Expression profile based gene clusters for ischemic stroke detection Whole blood gene clusters for ischemic stroke detection

Abstract

In microarray studies alterations in gene expression in circulating leukocytes have shown utility for ischemic stroke diagnosis. We studied forty candidate markers identified in three gene expression profiles to (1) quantitate individual transcript expression, (2) identify transcript clusters and (3) assess the clinical diagnostic utility of the clusters identified for ischemic stroke detection. Using high throughput next generation qPCR 16 of the 40 transcripts were significantly up-regulated in stroke patients relative to control subjects (p<0.05). Six clusters of between 5 and 7 transcripts discriminated between stroke and control (p values between 1.01e-9 and 0.03). A 7 transcript cluster containing PLBD1, PYGL, BST1, DUSP1, FOS, VCAN and FCGR1A showed high accuracy for stroke classification (AUC=0.854). These results validate and improve upon the diagnostic value of transcripts identified in microarray studies for ischemic stroke. The clusters identified show promise for acute ischemic stroke detection.

1. Introduction

Stroke is a leading cause of death and disability in the community and new diagnostics and therapeutics are greatly needed[1]. Inflammation and immune response after stroke impacts significantly on tissue and clinical outcome[2,3]. Application of molecular and cellular approaches to study the immune system in stroke may offer new diagnostic and therapeutic approaches.

Using microarrays that contained between 22,000 and 54,000 oligonucleotide probes, genomic profiling has been applied to the circulating leukocytes of human stroke patients[4-7]. Peripheral blood mononuclear cells (PBMCs)[4,7] and whole blood samples[5,6] were used for these studies. In three independent analyses 22, 18 and 9 transcripts showed utility for stroke detection[4-6]. In these studies ribonucleic acid (RNA) was sampled between 3 and 72 hours after stroke onset. Different microarrays from two companies (Affymetrix and Illumina) were used and therefore signal intensity was assessed differently for each study. Despite these methodological and experimental differences there was overlap among the transcripts identified and panels were able to be applied between the study cohorts[4-7].

These microarray studies raised the possibility of added diagnostic utility in stroke from genomic profiling of circulating leukocytes to clinical and neuroimaging information during the time window for thrombolytic therapy[8-11]. Expression changes were seen as early as 3 hours post stroke and persisted at 5 and 24 hours[5]. However, further translation and application of these microarray results has been hindered by data normalization issues, cost, high turnaround time and the limited availability of arrays. While providing unprecedented coverage of the transcriptome, microarray data are also limited by low sensitivity and low accuracy for transcripts expressed at low levels[12,13].

The majority of these stroke-related transcripts were not validated with standard quantitative polymerase chain reactions (qPCR) – the gold standard for measuring gene expression. qPCR- based approaches are more likely than microarrays to be applied and developed for rapid assays and automated point of care systems that would be needed for early stroke diagnosis[14,15]. Compared to microarrays qPCR approaches are characterized by shorter assay turnaround times and high sensitivity, with a theoretical limit of detection of a single copy of messenger ribonucleic acid (mRNA) target[16]. Until now standard reverse transcription (RT)-qPCR has been feasible for studying 6 genes at most from typical clinical samples.

Recently next generation microfluidic high throughput qPCR approaches have become available. These methods, known as high throughput RT-qPCR (HT RT-qPCR) or nanofluidic qPCR, permit the rapid quantification of multiple transcripts using small sample volumes[17,18] with very high sensitivity. Plates can contain up to 96 samples in which 96 transcripts can be simultaneously studied in 9,216 reactions. We have applied HT RT-qPCR to forty candidate markers identified in the three prior gene expression profiling studies to (1) quantitate individual transcript expression, (2) identify transcript clusters and (3) assess the clinical diagnostic utility of the clusters identified for ischemic stroke detection.

2. Methods

2.1. Study Subjects

Peripheral blood samples were obtained from 18 ischemic stroke patients admitted to the University Hospital of Brooklyn at SUNY Downstate Medical Center and at Long Island College Hospital and 15 gender and race matched control subjects recruited from the local community. The median time of blood draw was 36 hours post stroke onset. Stroke was diagnosed according to World Health Organization stroke criteria. The Institutional Review Board at the State University of New York (SUNY) approved the study and all study participants or their authorized representatives gave full and signed informed consent.

The study inclusion criteria were: over 18 years of age and acute ischemic stroke. The exclusion criteria were: current immunological diseases, taking steroid or immunosuppressive therapies, severe allergies, acute infection and severe anemia. The following clinical data were recorded: age, gender, race, self–reported risk factors, National Institutes of Health Stroke Scale (NIHSS) score in the stroke subjects and complete blood counts (CBC), including total white blood cell count and white cell differential counts. Hypertension was defined as a prior (at any time in the past) diagnosis of hypertension by the subject’s physician or currently receiving treatment for hypertension. Diabetes was defined as a past medical history of known diabetes mellitus. Coronary artery disease was defined as a physician-diagnosed past history of ischemic heart disease or angina. Hyperlipidemia was defined as a past history of documented elevation in total cholesterol (>200mg/dl). Smoking was defined as current or prior smoking. Atrial fibrillation was defined as a past or current history of physician-diagnosed atrial fibrillation.

2.2. Primer Selection and Development

40 transcripts identified in 3 previously published studies[4-6] were selected for analysis (Supplementary Table 1 and Supplementary Figure 1) . The 3 studies had identified 9, 18 and 22 genes within panels with some overlap among the studies. Hox 1.11, transcript identified in Tang’s et al. study[5], was not studied because it is a non-coding RNA sequence. Hypothetical protein FLJ22662 Laminin A motif from the Moore list[4] is now termed phospholipase B domain containing 1 (PLBD1) according to current nomenclature. Two variants of CD14 were studied to give a total of 41 transcripts that were tested. The complete primer characteristics were published earlier[18]. The RT-qPCR primers were self-designed, commercially synthesized by Invitrogen and wet tested using regular RT-qPCR (StepOnePlus Real-Time PCR Systems; Applied Biosystems).

2.3. Sample Processing

Where applicable the conduct and reporting of the study are in accordance with the Minimum Information for Publication of Quantitative Real-Time PCR Experiments criteria[19]. RNA was extracted using column separation (All-in-One Kit; Norgen Biotek, Thorold, Ontario, Canada) from 100μl of whole blood collected on ethylenediaminetetraacetic acid tubes (ETDA). Cell count (millions of cells per μl) was based on white blood cell (WBC) count from laboratory CBC obtained for each study subject. cDNA was synthetized using the High Capacity cDNA Reverse Transcription Kit (Life Technologies, Carlsbad, CA), based on random hexamers, according to the manufacturer’s protocol. In addition to study samples two commercial cDNA samples (Universal cDNA Reverse Transcribed by Random Hexamer: Human Normal Tissues; Biochain, Newark, CA) were run on each plate to perform normalization. HT RT-qPCR was run on the BioMark HD System, using 96 × 96 Fluidigm Dynamic Arrays (Fluidigm, South San Francisco, CA). Three plates were used for this study. The percent present calls were over 90%. Detailed methods have been published previously[18].

2.4. Gene Expression Data Analyses and Development of the Gene Classifier

Gene expression for each sample was measured using the input sample quantity method [20] after adjusting for the input cell count and normalizing to a standard volume of a standard cDNA sample (Universal cDNA Reverse Transcribed by Random Hexamer: Human Normal Tissues; Biochain, Newark, CA). The normalized copy number for each sample was obtained according to the equation:

$$X_c=\frac{{(1+E)}^{(nCq,cDNA-nCq,X)}}{CC}$$

where X_c is the transcript number per cell, E is the efficiency of target cDNA amplification, nCq,cDNA and nCq,X are the cycle number at which amplification crosses the threshold respectively for standard cDNA sample and for sample X, cc is the number of cells used for RNA extraction based on CBC result. The results for the stroke patients and control subjects were then compared. A 7 gene classifier was identified from a hierarchical cluster analysis. The upper level of normal for the expression of each transcript was defined as a value above the third quartile in the control subjects.

We have previously presented graphical results, based only on Cq values normalized to cell count, of four stroke related transcripts in a cohort of hemorrhagic and ischemic stroke patient [18].

2.5. Statistical Analyses

The data were analyzed using R version 2.15.1. Shapiro’s tests were used to assess for normality of the data. For grouped and categorical data t tests, Mann Whiney U, Wilcoxon rank sum and Student’s t tests were used to compare groups. Chi-square tests were used to compare categorical values. Spearman correlation coefficients were used to test the association of transcript expression with age and time of blood draw. Corrections for multiple comparisons used the Benjamini and Hochberg (false discovery rate, [FDR]) and Bonferroni algorithms. The hierarchical cluster analysis - a non-supervised technique to detect hidden associations in the data - used Ward’s method and log-transformed data. The Ward algorithm employs an Euclidian distance measure. A cutoff (“height”) level at “9” was used to give the 7 Clusters. Receiver operating curve analysis and sensitivity and specificity analyses were used to test diagnostic value of the 7 transcript cluster. p-values <0.05 were considered statistically significant.

3. Results

The patients and controls were matched on gender, race and stroke risk factors (Table 1). The mean age of the stroke patients was 71 years and of the controls was 58 years (p=0.004).

Table 1.

Clinical and laboratory characteristics of patients and controls

Factor	All (n=33)	Stroke (n=18)	Control (n=15)	p
Age	65.4±14.3	71.6±13.0	58.1±12.3	0.004
Gender– male	14 (42)	7 (39)	7 (47)	0.9
Race– black	30 (91)	17 (94)	13 (87)	0.9
Risk factors
Hypertension	28 (85)	17 (94)	11 (73)	0.2
Diabetes	15 (45)	8 (39)	7 (53)	0.6
Coronary artery disease	8 (24)	5 (28)	3 (20)	0.9
Smoking history	7 (21)	5 (28)	2 (13)	0.6
Atrial fibrillation	4 (12)	4 (22)	0 (0)	0.2
Hyperlipidemia	16 (48)	8 (44)	8 (53)	0.9
Medications
Diuretics	9 (27)	6 (33)	3 (15)	0.6
ACEIs/ARBs	9 (27)	7 (39)	2 (13)	0.2
Beta blockers	21 (64)	14 (78)	7 (47)	0.1
Calcium channel blockers	8 (24)	5 (28)	3 (20)	0.9
Anti-thrombotics	18 (54)	10 (55)	8 (53)	1.0
Statins	14 (42)	7 (39)	7 (47)	0.9
WBC count (109 cells/liter)	6.9±2.4	7.45±2.2	6.18±2.6	0.2
Stroke-Related
Time of blood draw (hours) Infarct volume (mm3)	N/A N/A	36.0 (23.0, 48.0) 5404.0 (1,207.0, 22,870.0))	N/A N/A	N/A N/A
NIHSS score	N/A	7.5 (4.2, 10.0)	N/A	N/A

Results are mean ±SD and median (interquartile range) for continuous factors and numbers (%) for categorical factors. ACEI – angiotensin converting enzyme inhibitor, ARB - angiotensin receptor blocker, WBC-white blood cell, N/A – not applicable, NIHSS – National Institutes of Health Stroke Scale.

3.1. Whole Blood Expression of Stroke-Related Transcripts

16 genes were significantly up-regulated in the stroke patients relative to the control subjects (p<0.05, Wilcoxon rank sum test, Table 2). The fold change differences for the 16 transcripts ranged from 6.4 for BST1 to 1.7 for IQGAP1 (Figure 1). Nine genes were altered at the p <0.01 level: these were CD93, S100A9, CYBB, S100A12, BST1, PLBD1, PYGL, ADM and CKAP4. All of these 9 genes remained significant after corrections for multiple comparisons using the FDR method and two (S100A12 and CD93) using the Bonferroni method.

Table 2.

Comparison of 41 transcripts between stroke and control subjects

Transcript	Cellular source (reference)	Fold change	p value	Adjusted p value*	Adjusted p value**
CD163	PBMC4	2.22	0.069	0.14	1.0
PLBD1	PBMC4	3.18	0.0034	0.03	0.14
ADM	PBMC4	1.85	0.0066	0.03	0.27
KIAA0146	PBMC4	1.21	0.43	0.52	1.0
APLP2	PBMC4	1.08	0.56	0.62	1.0
NPL	PBMC4, WB5	1.67	0.094	0.16	1.0
FOS	PBMC4	2.64	0.043	0.10	1.0
TLR2	PBMC4	1.37	0.57	0.62	1.0
NAIP	PBMC4	1.71	0.24	0.34	1.0
CD36	PBMC4	2.11	0.29	0.10	1.0
DUSP1	PBMC4	2.89	0.033	0.10	1.0
ENTPD1	PBMC4	2.03	0.039	0.10	1.0
VCAN	PBMC4, WB6	2.36	0.058	0.13	1.0
CYBB	PBMC4	2.61	0.0083	0.04	0.34
IL13RA1	PBMC4	1.58	0.10	0.16	1.0
LTA4H	PBMC4	1.61	0.20	0.30	1.0
ETS2	PBMC4, WB5	2.86	0.017	0.07	0.70
CD14-1	PBMC4	1.93	0.065	0.14	1.0
CD14-2	PBMC4	1.39	0.74	0.78	1.0
BST1	PBMC4	6.42	0.0035	0.03	0.14
CD93	PBMC4	2.11	0.00086	0.02	0.03
PILRA	PBMC4	1.29	0.56	0.62	1.0
FCGR1A	PBMC4	3.28	0.076	0.14	1.0
CKAP4	WB5	1.93	0.0040	0.03	0.14
S100A9	WB5	3.84	0.0014	0.02	0.06
MMP9	WB5,6	2.21	0.10	0.16	1.0
S100P	WB5	2.67	0.0399	0.10	1.0
F5-1	WB5	2.14	0.034	0.10	1.0
FPR1	WB5	1.79	0.07	0.14	1.0
S100A12	WB5,6	2.93	0.000593	0.02	0.02
RNASE2	WB5	1.06	0.84	0.86	1.0
ARG1	WB5,6	1.34	0.34	0.42	1.0
CA4	WB5,6	1.74	0.17	0.27	1.0
LY96	WB5,6	1.41	0.27	0.36	1.0
SLC16A6	WB5	1.64	0.23	0.34	1.0
HIST2H2AA3	WB5	1,48	0.25	0.34	1.0
BCL6	WB5	0.97	0.58	0.62	1.0
PYGL	WB5	2.55	0.0059	0.03	0.24
CCR7	WB6	0.995	0.96	0.96	1.0
IQGAP1	WB6	1.67	0.04	0.10	1.0
ORM1	WB6	1.28	0.31	0.40	1.0

^*FDR method,

^**Bonferroni method

Wilcoxon rank sum tests and t tests used for analyses

Figure 1. Stroke-Related Transcript Fold Changes Transcripts with p<0.05.

This bar graph shows the fold change levels for the 16 up-regulated transcripts in ischemic stroke.

41% (9/22) of genes from the PBMC list were significantly altered. 38% (8/21) from the whole blood gene lists were significantly altered, 7 genes were on Tang et al. list and 2 were on Barr et al. list. One transcript was common to both the whole blood and PBMC lists (this was ETS2) and one transcript was common to both WB lists (S100A12). Although, modest correlations with age for two transcripts were identified - FOS (rho=0.42, p=0.02) and PYGL (rho=0.43, p=0.02), after corrections for multiple comparisons these correlations were no longer significant. There was no correlation of transcript copy number with gender or with the time of blood draw.

3.2. Clusters of Genes in Whole Blood

Seven clusters of between 3 and 7 transcripts were identified in a hierarchical cluster analysis. (Figure 2). Six of these showed significant discrimination between stroke and control with p values for the six clusters ranging between 1.10e-9 for Cluster 1 and 0.037 for Cluster 7 (Table 3). After correction for multiple comparisons using the Bonferroni method all but one remained significant. The clusters consisted of transcripts from both whole blood and PBMC studies. Based on the demonstration of the most significant discrimination between stroke and control Cluster 1 was selected for further study.

Figure 2. Heatmap and Hierarchical Cluster Analysis.

This heatmap and hierarchical cluster analysis illustrates gene expression levels for the 41 studied genes in the control subjects (C) and stroke patients (S). Seven clusters (cl.1 to cl.7) are highlighted by seven squares of different color. Data are log-transformed. This demonstrates elevated expression of many transcripts in stroke patients relative to controls.

Table 3.

Transcript clusters identified in hierarchical cluster analysis

Transcripts	Cellular sources (number of genes)	P value, of cluster, stroke versus control	Adjusted p value*	Adjusted p value**
Cluster 1 7 genes PLBD1, PYGL, FOS, DUSP1, BST1, VCAN, FCGR1A	PBMC panel (5), both PBMC and WB panels (1), WB panel (1)	1.01e-9	7.04 e-9	7.04e-9
Cluster 2 7 genes PILRA, BCL6, FPR1, LY96, S100A9, S100A12, MMP9	WB panels (6), PBMC panel (1)	1.50e-6	5.26e-6	1.05e-5
Cluster 3 7 genes NPL, IQGAP1, CYBB, SLC16A6, LTA4H, CA4, CD14-1	PBMC panel (3), PBMC and WB panels (1), WB panels (3)	1.52e-5	3.55e-5	1.06e-4
Cluster 4 6 genes ADM, HIST2H2AA3, CD93, CKAP4, CD14- 2, TLR2	PBMC panel (4), WB panel (2)	0.0016	0.0022	0.011
Cluster 5 3 genes NAIP, RNASE2, CCR7	WB panels (2), PBMC panel (1)	0.40	0.40	1.0
Cluster 6 6 genes CD163, S100P, F5, ETS2, ARG1, ORM1	WB panels (4), PBMC panel (1), WB and PBMC panels (1)	0.00025	4.0e-4	1.61e-3
Cluster 7 5 genes APLP2, IL13RA1, ENTPD1, KIAA0146, CD36	PBMC panel (5)	0.037	0.042	0.26

Wilcoxon rank sum tests used for analyses ,

^*FDR.

^**Bonferroni

3.3. Performance of a 7 Gene Cluster for Stroke Classification

The 7 transcript Cluster 1 consisted of PLBD1, PYGL, BST1, DUSP1, FOS, VCAN and FCGR1A (Table 3 and Table 4). Five of these transcripts had differed significantly between stroke and control. The upper threshold levels for each of the transcripts were based on the third quartile in the control subjects (Table 4 and Figure 3A). Absent calls were noted for BST1 and FCGR1A in a number of the control subjects (this could reflect low or absent transcript expression). The number of subjects with elevated expression of each transcript is shown in Table 4. The proportion of patients with elevated expression of between 0-7 transcripts is shown in Figure 3B.

Table 4.

Performance of a 7 gene cluster - Cluster 1

Transcript	Threshold	Stroke Number of subjects with elevated transcript copy number (%)	Control Number of subjects with elevated transcript copy number (%)	p
PLBD1	>0.0144	15/18 (83%)	3/14 (21%)	0.0017
PYGL2	>0.0115	13/17 (76%)	3/12 (25%)	0.02
FOS	>0.0122	10/17 (59%)	3/13 (23%)	0.11
DUSP1	>0.0052	13/17 (76%)	2/11 (18%)	0.008
BST1	>0.0073	14/16 (88%)	2/7 (28%)	0.02
VCAN	>0.0101	12/17 (70%)	3/9 (33%)	0.16
FCGR1A	>0.0202	10/14 (71%)	2/6 (33%)	0.27
7 transcripts in Cluster 1	3 or more transcripts elevated	15/18 (83%)	3/15 (20%)	0.001

Figure 3. Characteristics of 7 transcript classifier for ischemic stroke detection.

(A) Boxplots demonstrating the threshold values for defining elevated expression of each of the transcripts (PLBD1, PYGL, FOS, DUSP1, BST1, VCAN, FCGR1A). The threshold was set at above the third quartile value in the control group (dashed line on each boxplot). The threshold value was the normalized transcript copy number. (B) Bar graphs depicting the number of transcripts elevated in the stroke patients and the control subjects. In the stroke bar the value for the 7 transcript elevation represents the 5 stroke patients who had all 7 transcripts elevated, the value for 6 transcripts represents the 5 patients who had 6 transcripts elevated, the value for 5 transcripts elevated represents the one stroke patient who had 5 transcripts elevated, etc. In Cluster 1, 83% (15/18) of the stroke patients had 3 or more of the 7 transcripts elevated while 20% (3/15) of the control group showed elevation of 3 or more of the 7 transcripts. Hence the sensitivity was 83% and the specificity was 80%. (C) ROC Analysis for Cluster 1 for Stroke Classification revealed that the AUC was 0.854. Elevation of 3 or more transcripts gave the greatest sensitivity and specificity.

Elevated whole blood expression of at least 3 transcripts in this 7 gene cluster classified stroke with a sensitivity of 83% and a specificity of 80% (Figure 3B). The overall accuracy of the 7 gene classifier was high (AUC=0.854, Figure 3C).

3.4. Performance of Three Previously Reported Transcript Panels

The Moore et al. transcripts list[4], identified in PBMCs, showed a highly significant discrimination between stroke and control (p=1.01e-9). The p values for the Tang et al. list[5] and the Barr et al. list[6], identified in whole blood, were 1.05e-5 and 0.02 respectively.

4. Discussion

The diagnostic utility of gene expression changes in acute ischemic stroke has been studied in a number of prior microarray studies[4-7,21,22]. However these microarray results were never validated with qPCR – the gold standard for measuring gene expression. The current study was based on three studies where gene panels had been identified using the Prediction Analysis for Microarrays[4-6] algorithm. The Grond-Ginsbach et al. study[7] was not included as only one transcript was identified and pooled samples were used. Results from Oh et al.[21] were published after this study commenced. In several other microarray studies the utility of gene expression was investigated: for the evaluation of the risk of hemorrhagic transformation[23], defining stroke etiology[24-26] and in studying gender related gene expression changes in stroke patients[27,28].

Using HT RT-qPCR, for the first time – a new qPCR based platform that has the advantages of high accuracy and sensitivity - we have found that 40% of the transcripts were up-regulated in stroke. It is arguable as to whether corrections for multiple comparisons were needed as these transcripts were a priori specified. Nevertheless even after correction for multiple comparisons expression of a small number of transcripts were still significantly different between stroke and control. Although, the hierarchical cluster analysis was not used previously it proved to be very successful in detecting association of studied transcripts with stroke. This analysis grouped genes into 7 clusters. These clusters were highly significantly different between the stroke patients and the control subjects, with 5 remaining highly significant after stringent correction for multiple comparisons (p values as low as 7.04e-9). The cluster of 7 genes - PLBD1, PYGL, BST1, DUSP1, FOS, VCAN and FCGR1A - classified stroke with high sensitivity and specificity, respectively 80% and 83%. The similar expression of genes within a cluster in the stroke patients and control subjects, with comparable differences between two groups permitted the analysis of the expression of all genes within a cluster together. Furthermore, the quantitative information on copy number permitted threshold levels of normal and abnormal expression to be established in the control subjects.

Of interest is that while whole blood samples were used in this study, the 16 significantly altered transcripts had been identified in whole blood and in PBMCs, and that transcripts within each of the 7 identified clusters came from both whole blood and PBMC gene lists. These two cell populations overlap substantially because whole blood is composed of PBMC and polymorphonuclear leukocytes (granulocytes). Neutrophils are the main cell population within polymorphonuclear leukocytes and represent the most numerous nucleated cell fraction in whole blood, however their RNA content is almost three times lower than in PBMC[18] . The overlap between panels and detection of PBMC gene alterations in whole blood samples supports the validity of the microarray results.

Accurate and rapid stroke diagnosis is crucial for timely and effective treatment in the acute phase. Diagnosis is also necessary in subacute and delayed phase to evaluate future risks and for optimal prevention strategies. Timely diagnosis is absolutely essential for treating patients with tissue plasminogen activator- the only FDA approved treatment of ischemic stroke. However, this treatment improves the chances of recovering from stroke only if administered within 3 to 4.5 hours. Stroke diagnosis may not be conclusive in the acute phase of stroke, especially in stroke mimics, while in subacute or chronic stroke, silent stroke, inconclusive brain imaging or atypical stroke presentation may confound stroke diagnosis. Hence, additional tests, such as molecular tests, that could confirm a diagnosis of stroke, or add complementary information, are much needed.

Molecular diagnostic tests based on gene expression patterns are now available in number of diseases, including breast, colon, lung, prostate and thyroid cancers. These tests have been based on microarray identified panels and detection of overall signal intensities rather than measurements of the expression of individual genes[14,29,30]. For stroke diagnosis, methods that are highly complex, labor intensive and require expensive equipment are difficult to be applied to clinical practice and need to be available rapidly. HT RT-qPCR is very promising and can address this problem[18]. HT RT-qPCR permits absolute quantification and measuring gene expression adjusted to the input cell count, is independent of control genes. An HT RT-qPCR identified classifier may be used to develop a point-of-care system for stroke diagnosis. Until now stroke gene expression panels established in microarray studies consisted of 18 to 79 genes[4,5,22]. Clusters of 5-7 genes established using HT RT-qPCR are more feasibly applied in clinical setting, where short turnaround times and low detection limits are crucial. We have recently discussed the requirements of this system and described a highly sensitive gene expression profiling method that can be measured almost in real time[15].

5. Conclusions

In summary, a proportion of previously reported genes in microarray studies in stroke were replicable using HT RT-qPCR and all except 3 were grouped together to form gene clusters highly significant for ischemic stroke detection. Grouping genes in clusters allowed the identification of gene expression classifiers that could be used in a point-of-care system. These results show the promise and potential for continuing studies of gene expression profiling in stroke and for further assessment of the sensitivity and specificity of transcript clusters for ischemic stroke detection and diagnosis. Further studies will examine the gene expression changes in terms of cellular source, time course and relation to clinical outcome.

Highlights.

• Next generation high throughput qPCR was used to validate microarray results 40% of whole blood ischemic stroke-related transcripts were validated

• Six clusters of between 5 and 7 genes were identified for stroke detection

• A 7-gene cluster is highly diagnostic for ischemic stroke (p=1.01x10⁻⁹; AUC=0.85)

Abbreviations

HT RT-qPCR

high throughput reverse transcription quantitative polymerase chain reactions

IS

ischemic stroke

NIHSS

National Institutes of Health Stroke Scale

CBC

complete blood counts

WBC

white blood cell

FDR

false discovery rate