program of China (Grant No. 2018YFC2000703); the Research Unit of Prospective Cohort of Cardiovascular Diseases and Cancers of Chinese Academy of Medical Sciences (2019RU038); the Jiangsu Provincial Fourth “333 Project”, the Priority Academic Program Development of Jiangsu Higher Education Institutions (Public Health and Preventive Medicine).

Disclosures

None.

Supporting information

Data S1

Tables S1–S15

Figures S1–S6