Figures
Abstract
Background
Heart failure (HF) prevalence is increasing in the United States. Mechanical Circulatory Support (MCS) therapy is an option for Advanced HF (AdHF) patients. Perioperatively, multiorgan dysfunction (MOD) is linked to the effects of device implantation, augmented by preexisting HF. Early recognition of MOD allows for better diagnosis, treatment, and risk prediction. Gene expression profiling (GEP) was used to evaluate clinical phenotypes of peripheral blood mononuclear cells (PBMC) transcriptomes obtained from patients' blood samples. Whole blood (WB) samples are clinically more feasible, but their performance in comparison to PBMC samples has not been determined.
Methods
We collected blood samples from 31 HF patients (57±15 years old) undergoing cardiothoracic surgery and 7 healthy age-matched controls, between 2010 and 2011, at a single institution. WB and PBMC samples were collected at a single timepoint postoperatively (median day 8 postoperatively) (25–75% IQR 7–14 days) and subjected to Illumina single color Human BeadChip HT12 v4 whole genome expression array analysis. The Sequential Organ Failure Assessment (SOFA) score was used to characterize the severity of MOD into low (≤ 4 points), intermediate (5–11), and high (≥ 12) risk categories correlating with GEP.
Results
Results indicate that the direction of change in GEP of individuals with MOD as compared to controls is similar when determined from PBMC versus WB. The main enriched terms by Gene Ontology (GO) analysis included those involved in the inflammatory response, apoptosis, and other stress response related pathways. The data revealed 35 significant GO categories and 26 pathways overlapping between PBMC and WB. Additionally, class prediction using machine learning tools demonstrated that the subset of significant genes shared by PBMC and WB are sufficient to train as a predictor separating the SOFA groups.
Citation: Bondar G, Cadeiras M, Wisniewski N, Maque J, Chittoor J, Chang E, et al. (2014) Comparison of Whole Blood and Peripheral Blood Mononuclear Cell Gene Expression for Evaluation of the Perioperative Inflammatory Response in Patients with Advanced Heart Failure. PLoS ONE 9(12): e115097. https://doi.org/10.1371/journal.pone.0115097
Editor: Mark A. Sussman, San Diego State University, United States of America
Received: August 13, 2014; Accepted: November 14, 2014; Published: December 17, 2014
Copyright: © 2014 Bondar et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: The authors confirm that all data underlying the findings are fully available without restriction. All relevant data have been deposited to Figshare (http://dx.doi.org/10.6084/m9.figshare.1246742).
Funding: This research was supported through R21 HL 120040-01A1 NHLBI, the kind Advanced Heart Failure Research gifts from Philip Geier, John Tocco, Robert Milo, and Juan Mulder. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors declare that they have no competing interests.
Introduction
Heart failure (HF) affects more than 5 million people in the United States. Mechanical Circulatory Support (MCS) therapy is a current temporary strategy for patients with end-stage HF who are not candidates for heart transplantation. Because of the limited availability of donor organs and the approval of MCS for long term (destination) therapy, the use of MCS therapy has grown rapidly over the last 10 years. MCS therapy is considered in patients who are no longer responsive to medical treatment, patients that are not candidates for heart transplantation (as destination or lifetime therapy), patients who are awaiting a heart transplant and/or are becoming too sick due to progressive heart failure, and also in patients with HF in whom myocardial function is expected to return to normal in a short period of time (as a bridge to recovery) [1]. Outcomes after MCS therapy have significantly improved, yet 10–20% of patients die during the first year post MCS-implantation, usually from sepsis and Multi Organ Dysfunction Syndrome (MOD) [2]. MOD is linked to an altered immune response induced by the device and the surgical procedure, and is influenced by the preexisting HF [3] [4] [5]. Multiple abnormal immune functions describe the critically ill patient: aberrant systemic inflammatory response syndrome (SIRS), altered production of antibodies, abnormal lymphocyte response regulation, release of chemical mediators including cytokines, nitric oxide, endothelin, and prostaglandins [6] [7] [8].
Risk prediction tools are commonly used in clinical practice to estimate outcomes. Among them is the Sequential Organ Failure Assessment (SOFA) score, which is useful to predict MOD following MCS therapy. Mortality and length of stay in the ICU and the hospital can be estimated by cross-sectional values of initial score or highest score as well as temporal changes in the SOFA score. Survival is limited for MCS patients with MOD who have high SOFA scores [9]. Early recognition of MOD has important implications in diagnosis, treatment and risk stratification of patients with AdHF undergoing high risk cardiovascular interventions such as MCS, high risk cardiac revascularization, or valve replacement.
However, the SOFA score as well as all other currently available risk scoring tools for MOD ICU patients do not incorporate parameters of immune function, i.e. indicators of the inflammatory response, even though MOD is linked to an exaggerated leukocyte-mediated SIRS. Therefore, we propose a comprehensive evaluation of the immune response associated with MCS implantation and states of multiorgan injury. After completion of the human genome project, global (whole transcriptome) methods of gene expression profiling (GEP) of various tissues and blood cell types have become available for genome-wide evaluation of clinical phenotypes that can now be used to improve clinical evaluation in multiple disease settings. [10] [11] [12] [13] [14] [15]. We have previously developed a PBMC GEP test to rule out heart transplant rejection [27], implemented the test clinically, showed its clinical utility [43] and showed its potential to predict clinical events after heart transplantation [42], and demonstrated the feasibility of PBMC whole transcriptome GEP in AdHF-patients undergoing MCS-implantation [44]. In this setting, evaluation of the information contained in the PBMC transcriptome may provide a promising solution to this important missing point in the assessment of the critically ill patient.
In the ICU-setting, there are specific challenges to overcome in the development of novel genomic diagnostics. First, several methodological challenges apply to PBMC GEP, which may interfere with reproducibility through the addition of systematic bias when used in the setting of multicenter studies [6] [23] [24] [25] [26] [27] [28] [29]. Second, when approaching the evaluation of critically ill patients, the amount of blood that is obtained may add a complication to the process of recovery. PBMC processing for leukocyte isolation requires 8 ml of blood while WB methods require only 2.5 ml. These patients require multiple blood draws and efforts to minimize loss that is considered in the best interest of the patients. Third, these patients are exposed to blood transfusion, medications, and different degrees of oxygen saturation and fluid loading which could potentially cause error in the results. (Fig. 1).
It is known that there is a significant overlap between WB and PBMC gene expression [16]. A study by Whitney et al (2003) found that the proportions of the variety of cell types in peripheral blood can affect the GEP of a patient and that these proportions can rapidly change after an inflammatory response [18]. Debey et al (2004) compared the effect of different isolation techniques in gene expression experiments and found that PBMC analysis yields a more sensitive diagnosis as compared to WB [19]. However, Freezor et al (2004) and Palmer et al (2006) discovered that the process was laborious and could affect the GEP [20] [21]. Most recently, Min et al (2012) evaluated the variability of GEP in human blood and lymphoblastoid cell lines. They showed a significant overlap in the gene expression profiles between WB and PBMC [22].
An important aspect concerning the applicability of new techniques that can be rapidly translated into clinical practice involves its feasibility, reproducibility, cost effectiveness, and ease of implementation, and application. For practical and methodological reasons, such studies are usually performed using purified PBMC samples. However, use of PBMC is labor-intensive and requires several methodological steps that must be strictly followed [16] [21] [23]. As an alternative, easy-to-use WB gene expression methods have been developed for clinical applications [30]. While comparisons have already been done for gene expression of PBMC and WB supporting the use of both methodologies, it is not known whether the same principles can be applied to critically ill patients undergoing complex surgical procedures [33]. These patients are usually exposed to multiple factors such as repeat blood transfusions, hemodilution, stressors and drugs which could interfere with the applicability of WB gene expression methodologies to this specific population.
Therefore, we hypothesize that mRNA profile of WB from patients with MCS can be used as a surrogate for mRNA profile of PBMC of patients with MCS to identify transcriptome fingerprints for biomarker discovery in various degrees of MOD [30], [31], [37].
Methods
Ethics statement
This study was approved by the Columbia University Office of Human Research Protection Program IRB # 2206 and all patients signed an informed consent.
Patients
We collected blood samples from 31 consecutive HF patients (57±15 years old) undergoing cardiothoracic surgery (n = 29 MCS, n = 1 coronary artery bypass surgery (CABG) + aortic valve replacement (AVR), n = 1 transcatheter aortic valve replacement (TAVR) and 7 healthy age-matched controls). Samples were obtained between March 2010 and May 2011 at a single institution. PBMC was available for 30 patients and 7 controls, and WB was available for 31 patients and 3 controls. Main characteristics of the study samples are summarized in Table 1.
Samples were collected at a single time-point postoperatively at a median of 8 days after device implantation (25–75% IQR 7 – 14 days). The variability of the day 8 sampling time point was dictated by the clinical circumstances. To assess for MOD, we utilized the SOFA score. The SOFA score is a simple, validated, and widely accepted tool that can be easily obtained at a patient's bedside and is used to assess patient's disease severity and predict survival in the critical care unit [40]. This score has been applied to different study populations including patients with MCS [41]. SOFA is an integer score that assigns a numerical variable to each of 6 major organ systems to quantify the severity of organ failure. Values range between 0 and 4. Each system's value is summed into a single SOFA score. Therefore the sum score ranges between 0 to 24 and correlates with the severity of MOD and clinical outcomes. Following our previous studies [42], [43], [44], [45], we divided the population into SOFA score risk subgroups which were defined as low (≤ 4 points), intermediate (5–11), and high (≥ 12). SOFA score was computed for each patient at the time that gene expression samples were obtained.
Sample collection and RNA isolation
Blood was drawn and collected into PaxGene tubes (Becton Dickinson, Franklin Lakes, NJ) for the WB mRNA gene expression analysis or into a CPT tube (Becton Dickinson, Franklin Lakes, NJ) for PBMC cell isolation. For the WB samples, 2.5ml of blood were collected using PaxGene tubes. Samples were incubated at room temperature for 2 hours for RNA stabilization and then stored at −80°C. RNA was extracted from WB using the PAXgene Blood RNA Kit (Qiagen, Valencia, CA) according to the manufacturer's protocol which includes a globin reduction step. Briefly, samples were removed from −80°C and incubated at room temperature for 2 hours. Following lysis, the tubes were centrifuged for 10 min at 5,000×g, the supernatant was discarded and 500µL of RNase-free water was added to the pellet. Subsequently, the tubes were vortexed thoroughly to re-suspend the pellet, centrifuged for 10 min at 5000×g and the entire supernatant was discarded. The pellet was re-suspended in 360µL of buffer BR1 by vortexing and further purification of RNA was done following the manufacturer's protocol with on-column DNase digestion.
For the PBMC samples, mononuclear cells were isolated from 8 ml of blood collected by Vacutainer cell preparation tubes (CPT) with sodium citrate (Becton Dickinson, Franklin Lakes, NJ), resuspended in RNeasy Lysis Buffer (RLT, Qiagen, Valencia, CA) within 2 hours of phlebotomy. Total RNA was isolated from each sample (RNeasy, Qiagen, Valencia, CA). Quality of the purified RNA was verified on an Agilent 2100 Bioanalyzer (Agilent Technologies, Palo Alto, CA); RNA concentrations were determined using a NanoDrop ND-1000 spectrophotometer (NanoDrop Technologies, Wilmington, DE).
Microarray data processing and analysis
After RNA extraction, quantification and quality assessment, total mRNA was amplified and hybridized on the Illumina single color Human BeadChip HT12 v4 whole genome expression array. Each array targets more than 47,000 probes derived from the National Center for Biotechnology Information Reference Sequence (NCBI) RefSeq Release 38 (November 7, 2009) and other sources. All samples were processed in a single core facility. After hybridization and microarray chip processing, feature extraction was carried out in the Illumina Beadarray platform (Illumina, San Diego, CA). Data was then subjected to quantile normalization using GenomeStudio (Illumina, San Diego, CA) and normalized expression files for PBMC and WB were used hereafter for comparison.
Filtering was done against background and only those probes having more than 20% variation in the gene expression were retained. Probes mapping to the same gene transcript were not averaged. Data was then transferred to GeneSpring GX 12 (Agilent, Palo Alto, CA). Sample data was normalized using the Quantile normalization method and quality control carried out as implemented in GeneSpring [46]. Given that the starting dataset showed a significant amount of noise in the signal for the WB samples, data was further assessed for the coefficient of variation for each entity within risk groups so only those with a coefficient lower than 0.75 in at least one group were kept for subsequent analysis. This criterion was employed to reduce the intra-group variability, thereby selecting the most reliable entities for potential biomarker candidates.
Quantitative Real-time Polymerase Chain Reaction (RT-qPCR) and Validation
Microarray data were validated by Quantitative PCR on PBMC and WB obtained from an independent set of samples. Total RNA from PBMC cells were purified using RNeasy Mini Kit (Qiagen, Valencia, CA) and total RNA from WB cells were purified using the MagNA pure Compact System (Roche, Pleasanton, CA). CDNA was synthesized with iScript supermix for RT-qPCR (BioRad, Hercules, CA). RT-qPCR analysis was carried out with iTaq SYBR green supermix (BioRad, Hercules, CA) on the 7500 Fast Real-time PCR system (Applied Biosystems, Foster City, CA). 18S rRNA levels were used as an internal control for real-time PCR. Sequences of the primer pairs used were as follows: 18S rRNA, 5′-GTAACCCGTTGAACCCATT-3′ and 5′-CCATCCAATCGGTAGTAGCG-3′; IL11RA, 5′-ACTTCCTGCTCAAGTTCCGT-3′ and 5′-GGCACTGACTCGTACAGCAT-3′; IL2RB, 5′-TTCTAGCGTCAGTGCTGGAG-3′ and 5′-CCTCAGAGATCCCAAAGGAA-3′; TRAF3IP3, 5′- GAGGCTCTGAAGGAGGACTG-3′ and 5′- TATCTGCTCCCTGCAGTTTG-3′; LAT2, 5′- CTACCCACCTGTCACCTCCT-3′ and 5′- CTGTTGGCACCATCAGAATC-3′; CCL5, 5′- AAGGAAGTCAGCATGCCTCT-3′ and 5′- TTTGCCAGTAAGCTCCTGTG-3′; GNLY, 5′-TTCTAGCGTCAGTGCTGGAG-3′ and 5′- ATGCCTTTCACACCCTGTTT-3′; MAP4K1, 5′- AAGATCCAGGACACCAAAGG-3′ and 5′- CTGGTACCACTGAAGCAGGA-3′; ABLIM1, 5′- GTGCAGTTCCCATGAGTCAC-3′ and 5′- GGACAATGGTTTCCTCTGCT-3′; CD96, 5′- AACACCATGGCTGTCACACT-3′ and 5′- AGGCTCGATGGTTCTCAACT-3′; GRB10, 5′- GCAGCCAGTCAGTCTTTCAA-3′ and 5′- GCAGCCAGTCAGTCTTTCAA-3′.
Statistical analysis of gene expression
Normalized gene expression samples of normal controls were used as the reference against which samples from patients with low, medium, and high SOFA scores were compared. Statistical comparison was done by the Kruskal-Wallis analysis of variance method and corrected for multiple hypotheses testing using the Benjamini-Hochberg method. Genes that satisfied the significance criteria of less than 5% false discovery rate (FDR) were further analyzed for the enrichment of gene ontology (GO) and subjected to pathway analysis. For the GO analysis, we used High-Throughput GoMiner (HTGM) [47] (http://discover.nci.nih.gov/) to analyze the enrichment of GO categories by up- and down-regulated genes for both WB and PBMC [48]. HTGM analyzes data from all microarrays in a study, provides diagnostics for data interpretation, and allows visualization in the form of clustered heat maps [49]. Normally, the input to HTGM consists of a total-genes file (representing the entire Microarray or a randomly generated whole genome seed) and a changed-genes file (representing the genes with altered expression) relevant to the study purpose. The output generated by HTGM includes a summary of the results, a matrix whose rows are categories and whose columns are names of changed gene for hierarchical clustering of experiments and categories, and a statistical summary for each category including one-sided Fisher exact p-value and an FDR. Hierarchical clustering of enriched categories and changed genes allows determining which categories achieved statistical significance by virtue of containing essentially the same set of changed genes. For pathway analysis, we used the algorithm incorporated in Genespring, testing for differentially expressed genomic pathways based on the available repositories including Biopax, Wiki, and Reactome. A pathway list was obtained based on Entrez Gene ID and UniGene ID.
To assess the performance of PBMC and WB for biomarker development, we examined the accuracy of predicting the high-risk SOFA group subjects across data sets using the Support Vector Machine (SVM) Class Prediction module in GeneSpring GX 12 (Agilent, Palo Alto, CA), we trained linear kernel SVMs on the PBMC and WB gene datasets defined by the significance criteria described above, as well as the gene set defined by the intersection of these two sets and evaluated the misclassification rates between SOFA groups when each of these trained SVMs were applied to either dataset in its entirety.
Results
PBMC versus WB Differential Gene Expression (HF versus Controls)
A general workflow of the analysis depicting differential gene expression for both groups is presented in Fig. 2.
The work presented here is the comparison between PBMC and WB whole transcriptome GEP and their correlation to the SOFA score. The combined risk groups of all 31 HF patients exhibited 2605 gene transcripts (2305 genes differentially expressed) in the PBMC and 446 (400 genes differentially expressed) in the WB dataset that had at least a 1.5 fold change in their expression level compared to healthy age-matched controls. While substantially more gene transcripts were differentially expressed in the PBMC compared to WB, there was a significant overlap among them with 333 genes (83.2%) shared by WB and PBMC. The direction of gene expression between WB and PBMC was consistent for 229 genes (68.7%) in the low risk compared to the control group, 245 genes (73.5%) in the medium risk group, and 251 genes (75.4%) in the high risk group. A list of the highest differentially expressed genes overlapping in WB and PBMC is provided in Table 2. A list of the overlapping genes and their Pearson correlations with the SOFA score, in both WB and PBMC, is included in the S2 Table. A complete list of genes both overlapping and non-overlapping for each dataset is also provided in the S1 Table.
PBMC versus WB Biomarker Candidate Gene Discovery
Expression patterns of the highest differentially expressed genes have been clustered as a heatmap for visualization purposes and provided in Fig. 3.
This shows that patterns of the highest differentially expressed genes have similar profiles for PBMC and WB. For these genes, we used text mining algorithms as implemented in GeneSpring to identify those with the highest connectivity and most likely to have an important role in the inflammation process. These genes included IL11RA, CCLS, GNLY, MAP4K1, LAT2, CD96, GRB10, and IL2RB.
Validation of PBMC and WB differentially regulated genes by RT-qPCR
To validate the microarray results in this study we used an alternative approach. We performed RT-qPCR to assay the 10 highest ranked genes, ranked first by statistical significance, secondly by correlation between WB and PBMC microarray expression levels, and finally by biological relevance (Table 3).
To perform this validation, we analyzed RT-qPCR of 8 samples taken from across control, low/medium, and high SOFA score groups. Fold changes were calculated against control, averaged in each group by their logarithms, and transformed back. We compared the microarray and RT-qPCR results of expression of 10 genes between the SOFA score sub-groupings within the PBMC and WB groupings (Fig. 4).
Darker colors correspond to higher magnitude of fold change. Fold changes showed similar patterns between all 10 genes, 9 of which are down-regulated, and 1 of which is up-regulated. 7 of 10 genes showed similar patterns between PBMC and WB and also within SOFA score groups according to our microarray and PCR results. 2 of 10 genes (IL11RA and GRB10) show an opposite result in HIGH PCR (PBMC) and low/med array (WB) correspondingly. Only 1 gene, MAP4K1, shows an opposite result between SOFA score groups.
PBMC versus WB Gene Ontology analysis
GO analysis identified 314 terms enriched by 1327 genes. 35 of these terms overlapped between WB and PBMC. GO categories included regulation of the immune response and related pathways including the innate and the adaptive immune responses. Interestingly, there was a pronounced enrichment of innate response GO terms by up-regulated genes and of the adaptive immune response by down-regulated genes as previously described in the setting of critical illness [16]. Other enriched GO terms were related to programmed cell death and the coagulation cascade, both well known to be involved in the biology of critical illness [17]. The identification of these processes in the context of the known biology supports the validity of our findings as opposed to random enrichment of GO categories. The overall enrichment found in the PBMC dataset compared to WB showed that the majority of genes, while different, were likely involved in similar biological processes. Thus, GO analysis suggests that for both PBMC and WB, expressed genes are involved in inflammation, apoptosis, and stress response related pathways among others. There is a significant difference in the number of GO categories but not in the processes represented by these categories. The lack of a significant enrichment of GO categories or members of a specific gene pathway in the WB dataset could be explained by different power assumptions when starting from a smaller number of genes. However, differences in the direction of gene expression could be explained by multiple mechanisms that reside in the interface between WB composition and PBMC and need to be further investigated. Genes in the overlap dataset include members of pathways such as Apoptosis (BCL2, MAGED1, and TNFRSF10B), Nephrin interactions (FYN), CD28 and CD28 co-stimulation (CD28r), Cell-extracellular matrix interactions (BCL2, LAT2, KLFR1, KLRD1, TRAT1, PTCRA, KLRB1, and ICAM2), Antiviral mechanism mediated by Interferon gamma signaling (IL1R2 and IRAK3), and Lipoprotein metabolism including LDL-mediated lipid transport, and Chylomicron-mediated lipid transport (LPL, VLDL, and TGIHDL). The most relevant GO categories enriched in PBMC and WB are described in Table 4.
Of particular interest in the heatmap (Fig. 3) are the areas where multiple GO categories overlap or “cross-talk”, suggesting that mechanisms involved for a specific set of genes are common to different biological processes. Those GO terms were related to molecular mechanisms associated with inflammation leading to T-cell activation, activation of cytokine signaling cascades, and regulation of programmed cell death. Other important biological processes identified in the non-overlapping genes included “cross-talking” GO categories involving regulation of peptidase, caspase, and endopeptidase activity. A complete list of overlapping and non-overlapping GO categories is provided in S3 Table.
PBMC versus WB Pathway analysis
The total number of differentially expressed pathways associated with up- or down-regulated genes was 26 in the WB dataset and 265 in the PBMC dataset. A total of 137 genes present in the WB pathway enrichment analysis were also included in the list of genes from PBMC pathways (466). The enrichment analysis revealed that 26 pathways were associated with up-regulated genes based on a corrected 2-tailed p-value <0.05 (5% FDR). Pathways differentially regulated in WB that were common to PBMC included the adaptive immune system. Out of 40 genes listed under the adaptive immune system pathway, 10 genes (25%) in the WB dataset were differentially expressed and 15 genes (37.5%) in the PBMC dataset were differentially expressed (p-values<0.001). Genes in this pathway including ITK, LAT, LCK, and TRAT1 were found to be up-regulated in both PBMC and WB. Eleven genes were up-regulated in PBMC but down-regulated in WB. The T-Cell receptor signaling pathway showed an enrichment of 30% (7/23 genes, (p = 3.8×10−09). Other important pathways included apoptosis, homeostasis, and those associated with HIV infections, CD28 and GPCR signaling. Significantly enriched pathways and corresponding gene expression are listed in Table 5.
For the pathway analysis, we found that of the 34 significant genes that overlapped between PBMC and WB pathway groups, 26 had concordantly regulated patterns in both groups. The other 8 genes showed an opposite expression in PBMC and WB datasets. Results observed in the WB samples were similar in several aspects to those in the PBMC datasets, supporting the role of these genes as potential biomarkers for early diagnosis and monitoring of MOD. Variations in the direction of gene expression suggest that perturbations in multiple signaling and cellular mechanisms occur in a comparable way in PBMC as in WB.
Performance of classification using PBMC and WB
We employed a support vector machine (SVM) to examine the accuracy of predicting the high-risk SOFA group subjects across data sets. Misclassification rates between the SOFA groups, when each of these trained SVMs was applied to either dataset, are shown as confusion matrices in Table 6.
We found that the SVM trained on the WB genes were accurate in predicting the medium and high risk SOFA subjects in the PBMC dataset (a), and the overlapping PBMC genes were accurate in predicting the medium and high risk SOFA subjects in the WB dataset (d). Also, training on only the overlapping PBMC genes was sufficient to accurately predict all patients in the PBMC group (c). However, training on all PBMC genes was not useful in predicting in the WB group because the SVM relies heavily on information about genes that were not significant in the WB dataset (b).
Discussion
In this paper, we show that WB can be used as a surrogate of PBMC expression in a set of critically ill patients who underwent AdHF cardiac surgery. We validate this result with independent set of samples using RT-qPCR. Our results indicate that the direction of change in gene expression profiles of individuals with MOD as compared to controls is similar when determined from PBMC versus WB. In each of the low, medium and high risk SOFA score patients; there was consistency in the direction of gene expression changes. The main enriched terms by GO analysis included those involved in the inflammatory response, apoptosis, and other stress response related pathways. The data revealed that there were 35 significant GO categories and 26 pathways overlapping between PBMC and WB. Additionally, class prediction using machine learning tools demonstrate that the subset of significant genes shared by both PBMC and WB are sufficient to train a predictor separating the SOFA groups.
Overall, multiple explanations can be postulated to support our findings. First, the differences between PBMC and WB could be secondary to lower signal-to-noise ratios in WB, leading to a higher number of false positive rates when tested with multiple comparisons. Thus, a way to reduce the ratio of true positives to true negatives required an additional step in filtering those transcripts with high intraphenotypic variability that would otherwise be removed by multiple hypothesis testing. As a positive consequence, only those transcripts that are more stable across phenotypes are retained for the analysis, providing a more reproducible signal with lower enrichment but higher correlated findings in the PBMC dataset.
Second, while most relevant changes in gene expression are captured by both methods, variability seems larger in the WB leading to many transcripts not being significant when tested with multiple hypotheses.
Third, the globin mRNA depletion method used in the PAXgene stabilized RNA leads to variable changes in gene expression based on different concentrations of WB and content of globin transcripts. Globin reduction protocols [32] [34] [35] [36] [37] have been used to improve microarray data quality by reducing data variability with increased detection rate of expressed genes and improved overlap with the expression signatures of isolated PBMC preparations.
Fourth, the differential composition of cell populations in PBMC and WB might account for differences in GEP patterns. While PBMC seems to directly identify changes in a well-defined immunological set of cells, such an assumption does not apply to WB. In WB, more diverse cell populations including neutrophils, eosinophils, platelets and reticulocytes among other constituents in the preparations, reduce the specific cell population of the study. While they have the potential to be used as biomarkers with the goal of classifying phenotypes, the biological meaning of the results is highly questionable when proposing mechanistic explanations. A comparative analysis revealed that nearly 2,000 genes with at least a 2-fold average difference in expression between WB and PBMCs could reflect the difference in cellular composition [18] [22]. Gene expression associated with WB displays a more pronounced pattern of up-regulation (Table 5). Theoretically, in both PBMC and WB sampling, analysis is possible using microfluidics-based flow cytometry-guided cell sorting followed by subpopulation-specific GEP [38] [39]. However, the intention of our approach, in creating a clinically feasible protocol, is to use the integrated mixed PBMC population in the original PBMC protocol and compare it with the WB-approach.
Fifth, the observed differences could also be explained by different RNA processing methods that have an effect on gene expression profile in WB and PBMC samples.
Sixth, the results in RT-qPCR showed non-significant differences between the 10 genes when compared to the microarray results.
Lastly, we have obtained high classification rates using both PBMC and WB. Although PBMC provided more accurate results, classification rates for the moderate and high risk groups were well discriminated by WB which provides significant advantages from a methodological and clinical implementation perspectives.
Limitations
Our findings should be interpreted in consideration of several study limitations including: 1) limited sample size, 2) lack of power to assess gender specific variation, 3) variation in the time points of collection dictated by clinical setting, 4) use of the SOFA score which has not been specifically developed for the MCS population, and 5) use of cross-sectional single time-point evaluations.
Conclusions
We conducted a study to evaluate the comparative usefulness of WB and PBMC transcriptome analysis in critically ill patients undergoing a complex and high risk cardiovascular intervention such as MCS therapy. We found that although there was a significant difference in sensitivity, molecular fingerprints of WB and PBMC had a good overlap and concordance in their gene expression with common pathways and mechanisms represented by these genes.
WB as well as PBMC is useful to delineate the inflammatory response associated with MOD after MCS. While PBMC classification outperformed WB, technical refinement and development of larger, prospective studies are warranted to further develop WB biomarkers in critical care settings.
Advantages of PBMC testing include better biological inquiry at the cost of methodological challenges with difficult implementation in the real world clinical setting. While WB potentially carries a reduced value in classification performance and reflection of actual biology, it has the advantage of easier implementation, requires a smaller volume, and is more resistant to operational variation. Even though the critical care setting is associated with many complex interventions, these do not appear to interfere with the ability of using WB biomarkers to identify high risk patients.
Supporting Information
S1 Table.
Overlapping up- and down-regulated genes expressed in PBMC and WB between patients after intervention.
https://doi.org/10.1371/journal.pone.0115097.s001
(DOCX)
S2 Table.
Correlations between expression in PBMC and WB, and SOFA score.
https://doi.org/10.1371/journal.pone.0115097.s002
(DOCX)
S3 Table.
GO categories overlapping between PBMC and WB after intervention.
https://doi.org/10.1371/journal.pone.0115097.s003
(DOCX)
Author Contributions
Conceived and designed the experiments: GB MC NW KS PP ER MD. Performed the experiments: JM JC EC MB CS. Analyzed the data: GB MC NW EC MD. Contributed reagents/materials/analysis tools: GB MC NW. Wrote the paper: GB MC NW MD. Drafted the manuscript: JM JC EC MB CS. Carried out the molecular genetic studies: KS. Helped develop the overall study hypothesis: PP ER MD.
References
- 1. Kirklin JK, Naftel DC, Pagani FD, Kormos RL, Stevenson L, et al. (2012) Long-term mechanical circulatory support (destination therapy): on track to compete with heart transplantation? J Thorac Cardiovasc Surg. Sep 144(3):584–603 discussion 597–8 doi:https://doi.org/10.1016/j.jtcvs.2012.05.044. Epub 2012 Jul 15. 22795459; PubMed Central PMCID: PMC3443856
- 2. Kirklin JK, Naftel DC, Pagani FD, Kormos RL, Stevenson LW, et al. (2014) Sixth INTERMACS annual report: A 10,000-patient database. J Heart Lung Transplant. 2014 Jun 33(6):555–64 doi:https://doi.org/10.1016/j.healun.2014.04.010.
- 3.
Barnes K (2008) Complications in patients with ventricular assist devices. Dimens Crit Care Nurs. 2008 Nov-Dec;27(6):233–41; quiz 242-3. doi:10.1097/01.DCC.0000338867.24293.20. Review. PubMed PMID: 18953188.
- 4.
Yndestad A, Damås JK, Oie E, Ueland T, Gullestad L, et al. (2006) Systemic inflammation in heart failure-the whys and wherefores. Heart Fail Rev. 2006. Mar;11(1):83–92. Review. PubMed PMID: 16819581.
- 5. Shahzad K, Aziz QA, Leva JP, Cadeiras M, Ho EK, et al. (2010) New-onset graft dysfunction after heart transplantation—incidence and mechanism-related outcomes. J Heart Lung Transplant. 2011 Feb 30(2):194–203 doi:https://doi.org/10.1016/j.healun.2010.08.026. Epub 2010 Oct 16. 20952209
- 6. Deng MC, Edwards LB, Hertz MI, Rowe AW, Keck BM, et al. (2005) International Society for Heart and Lung Transplantation. Mechanical circulatory support device database of the International Society for Heart and Lung Transplantation: third annual report-2005. J Heart Lung Transplant. 2005 Sep 24(9):1182–7 PubMed PMID 16143231
- 7.
Redmond A, McDevitt M, Barnes S (2004) Acute renal failure: recognition and treatment in ward patients. Nurs Stand. 2004 Feb 11–17;18(22):46–53; quiz 54–5.Review. PubMed PMID: 14999985.
- 8. Redmond A, McDevitt M, Barnes S. (2004) Acute renal failure: recognition and treatment in ward patients. Nurs Stand. 2004 18(22):46–53.
- 9. Vincent JL, Moreno R, Takala J, Willatts S, De Mendonça A, et al. (1996) The SOFA (Sepsis-related Organ Failure Assessment) score to describe organ dysfunction/failure. On behalf of the Working Group on Sepsis-Related Problems of the European Society of Intensive Care Medicine. Intensive Care Med. 1996 Jul; 22(7):707–10 PubMed PMID 8844239
- 10. Xie L, Weichel B, Ohm JE, Zhang K (2011) An integrative analysis of DNA methylation and RNA-Seq data for human heart, kidney and liver. BMC Syst Biol. 2011 Dec 3 5 Suppl 3S4 doi:;10.1186/1752-0509-5-S3-S4. Epub 2011 Dec 23. PubMed MID: 22784623; PubMed Central PMCID: PMC3287572.
- 11. Munch EM, Harris RA, Mohammad M, Benham AL, Pejerrey SM, et al. (2013) Transcriptome Profiling of microRNA by Next-Gen Deep Sequencing Reveals Known and Novel miRNA Species in the Lipid Fraction of Human Breast Milk. PLoS One. 2013;8(2):e50564 doi:https://doi.org/10.1371/journal.pone.0050564. Epub 2013 Feb 13. PubMed PMID 23418415; PubMed Central PMCID: PMC3572105
- 12.
Cine N, Limtrakul P, Sunnetci D, Nagy B, Savli H (2013) Effects of curcumin on global gene expression profiles in the highly invasive human breast carcinoma cell line MDA-MB 231: A gene network-based microarray analysis. Exp Ther Med. 2013 Jan;5(1):23–27. Epub 2012 Oct 19. PubMed PMID: 23251236; PubMed Central PMCID: PMC3524226.
- 13.
Ma XJ, Salunga R, Tuggle JT, Gaudet J, Enright E, et al. (2003) Gene expression profiles of human breast cancer progression. Proc Natl Acad Sci U S A. 2003 May 13;100(10):5974–9. Epub 2003 Apr 24. PubMed PMID: 12714683; PubMed Central PMCID: PMC156311.
- 14.
Mehra MR, Uber PA, Walther D, Vesely M, Wohlgemuth JG, et al. (2006) Gene expression profiles and B-type natriuretic peptide elevation in heart transplantation: more than a hemodynamic marker. Circulation. 2006 Jul 4;114(1 Suppl):I21–6. PubMed PMID: 16820574.
- 15. Erickson LM, Pan F, Ebbs A, Kobayashi M, Jiang H (2003) Microarray-based gene expression profiles of allograft rejection and immunosuppression in the rat heart transplantation model. Transplantation. 2003 Aug 15 76(3):582–8 12923448
- 16. Joehanes R, Johnson AD, Barb JJ, Raghavachari N, Liu P, et al. (2012) Gene expression analysis of whole blood, peripheral blood mononuclear cells, and lymphoblastoid cell lines from the Framingham Heart Study. Physiol Genomics. 2012 Jan 18 44(1):59–75 doi:https://doi.org/10.1152/physiolgenomics.00130.2011. Epub 2011 Nov 1. 22045913; PubMed Central PMCID: PMC3289123
- 17.
Draghici S, Khatri P, Tarca AL, Amin K, Done A, et al. (2007) A systems biology approach for pathway level analysis. Genome Res. 2007 Oct;17(10):1537–45. Epub 2007 Sep 4. PubMed PMID: 17785539; PubMed Central PMCID:PMC1987343.
- 18.
Whitney AR, Diehn M, Popper SJ, Alizadeh AA, Boldrick JC, et al. (2003) Individuality and variation in gene expression patterns in human blood. Proc Natl Acad Sci U S A. 2003 Feb 18;100(4):1896–901. Epub 2003 Feb 10. PubMed PMID: 12578971; PubMed Central PMCID: PMC149930.
- 19. Debey S, Schoenbeck U, Hellmich M, Gathof BS, Pillai R, et al. (2004) Comparison of different isolation techniques prior gene expression profiling of blood derived cells: impact on physiological responses, on overall expression and the role of different cell types. Pharmacogenomics J. 2004 4(3):193–207 15037859
- 20. Feezor RJ, Baker HV, Mindrinos M, Hayden D, Tannahill CL, et al. (2004) Inflammation and Host Response to Injury, Large-Scale Collaborative Research Program. Whole blood and leukocyte RNA isolation for gene expression analyses. Physiol Genomics. 2004 Nov 17 19(3):247–54 PubMed PMID: 15548831
- 21. Palmer C, Diehn M, Alizadeh AA, Brown PO (2006) Cell-type specific gene expression profiles of leukocytes in human peripheral blood. BMC Genomics. 2006 May 16 7:115 16704732; PubMed Central PMCID: PMC1479811
- 22. Min JL, Barrett A, Watts T, Pettersson FH, Lockstone HE, et al. (2010) Variability of gene expression profiles in human blood and lymphoblastoid cell lines. BMC Genomics. 2010 Feb 8 11:96 doi:https://doi.org/10.1186/1471-2164-11-96. 20141636; PubMed Central PMCID: PMC2841682
- 23. Yentis SM, Rowbottom AW, Riches PG (1995) Detection of cytoplasmic IL-1 beta in peripheral blood mononuclear cells from intensive care unit (ICU) patients. Clin Exp Immunol. 1995 May 100(2):330–5 PubMed PMID: 7743673; PubMed Central PMCID:PMC1534328
- 24. Cadeiras M, Nagaraj H, Kirklin JK, Holman WL, Pamboukian SV, et al. (2011) Use of a vascular plug to percutaneously occlude the outflow conduit of a dysfunctional left ventricular assist device. J Heart Lung Transplant. 2011 Jun 30(6):733–4 doi:https://doi.org/10.1016/j.healun.2011.01.723. Epub 2011 Mar 26. PubMed PMID: 21441042
- 25.
Cabré L, Mancebo J, Solsona JF, Saura P, Gich I, et al. (2005) Bioethics Working Group of the SEMICYUC. Multicenter study of the multiple organ dysfunction syndrome in intensive care units: the usefulness of Sequential Organ Failure Assessment scores in decision making. Intensive Care Med. 2005 Jul;31(7):927–33. Epub 2005 Apr 26. PubMed PMID: 15856171.
- 26. Dobbin KK, Beer DG, Meyerson M, Yeatman TJ, Gerald WL, et al. (2005) Interlaboratory comparability study of cancer gene expression analysis using oligonucleotide microarrays. Clin Cancer Res. 2005 Jan 15;11(2 Pt 1):565–72 15701842
- 27. Shahzad K, Cadeiras M, Memon S, Zeeberg B, Klingler T, et al. (2010) Gene Expression Signatures of Peripheral Blood Mononuclear Cells during the Early Post-Transplant Period in Patients Developing Cardiac Allograft Vasculopathy. J Transplant. 2010;2010:719696. doi:https://doi.org/10.1155/2010/719696. Epub 2010 Dec 28. PubMed PMID: 21234308; PubMed Central PMCID: PMC3018628.
- 28.
Jison ML, Munson PJ, Barb JJ, Suffredini AF, Talwar S, et al. (2004) Blood mononuclear cell gene expression profiles characterize the oxidant, hemolytic, and inflammatory stress of sickle cell disease. Blood. 2004 Jul 1;104(1):270–80. Epub 2004 Mar 18. PubMed PMID: 15031206.
- 29. Burczynski ME, Twine NC, Dukart G, Marshall B, Hidalgo M, et al. (2005) Transcriptional profiles in peripheral blood mononuclear cells prognostic of clinical outcomes in patients with advanced renal cell carcinoma. Clin Cancer Res. 2005 Feb 1 11(3):1181–9 PubMed PMID: 15709187
- 30. Rainen L, Oelmueller U, Jurgensen S, Wyrich R, Ballas C, et al. (2002) Stabilization of mRNA expression in whole blood samples. Clin Chem. 2002 Nov 48(11):1883–90 PubMed PMID 12406972
- 31. Raghavachari N, Xu X, Munson PJ, Gladwin MT (2009) Characterization of whole blood gene expression profiles as a sequel to globin mRNA reduction in patients with sickle cell disease. PLoS One. 2009 Aug 3; 4(8):e6484 doi:https://doi.org/10.1371/journal.pone.0006484. 19649296; PubMed Central PMCID: PMC2714456
- 32.
(2003) An Analysis of Blood Processing Methods to Prepare Samples for GeneChip Expression Profiling. Affymetrix Technical Note.
- 33. Debey-Pascher S, Hofmann A, Kreusch F, Schuler G, Schuler-Thurner B, et al. (2011) RNA-stabilized whole blood samples but not peripheral blood mononuclear cells can be stored for prolonged time periods prior to transcriptome analysis. J Mol Diagn. 2011 Jul; 13(4):452–60 doi:https://doi.org/10.1016/j.jmoldx.2011.03.006. 21704280; PubMed Central PMCID: PMC3123794
- 34.
(2003) Globin Reduction Protocol: A Method for Processing Whole Blood RNA Samples for Improved Array Results. Affymetrix Technical Note.
- 35.
Globin RNA Reduction in Blood Samples. A Quintiles Company.
- 36. Rainen L, Oelmueller U, Jurgensen S, Wyrich R, Ballas C, et al. (2002) Stabilization of mRNA expression in whole blood samples. Clin Chem. 2002 Nov; 48(11):1883–90 PubMed PMID: 12406972
- 37.
Li L, Ying L, Naesens M, Xiao W, Sigdel T, et al. (2008) Interference of globin genes with biomarker discovery for allograft rejection in peripheral blood samples. Physiol Genomics. 2008 Jan 17;32(2):190–7. Epub 2007 Oct 30. PubMed PMID: 17971501.
- 38. Kotz KT, Xiao W, Miller-Graziano C, Qian WJ, Russom A, et al. (2010) Inflammation and the Host Response to Injury Collaborative Research Program. Clinical microfluidics for neutrophil genomics and proteomics. Nat Med. 2010 Sep; 16(9):1042–7 doi:https://doi.org/10.1038/nm.2205. Epub 2010 Aug 29. PubMed PMID: 20802500; PubMed Central PMCID: PMC3136804
- 39. Rosenbach AE, Koria P, Goverman J, Kotz KT, Gupta A, et al. (2011) Microfluidics for T- lymphocyte cell separation and inflammation monitoring in burn patients. Clin Transl Sci. 2011 Feb; 4(1):63–8 doi:https://doi.org/10.1111/j.1752-8062.2010.00255.x. PubMed PMID: 21348958; PubMed Central PMCID:PMC3057059
- 40. Vincent JL, Moreno R, Takala J, Willatts S, De Mendonça A, et al. (1996) The SOFA (Sepsis-related Organ Failure Assessment) score to describe organ dysfunction/failure. On behalf of the Working Group on Sepsis-Related Problems of the European Society of Intensive Care Medicine. Intensive Care Med. 1996 Jul 22(7):707–10 PubMed PMID 8844239
- 41.
Shahzad K, Latif F, Akashi H, Kato TS, Sinha A, et al. (2011). Cardiac Support and Multiorgan Dysfunction Syndrome, Ventricular Assist Devices, Dr. Jeffrey Shuhaiber (Ed.), ISBN: 978-953-307-164-0, InTech, Available from: http://www.intechopen.com/books/ventricular-assist-devices/cardiac-support-and-multiorgan-dysfunctionsyndrome
- 42. Deng MC, Elashoff B, Pham MX, Teuteberg JJ, Kfoury AG, et al. (2014) IMAGE Study Group. Utility of gene expression profiling score variability to predict clinical events in heart transplant recipients. Transplantation. 2014 Mar 27 97(6):708–14 doi:[]10.1097/01.TP.0000443897.29951.cf. PMID: 24637869 [PubMed - indexed for MEDLINE].
- 43. Pham MX, Teuteberg JJ, Kfoury AG, Starling RC, Deng MC, et al. (2010) for the IMAGE Study Group. Gene Expression Profiling for Rejection Surveillance After Cardiac Transplantation. N Engl J Med. 2010 362:1890–1900.
- 44. Sinha A, Shahzad K, Latif F, Cadeiras M, Von Bayern MP, et al. (2010) Peripheral blood mononuclear cell transcriptome profiles suggest T-cell immunosuppression after uncomplicated mechanical circulatory support device surgery. Hum Immunol. 2010 Feb 71(2):164–9 doi:https://doi.org/10.1016/j.humimm.2009.10.012.
- 45. Starling RC, Pham M, Valantine H, Miller L, Eisen H, et al. (2007) Molecular testing in the management of cardiac transplant recipients: initial clinical experience. J Heart Lung Transplant. 2007;26:204.
- 46.
Genespring Manual12.5. Agilent Technologies.
- 47.
Zeeberg BR, Qin H, Narasimhan S, Sunshine M, Cao H, et al. (2005) High-throughput GoMiner, an ‘industrial-strength’ integrative gene ontology tool for interpretation of multiple-microarray experiments, with application to studies of Common Variable Immune Deficiency (CVID) BMC Bioinformatics. 2005;6, article 168
- 48.
Harris MA, Clark J, Ireland A, Lomax J, Ashburner M, et al. (2004) Gene Ontology Consortium. The Gene Ontology (GO) database and informatics resource. Nucleic Acids Res. 2004 Jan 1;32(Database issue):D258–61. PubMed PMID: 14681407; PubMed Central PMCID: PMC308770.
- 49. Weinstein JN, Myers TG, O'Connor PM, Friend SH, Fornace AJ Jr, et al. (1997) An information-intensive approach to the molecular pharmacology of cancer. Science. 1997 Jan 17; 275(5298):343–9 PubMed PMID: 8994024