Peripheral Blood MCEMP1 Gene Expression as a Biomarker for Stroke Prognosis
Background and Purpose—A limitation when making early decisions on stroke management is the lack of rapid diagnostic and prognostic testing. Our study sought to identify peripheral blood RNA biomarkers associated with stroke. The secondary aims were to assess the discriminative capacity of RNA biomarkers for primary stroke type and stroke prognosis at 1-month.
Methods—Whole-blood gene expression profiling was conducted on the discovery cohort: 129 first-time stroke cases that had blood sampling within 5 days of symptom onset and 170 control participants with no history of stroke.
Results—Through multiple regression analysis, we determined that expression of the gene MCEMP1 had the strongest association with stroke of 11 181 genes tested. MCEMP1 increased by 2.4-fold in stroke when compared with controls (95% confidence interval, 2.0–2.8; P=8.2×10−22). In addition, expression was elevated in intracerebral hemorrhage when compared with ischemic stroke cases (P=3.9×10−4). MCEMP1 was also highest soon after symptom onset and had no association with stroke risk factors. Furthermore, MCEMP1 expression independently improved discrimination of 1-month outcome. Indeed, discrimination models for disability and mortality that included MCEMP1 expression, baseline modified Rankin Scale score, and primary stroke type improved discrimination when compared with a model without MCEMP1 (disability Net Reclassification Index, 0.76; P=3.0×10−6 and mortality Net Reclassification Index, 1.3; P=1.1×10−9). Significant associations with MCEMP1 were confirmed in an independent validation cohort of 28 stroke cases and 34 controls.
Conclusions—This study demonstrates that peripheral blood expression of MCEMP1 may have utility for stroke diagnosis and as a prognostic biomarker of stroke outcome at 1-month.
Stroke is the second leading cause of death worldwide and a major cause of disability.1,2 Stroke diagnosis is dependent on clinical assessment and neuroimaging. However, the lack of rapid diagnostic testing hinders patient management. Although an effective ischemic stroke treatment is available, tissue-type plasminogen activator,3 studies have observed an underuse by rural4 and emergency physician5,6 owing, in part, to diagnostic uncertainty, risk of hemorrhage, and the short therapeutic time window. However, a biomarker that establishes diagnosis of stroke and distinguishes hemorrhage from ischemia has the potential to minimize the time from symptom onset to treatment and improve patient outcomes. Determining a patient’s risk for disability or mortality may also be used to inform clinical decision making, evaluate risk–benefit, and optimize allocation of healthcare resources. Indeed, although one third of patients with stroke die or experience disability within the first month,2 clinical risk scores to predict patient outcome are infrequently used by clinicians because of lack of precision, validation, and complexity. Identification of biomarkers that quickly distinguish stroke cases from controls, ischemia from hemorrhage, and predict prognosis could improve patient management.
The advent of high-throughput genomic technology provides a novel, agnostic approach for biomarker discovery. RNA gene expression levels vary rapidly in response to physiological changes. Rapid point-of-care RNA tests are currently in development7; therefore, peripheral blood RNA may be used in the clinic in the future. Both animal8 and human9,10 studies have observed unique RNA expression changes in whole blood after ischemic stroke. However, previous clinical studies were conducted on a relatively small sample size, consisting of a maximum of 39 ischemic stroke cases and 25 controls9,10 and still require validation. In addition, these studies have only assessed RNA biomarkers for stroke diagnosis, but not for stroke prognosis. Because of the clinical need for stroke biomarkers and heterogeneity in stroke pathophysiology, large studies are crucial to robustly identify novel biomarkers and to assess clinical value.
In this report, we used a large discovery population, 299 participants from a multination case-control study (INTERSTROKE; 129 stroke cases and 170 controls), to identify novel RNA biomarkers of stroke. We then determined whether RNA biomarkers distinguished between primary stroke types and stroke outcome. Significant results were validated in an independent group of participants (28 stroke cases and 34 controls).
The INTERSTROKE study has been described in detail elsewhere.11 Briefly, INTERSTROKE was a large, international, standardized case–control study consisting of stroke cases and control participants from 22 countries. Stroke cases were patients admitted to the hospital with first-time acute stroke who presented within 5 days of symptom onset and within 72 hours of hospital admission. Distinction between stroke subtypes was confirmed with neuroimaging (computed tomography or magnetic resonance imaging). Control participants were recruited from the hospital or within the community and had no history of stroke. Three hundred and seventy-five INTERSTROKE participants recruited from 6 centers consented to the expression profiling substudy. Our expression study benefitted from the international recruitment by the increased ethnic diversity, and the greater prevalence of intracerebral hemorrhage (ICH) in South America12 that increased the number of samples available for analysis. Peripheral whole blood was collected into PAXgene Blood RNA tubes (PreAnalytiX) and stored at −80°C before sample processing.
Only 2 patients with subarachnoid hemorrhage had blood samples collected, and so were excluded from the analysis. Ninety-nine percent of participants were either Latin American or white, so we excluded participants of other ethnicities (n=9) to reduce potential population stratification. As a result, our study consisted of 364 participants. Initially, 302 participants (131 cases and 171 controls) were recruited and consecutively assigned for biomarker discovery. An additional 62 participants (28 cases and 34 controls) were recruited for independent validation
Sample Processing and Microarray Hybridization
Total RNA was isolated from the discovery cohort using the QIAsymphony PAXgene Blood RNA kit (Qiagen) according to the manufacturer’s protocol. RNA was isolated from the validation cohort using the MagMAX Stabilized Blood Tube RNA Isolation kit (LifeTech). RNA quality was assessed with Nanodrop2000 (Nanodrop) and 2100 Bioanalyzer (Agilent), then quantified using Quant-IT RiboGreen (LifeTech). Total RNA was amplified and biotinylated using the Illumina TotalPrep RNA Amplification Kit (LifeTech). Samples were then hybridized to Illumina HumanRef-8v4 BeadChips (Illumina) and scanned on the iScan System (Illumina) as per manufacturer protocol.
Microarray Data Preprocessing
The Illumina HumanRef-8v4 BeadChip interrogates expression of 34 694 unique genes using 47 323 probes. The raw sample probe profile and control probe profile were exported from GenomeStudio version1.9.0 (Illumina). All analysis was performed in R (http:www.//r-project.org).
In the discovery cohort, 3 samples did not pass quality control metrics and were excluded from further analysis. Data preprocessing involved background correction using the nongenomic control probes,13 quantile normalization and log2 transformation.14 Probes with detection P<0.01 in >50% of the samples were considered expressed.
Microarrays (and quantitative polymerase chain reaction [PCR]) measure relative rather than absolute gene expression, or in other words, the relative increase or decrease in the expression of a gene when compared with global expression (or housekeeping genes). Differential gene expression was thus reported as fold change (FC), with 95% confidence intervals (CIs). Regression models were used to identify RNA transcripts associated with stroke in the discovery cohort. Each model tested a single gene’s association with stroke while adjusting for sex, age, body mass index, ethnicity, and hybridization chip. The hybridization chip variable acted as a surrogate for batch effect and other unwanted technical variation.15 To correct for multiple hypotheses testing, a conservative Bonferroni correction was applied, setting the significance threshold at 0.05/11 181=4.5×10−6. As external validation, we assessed the significance of genes reported to be associated with stroke by Tang et al9 and Barr et al.10
Further analysis was conducted on the most significant transcript associated with stroke in the discovery cohort. Regression was used to assess the association between expression and stroke risk factors. The relationship between expression and hours from symptom onset was assessed using regression and t tests. Comparison of expression between controls and primary stroke types, hemorrhagic and ischemic stroke, and between ischemic subtypes was performed with t tests. We used ordinal logistic regression to evaluate the association between functional disability, measured as modified Rankin Scale (mRS) and gene expression. Our analysis used mRS recorded soon after the stroke (at baseline) and at the 1-month follow-up. One-month mRS was also dichotomized to represent either functional disability (mRS=0–2 versus mRS>2) or mortality (mRS=0–5 versus mRS=6). Using pROC,16 receiver operator curves were constructed from logistic regression models for the dichotomized outcomes. Area under the receiver operator curve (AUC) was determined as a measure of sensitivity and specificity. The odds ratio (OR), positive predictive value, and negative predicative value were determined based on the optimal univariate receiver operator curve expression threshold. The continuous Net Reclassification Index17 was calculated using Hmisc18 to compare multiple discrimination models. An Net Reclassification Index >0.6 was considered a strong improvement in discriminative capacity, 0.4 was intermediate, and 0.2 was considered weak.
Quantitative PCR Validation and Replication
For quantitative real-time PCR (qPCR), complementary DNA was synthesized using the QuantiTect Reverse Transcription Kit (Qiagen). TaqMan qPCR was performed on a Viia7 Real-Time System (LifeTech) where MCEMP1, monitored with Hs00545333_g1 (LifeTech), was normalized to ACTB, monitored with Hs01060665_g1 (LifeTech). Cycle threshold values were calculated automatically with default parameters, and FC was calculated using the δCT method.19 qPCR confirmed microarray results if Pearson correlation >0.8 and regression P<0.05. The independent validation cohort was analyzed using 1-sided t tests and ordinal logistic regression.
Between March 2007 and April 2010, 364 INTERSTROKE participants consented to the gene expression substudy. Biomarker discovery was conducted on 299 samples (129 stroke cases and 170 controls) that passed quality control. Sixty-two additional participants (28 stroke cases and 34 controls) were recruited as an independent validation cohort. Patient demographics for the discovery and validation cohorts are presented in Table I in the online-only Data Supplement. Among stroke cases in the discovery cohort, 19.4% (n=25) were ICH and 80.6% (n=104) were ischemic. On the basis of Trial of ORG 10172 in Acute Stroke Treatment criteria,20 21.7% of the ischemic strokes were classified as cardioembolic, 7.8% large vessel, 15.5% small vessel, 24% cryptogenic, and 11.6% other. Clinical features of stroke cases were similar to controls in the discovery cohort, except for presence of hypertension (P=0.02), migraine (P=4.5×10−3), and smoking (P=0.01), all of which were more common among stroke cases.
Association Between Gene Expression and Stroke
Microarray expression profiling was conducted in the discovery cohort (129 stroke cases and 170 controls). Each of the 11 181 RNA probes were tested for association with stroke and 19% were significantly associated after Bonferroni correction (P<4.5×10−6; Figure IA in the online-only Data Supplement). As external validation, we compared our significant associations with the 18 genes previously associated with stroke by Tang et al9 and 9 genes by Barr et al.10 We observed that 81.2% of genes identified by Tang et al9 and 77.8% of genes identified by Barr et al10 had genome-wide significant association with stroke in our data (Tables II and III in the online-only Data Supplement). The direction of effect was consistent between our study and previous reports for all replicated genes.
The Table presents the 10 most significant genes associated with stroke in our discovery cohort. The most significant gene identified was MCEMP1, which had a 2.4-fold expression increase in stroke cases when compared with controls (95% CI, 2.0–2.8; P=8.2×10−22; Figure 1A). The AUC of MCEMP1 for stroke was 0.81 (95% CI, 0.76–0.86; Figure IIA in the online-only Data Supplement). To test whether multiple probes were nonredundantly associated with stroke, we included MCEMP1 expression in the initial association models as a covariable and tested all 11 180 remaining probes. Under this model, the most significant gene was MSRA (P=4.6×10−6), which did not reach our threshold for statistical significance after Bonferroni correction (Figure IB in the online-only Data Supplement).
Differential expression of MCEMP1 was verified using qPCR in a subset of the discovery cohort (76 stroke cases and 66 controls). We observed high correlation between MCEMP1 expression levels measured by qPCR and microarray (r2=0.88 and P=4.8×10−48). Using qPCR, a 2.4-fold increase in MCEMP1 was detected in stroke cases when compared with controls (95% CI, 1.8–3.2; P=1.6×10−8; Figure III in the online-only Data Supplement).
MCEMP1 Expression Is Not Associated With Stroke Risk Factors
Restricting the analysis to healthy controls (n=170), we tested MCEMP1 for association with stroke risk factors, including age, sex, body mass index, ethnicity, hyperlipidemia, diabetes mellitus, atrial fibrillation, hypertension, migraine, and smoking status. After adjustment for multiple hypothesis testing, we observed no association between MCEMP1 and stroke risk factors (P>0.05/9=0.0056; Table IV in the online-only Data Supplement). A modest association between elevated MCEMP1 and hypertension was identified (FC=1.2; 95% CI, 1.1–1.4; P=8.8×10−3). However, adjusting the initial stroke association model for all available risk factors did not attenuate the association with MCEMP1 (FC=3.4; 95% CI, 2.4–4.9; P=2.6×10−11).
MCEMP1 Expression Is Associated With Time From Symptom Onset
INTERSTROKE case participants were recruited at varying times after symptom onset. Because the time from symptom onset to blood sampling varied between individuals, we were able to assess the temporal profile of MCEMP1 change after stroke. We identified a significant relationship where MCEMP1 decreased by 1% per hour from symptom onset (95% CI, 0.98–1.0; P=3.7×10−3; Figure 2), even after adjustment for stroke risk factors. Separating stroke cases by primary stroke type, we also observed that MCEMP1 decreased as time from symptom onset increased (Figure IV in the online-only Data Supplement). Furthermore, MCEMP1 was highest in samples collected <24 hours of symptoms onset when compared with controls (FC=5.3; 95% CI, 3.2–8.5; P=1.7×10−6) or stroke cases collected >24 hours (FC=1.9; 95% CI, 1.4–3.9; P=1.9×10−3).
MCEMP1 Expression Differs Between Stroke Types
MCEMP1 was increased by 4.5-fold in ICH cases when compared with controls (95% CI, 3.1–6.4; P=3.4×10−9) and by 2.1-fold in ischemic cases when compared with controls (95% CI, 1.8–2.6; P=3.4×10−13). Accordingly, a 2.1-fold increase in MCEMP1 was observed in ICH cases when compared with ischemic (95% CI, 1.4–3.1; P=3.9×10−4; Figure 1B). The area under the receiver operator curve for primary stroke type discrimination by MCEMP1 was 0.75 (95% CI, 0.65–0.85; Figure IIB in the online-only Data Supplement). Expression differences were also detected between ischemic stroke subtypes. MCEMP1 was elevated in cardioembolic (FC=1.5; 95% CI, 1.1–2.1; P=8.1×10−3) and large-vessel (FC=2.3; 95% CI, 1.2–4.1; P=0.012) stroke when compared with small-vessel stroke (Figure 1C).
Baseline and 1-Month mRS Associated With MCEMP1 Expression
Baseline mRS, measured soon after stroke, was associated with MCEMP1 expression (P=4.0×10−13; Figure V in the online-only Data Supplement). One unit MCEMP1 expression increase was associated with a 3.3 odds (95% CI, 2.4–4.5) increase in baseline mRS. The association remained significant after adjustment for stroke risk factors, stroke type, tissue-type plasminogen activator treatment, and hours from symptom onset (OR=3.1; 95% CI, 2.4–4.5; P=1.8×10−9).
One-month mRS was also associated with MCEMP1 expression (P=1.3×10−14; Figure VI in the online-only Data Supplement). One unit MCEMP1 expression increase was associated with a 3.4 odds (95% CI, 2.5–4.6) increase in 1-month mRS and the association remained significant after adjustment for stroke risk factors, primary stroke type, tissue-type plasminogen activator treatment, hours from symptom onset, and baseline mRS as a categorical variable (OR=1.8; 95% CI, 1.2–2.8; P=6.6×10−3). In fact, only MCEMP1 expression (P=6.6×10−3), baseline mRS (P=3.2×10−3-6.4×10−10), and primary stroke type (P=1.0×10−3) were independently associated with 1-month mRS.
MCEMP1 Expression Is Associated With Disability at 1-Month
To further explore the association, we dichotomized 1-month mRS into 2 groups: mRS of 0, 1, or 2 and mRS >2, representing the ability to live autonomously or not post stroke. Individuals with disability at 1-month had elevated baseline MCEMP1 when compared with controls (FC=4.7; 95% CI, 3.5–5.7; P=1.6×10−19) or individuals without disability (FC=3.2; 95% CI, 2.5–4.2; P=1.8×10−14; Figure 3A). A disability discrimination model, including MCEMP1 expression, primary stroke type, and baseline mRS as a categorical variable, strongly improved discrimination when compared with a model including only primary stroke type and baseline mRS (AUC with MCEMP1=0.96, AUC without MCEMP1=0.93, Net Reclassification Index=0.76; P=3.0×10−6). The optimal MCEMP1 threshold had a specificity of 80.3%, sensitivity of 86.2%, with corresponding positive predictive value of 78.1% and negative predicative value of 87.7%, for disability (Table V in the online-only Data Supplement). The OR for disability was 6.6 (95% CI, 1.9–22.7) in individuals with high versus low MCEMP1 expression after adjustment for stroke type and baseline mRS.
MCEMP1 Expression Is Associated With Mortality at 1-Month
MCEMP1 was also associated with 1-month mortality after adjustment for stroke risk factors, baseline mRS, primary stroke type, tissue-type plasminogen activator treatment, and hours from symptom onset (FC=3.8; 95% CI, 1.4–11.0; P=9.9×10−3; Figure 3B). Comparing univariate 1-month mortality discrimination models, MCEMP1 appeared more informative (AUC=0.88) than primary stroke type (AUC=0.80). Moreover, a model including MCEMP1, primary stroke type, and baseline mRS, as a categorical variable, strongly improved mortality discrimination when compared with a model without MCEMP1 (AUC with MCEMP1=0.97, AUC without MCEMP1=0.92, Net Reclassification Index=1.3; P=1.1×10−9). Selecting the optimal discrimination threshold, MCEMP1 had a specificity of 97.8%, sensitivity of 35.1%, positive predictive value of 86.7%, and negative predicative value of 78.9% for mortality (Table VI in the online-only Data Supplement). The OR for mortality was 20.7 (95% CI, 2.5–174.6) in individuals with high when compared with low expression, after adjustment for baseline mRS and stroke type.
Replication of MCEMP1 Associations in Validation Cohort
The significance of MCEMP1 was validated in a small independent cohort (28 stroke cases and 34 controls) using qPCR. Power calculations indicated that we had sufficient power (>99%) to detect an expression difference of 2.4-fold between cases and controls, similar to the discovery cohort, at a significance of P<0.05. We detected increased MCEMP1 in stroke cases when compared with controls (FC=1.6; P=0.039; Figure VII in the online-only Data Supplement). We also observed trends toward higher MCEMP1 in ICH cases than in controls (FC=5.6; P=0.05), higher expression in ischemic cases than in controls (FC=1.3; P=0.14), and higher expression in ICH cases than in ischemic (FC=4.4; P=0.074). Finally, both baseline mRS and 1-month mRS were associated with MCEMP1 expression (baseline, P=0.049; 1-month, P=3.3×10−3).
The present study evaluated peripheral blood gene expression in a subgroup of INTERSTROKE participants. We identified elevated expression of a novel gene, MCEMP1, in stroke cases when compared with controls. MCEMP1 decreased as time from symptom onset increased and expression was increased in hemorrhagic stroke cases when compared with ischemic. We also identified an association between 1-month mRS and MCEMP1, where increased functional disability and mortality were associated with increased MCEMP1. One-month prognosis discrimination models that included MCEMP1, primary stroke type, and baseline mRS improved discrimination when compared with similar models without MCEMP1. The associations between MCEMP1 with stroke, primary stroke type, and stroke prognosis were independently confirmed in the validation cohort.
Our results demonstrate that noninvasive measurement of MCEMP1 soon after stroke provides additional prognostic information to clinical characteristics. We observed that MCEMP1 decreased as time from symptom onset increased, suggesting that MCEMP1 may have utility for estimating time from symptom onset. In addition, identifying patients with poor prognosis may be beneficial for informed clinical decision making and assessing the risk–benefit ratio for acute therapies. Although several clinical scores have been proposed to predict outcome and mortality,21–23 these scores are not routinely used in the clinic, in part, because of their complexity. We have shown that a simple model including only baseline mRS, primary stroke type, and MCEMP1 expression may predict 1-month disability and mortality. Indeed, inclusion of MCEMP1 strongly improved discrimination of 1-month prognosis.
Mast cell expressed membrane protein 1 (MCEMP1), also known as C19ORF59, is a transmembrane protein expressed by mast cells,24 macrophages and other tissues.25 The exact function of MCEMP1 has yet to be determined; however, the gene’s promoter region contains nuclear factor kappa-light-chain-enhancer of activated B cell and nuclear factor of activated T-cell binding motifs, similar to many immune receptor genes.24 Although limited research has been conducted on MCEMP1, recent findings indicate an emerging role for brain resident mast cells in acute stroke. Experimental stroke studies have reported that mast cells are first responders to cerebral ischemia and act as early regulators of blood–brain barrier permeability.26–28 The increase in MCEMP1 expression observed in stroke cases may be the result of cerebral mast cell activation and mast cell–mediated blood–brain barrier disruption. Furthermore, the expression difference detected between the primary stroke types and ischemic stroke subtypes may indicate an association between MCEMP1 expression and infarct size.
Whole genome expression after ischemic stroke has been previously assessed9,10 in small studies including 15 to 39 stroke cases. We used a larger discovery cohort of 129 stroke cases and 170 controls, and verified expression of 77.8% to 81.2% of previously reported genes. To our knowledge, our study shows the largest proportion of overlap with previous reports, providing confidence in both novel and known results. In addition, a recent study reported significantly elevated MCEMP1 in 12 ischemic stroke cases when compared with 12 controls.29 Although the study had a small sample size and lacked replication, stroke samples were collected within 24 hours of symptom onset, thus further positioning MCEMP1 as a marker of acute stroke. Our study robustly identified MCEMP1 as a stroke biomarker in a significantly larger, multiethnic population and confirmed our findings in an independent validation cohort. The overlap and concordance between our study and previous works add credence to our findings and the significance of MCEMP1 as an acute stroke biomarker.
There are a few study limitations that warrant further discussion. First, our study lacks stroke mimics and nonstrokes such as transient ischemic attacks, migraines, seizure, and other neurological or inflammatory disorders. Thus, we were unable to assess the specificity of MCEMP1. Second, an acute stroke biomarker would have the greatest clinical utility if increased concentrations were detected shortly after symptom onset. In our discovery cohort, only 17 samples were collected within 24 hours of symptom onset, but we observed elevated MCEMP1 expression in these samples when compared with controls or stroke samples collected after 24 hours. Nonetheless, early sampling (<6 hours) will be required to assess the utility of MCEMP1 in guiding acute stroke treatment. Third, because of the nature of the INTERSTROKE study design, there were limited neurological imaging data available, and, consequently, the effect of infarct size on expression could not be assessed. However, MCEMP1 may provide useful information early after stroke onset at times where stroke volume is difficult to assess with plain computed tomographic scans. Additional studies including neurological imaging and National Institutes of Health Stroke Scale score will be useful to further characterize the association between MCEMP1 and stroke severity. Finally, our study focused on a single gene to differentiate between the various stroke groups, but there may be other genes with diagnostic potential.
The results of the study demonstrate that MCEMP1 expression has prognostic capacity beyond baseline mRS and stroke type. MCEMP1 may also have diagnostic capacity. These observations are promising given the currently limited number of simple clinical tools available to predict outcome and mortality and lack of an established noninvasive stroke biomarker. The results also point toward an important role for mast cells in stroke, and unraveling the biological mechanisms may lead to the identification of new therapeutic targets. Additional clinical studies including a stroke mimic cohort, early blood sampling, and measurement of stroke severity will help determine the diagnostic capacity and clinical utility of MCEMP1.
Sources of Funding
This study was supported by Canadian Stroke Network, New Investigator Funds (HHS), and Heart and Stroke Foundation of Canada. Dr Pare was supported by a Canada Research Chair and a CISCO Professorship in Integrated Health Biosystems. Dr Shoamanesh was supported by the Marta and Owen Boris Chair in Research and Stroke Care.
The online-only Data Supplement is available with this article at http://stroke.ahajournals.org/lookup/suppl/doi:10.1161/STROKEAHA.115.011854/-/DC1.
- Received October 16, 2015.
- Revision received January 7, 2016.
- Accepted January 12, 2016.
- © 2016 American Heart Association, Inc.
- Go AS,
- Mozaffarian D,
- Roger VL,
- Benjamin EJ,
- Berry JD,
- Borden WB,
- et al
- Emberson J,
- Lees KR,
- Lyden P,
- Blackwell L,
- Albers G,
- Bluhmki E,
- et al
- Scott PA,
- Xu Z,
- Meurer WJ,
- Frederiksen SM,
- Haan MN,
- Westfall MW,
- et al
- Tang Y,
- Xu H,
- Du X,
- Lit L,
- Walker W,
- Lu A,
- et al
- O’Donnell MJ,
- Xavier D,
- Liu L,
- Zhang H,
- Chin SL,
- Rao-Melacini P,
- et al
- Shi W,
- Oshlack A,
- Smyth GK.
- Johnson WE,
- Li C,
- Rabinovic A.
- Harrel FE Jr
- Saposnik G,
- Kapral MK,
- Liu Y,
- Hall R,
- O’Donnell M,
- Raptis S,
- et al
- O’Donnell MJ,
- Fang J,
- D’Uva C,
- Saposnik G,
- Gould L,
- McGrath E,
- et al
- Strbian D,
- Karjalainen-Lindsberg ML,
- Tatlisumak T,
- Lindsberg PJ.
- Lindsberg PJ,
- Strbian D,
- Karjalainen-Lindsberg ML.
- Mattila OS,
- Strbian D,
- Saksi J,
- Pikkarainen TO,
- Rantanen V,
- Tatlisumak T,
- et al