|
|
|
Original Paper
|
|
|||
|
Acta Biochim Biophys
Sin 2008, 40: 426-436 |
||||
|
doi:10.1111/j.1745-7270.2008.00418.x |
How is mRNA expression predictive
for protein expression? A correlation study on human circulating monocytes
Yanfang Guo1,2#,
Peng Xiao1,3#, Shufeng Lei1, Feiyan
Deng1,
Gary Guishan Xiao3, Yaozhong Liu3,
Xiangding Chen1, Liming Li1, Shan Wu1,
Yuan Chen1, Hui Jiang1, Lijun
Tan1,
Jingyun Xie4, Xuezhen Zhu2,
Songping Liang4, and Hongwen Deng1,2,5*
1 Laboratory of Molecular and Statistical
Genetics, College of Life Sciences, Hunan Normal University, Changsha 410081,
China
2 Key Laboratory of Biomedical Information Engineering,
Ministry of Education and Institute of Molecular Genetics, School of Life
Science and Technology, Xi'an Jiaotong University, Xi'an 710049, China
3 Osteoporosis Research Center and Department of
Biomedical Sciences, Creighton University Medical Center, Omaha, Nebraska
68131, USA
4 Key Laboratory of Protein Chemistry and
Developmental Biology of Education Committee, College of Life Sciences, Hunan
Normal University, Changsha 410081, China
5 Departments of Orthopedic Surgery and Basic
Medical Sciences, University of Missouri-Kansas City, Kansas City, Missouri
64108, USA
Received: February
20, 2008�������
Accepted: March 27,
2008
The work was
partially supported by the grants from the National Natural Science Foundation of
China (Nos. 30600364, 30470534, and 30230210) and the Scientific Research Fund
of the Hunan Provincial Education Department (No. 05B037).� It was also partially supported by the LB595
grant from the State of Nebraska, USA. HWD was partially supported by grants
from the US National Institutes of Health (Nos. K01 AR02170-01A2, R01 GM60402,
5R01 AR050496-02, and R21 AG027110-01A1)
Abbreviations:
2-DE, two-dimensional electrophoresis; CMCs, circulating monocytes; GO, gene
ontology; MS, mass spectrometry; MS/MS, tandem mass spectrometry
#These authors
contributed equally to this work
*Corresponding
author: Tel, 86-731-8872791; Fax, 86-731-8872791; Email, [email protected]
A key
assumption in studying mRNA expression is that it is informative in the prediction
of protein expression. However, only limited studies have explored the
mRNA-protein expression correlation in yeast or human tissues and the results
have been relatively inconsistent. We carried out correlation analyses on
mRNA-protein expressions in freshly isolated human circulating monocytes from
30 unrelated women. The expressed proteins for 71 genes were quantified and
identified by 2-D electrophoresis coupled with mass spectrometry. The
corresponding mRNA expressions were quantified by Affymetrix gene chips.
Significant correlation (r=0.235, P<0.0001) was observed for
the whole dataset including all studied genes and all samples. The correlations
varied in different biological categories of gene ontology. For example, the
highest correlation was achieved for genes of the extracellular region in terms
of cellular component (r=0.643, P<0.0001) and the lowest
correlation was obtained for genes of regulation (r=0.099, P=0.213)
in terms of biological process. In the genome, half of the samples showed
significant positive correlation for the 71 genes and significant correlation
was found between the average mRNA and the average protein expression levels in
all samples (r=0.296, P<0.01). However, at the study group
level, only five studied genes had significant positive correlation across all
the samples. Our results showed an overall positive correlation between mRNA
and protein expression levels. However, the moderate and varied correlations
suggest that mRNA expression might be sometimes useful, but certainly far from
perfect, in predicting protein expression levels.
Keywords������� protein expression; mRNA expression; circulating monocytes;
gene function; correlation
Functional genomics, dealing with the functional analysis of genes and their products [1], is an active research area in the post-genomic era. Genetic information stored in DNA is translated into proteins with the intermediate of mRNA, and proteins are directly involved in almost all processes of life by carrying out various biological functions. Extensive studies have been carried out at the mRNA level and protein level separately for illuminating functions of genes. The studies at the mRNA level are largely based on a key assumption that mRNA expression is informative in predicting protein expression level in relation to gene functions. However, the degree of validity of this key assumption has seldom been investigated, and is certainly far from being established.
Powerful high-throughput methods for measuring both mRNA and protein expressions are currently available, such as DNA microarray technologies for mRNA profiling, and 2-D electrophoresis (2-DE) or liquid chromatography coupled with mass spectrometry (MS) approaches for protein profiling. Those technologies make it possible to study gene expression at both transcriptome and proteome levels, and provide data for assessing statistical correlations between mRNA and protein expression levels.
So far, only a few studies have investigated the correlation of mRNA and protein expression levels in yeast [2-4] and human tissues [5-7], and the results have been relatively inconsistent.
In yeast, Gygi et al found that even similar mRNA expression patterns could be accompanied by large variations (more than 20-fold difference) of protein abundances, and vice versa [2]. Conversely, Futcher et al found a relatively high genomic mRNA and protein expression correlation (r=0.76) after normalizing the expression data using a logarithmic method [3]. Greenbaum et al found significant positive correlations between protein and mRNA expression in both global range and biological subcategories (such as cell cycle and cell rescue) (r=0.20-0.89), by merging many yeast protein-abundance datasets [4].
In humans, Anderson and Seilhamer examined 19 selected mRNA and corresponding protein abundances in a normal human liver expression database and found a positive correlation (r=0.48) [5]. However, a study on the expression correlation of three genes in 17 human prostate cancer tissue samples displayed no significant result for any gene [7]. Among 98 studied genes, 21 showed a significant positive correlation between mRNA and protein expression levels (r>0.2445, P<0.05) for a cohort of 85 lung cancer adenocarcinomas and normal lung samples [6].
The largely inconsistent results in both yeast and human tissues
suggested that gene expressions at the transcriptional and translational levels
might be different. Here, to identify the relationship between gene expression
at the mRNA and protein levels in vivo in normal human cells, we carried
out an exhaustive correlation analyses of mRNA and protein expression
abundances in human circulating monocytes (CMCs). Relative to using human
tissue samples that could be composed of various proportions of different cells,
we attempted to study a specific and relatively homogeneous human cell type,
trying to avoid potential confounders due to various mixtures of different
cells in tissue samples.
Materials and Methods
Subjects
The samples of this study came from an ongoing genetic project on osteoporosis [8] approved by the Research Administration of Hunan Normal University (Changsha, China). All subjects signed informed consent documents before entering the project. Thirty unrelated women were recruited and each donated 50 ml peripheral blood for this study. They were all randomly selected from a sample composed of 1000 healthy and normal adult Chinese Han females aged 20-45 years, living mainly in Changsha city in southern China. The sampling scheme [8] and exclusion criteria [9] have been detailed elsewhere. Briefly, the exclusion criteria included chronic disorders involving vital organs (heart, lung, liver, kidney, or brain), serious metabolic diseases such as diabetes, hypo- or hyperparathyroidism , hyperthyroidism, other skeletal diseases such as Paget抯 disease, osteogenesis imperfecta, rheumatoid arthritis, chronic use of drugs affecting bone metabolism such as corticosteroid therapy, anticonvulsant drugs, estrogens, thyroid hormone, and malnutrition conditions such as chronic diarrhea and chronic ulcerative colitis.
Monocyte extraction and
purification
Mononuclear cells were isolated from the blood samples using Lymphoprep solution (Axis-Shield, Oslo, Norway), and CMCs were isolated from mononuclear cells using Dynal monocyte negative isolation kit (Dynal Biotech, Oslo, Norway). The purity of the isolated monocyte sample was monitored by BD FACSCalibur flow cytometry (BD Biosciences, San Jose, USA) with fluorescence-labeled antibodies PE-CD14 and FITC-CD45, and determined to be as high as 90%. The obtained CMCs from each sample were divided into two equal portions for RNA and protein extraction.
Protein profiling and
identification
The experimental procedures for protein profiling and identification were detailed elsewhere [10]. Briefly, total cell lysates were obtained from monocyte samples and the proteins were quantified. To use as many samples as possible under our budget constraint, we used a sample pooling scheme for 2-DE [11]. Three protein samples were pooled with equal aliquots to form a mixed sample, thus 30 subjects were divided into 10 pooled samples for 2-DE. The composition of pooled samples is given in Table 1. Each pooled sample was subject to triplicate 2-DE, running on the IPGphor isoelectric focusing system (Amersham Pharmacia Biotech, Pittsburgh, USA) and Protein II xi cell electrophoresis apparatus (Bio-Rad Laboratories, Hercules, USA). PDQuest software (version 7.2; Bio-Rad Laboratories) was used to quantify the expression levels for the target protein spots on triplicate silver-stained 2-DE gels. In order to justify for variability due to silver staining, the intensity of each spot was normalized to the total intensity of all the spots in the gel. The normalized data were exported as expression values for further statistical analyses. The target spots were excised and trypsin digested into peptides. The extracted peptides were analyzed by MS and tandem MS (MS/MS) using a matrix-assisted laser desorption/ionization (MALDI)-time of flight/time of flight (TOF/TOF) Ultraflex instrument (Bruker Daltonik, Bremen, Germany). Finally, the MS and MS/MS data generated from the same protein spot were combined and searched against the NCBInr database of Homo sapiens (version 2.23.2005) using Mascot software (version 2.1; Matrix Science, Boston, USA). The proteins were identified using a probability-based Mowse algorithm [12]. A -10log(P) score, where P is the probability that the observed match is a random event, of >30 (for MS and MS/MS data), was regarded as significant.
mRNA quantification
The total RNA from monocytes was extracted using a Qiagen RNeasy Mini Kit (Qiagen, Valencia, USA) following the procedures recommended by the manufacturer. Individual samples were used for DNA microarray study. Preparation of cRNA, hybridization, and scanning of the Human Genome U133 Plus 2.0 GeneChip (Affymetrix, Santa Clara, USA) were carried out according to the manufacturer's protocol. The expression values were generated with Microarray Suite 5.0 software (Affymetrix). Expression data at the mRNA level were only available for 21 out of the 30 subject samples that had protein expression data (noted in Table 1). To match the corresponding protein expression abundances in each pooled sample in the protein analyses, the average mRNA expression value of the component individual samples in each pooled sample was calculated and used in subsequent analyses.
����
Statistical analysis
The relationship between the levels of mRNA and protein expression was examined using the Pearson's product-moment correlation coefficient statistical method. P values and interval estimation of correlation coefficients (r) were obtained through Fisher's Z transformation [13]. P<0.05 (two-sided) was considered statistically significant. The percentage of protein expression variation explained by corresponding mRNA expression was calculated as r2. As we used the pooling scheme in the protein expression study, it is important to prove that correlation analyses based on this scheme are feasible. In Appendix I (http://www.abbs.info/fulltxt/40-05/supplement08055.htm), we present a detailed analytical rationale justifying our analyses using averages in the pooled samples. Some of our pooled proteins and the corresponding mRNA samples are not perfectly matched so that, for three subjects composing one pooled protein sample, in a few cases only one or two of them were analyzed for mRNA expression (Table 1). We show in Appendix I that the correlation coefficients derived from this study are lower bounds approximating true values.
Six different types of correlation analyses were carried out, to reflect various aspects of correlation of mRNA-protein in the genomic and/or study group levels. To aid in comprehension of the computation and thus biological significance of these various types of correlations, the mathematical symbols used to denote the data and the resultant correlation coefficients are outlined below.
i: the ith gene, i=1, 2, , 71.
j: the jth pooled sample, j=1, 2, , 10.�
: the mRNA expression value for the ith gene in the jth pooled sample.
: the protein expression value for the ith gene in the jth pooled sample.
: the mRNA expression values for all 71 genes in the jth pooled sample.
: the protein expression values for all 71 genes in the jth pooled sample.
: the mRNA expression values in all 10 pooled samples for the ith gene.
: the protein expression values in all 10 pooled samples for the ith gene.
: the average mRNA expression value of the 10 samples for the ith gene.
: the average protein expression value of the 10 samples for the ith gene.
��� The analyses include the following:
1��� An all-against-all correlation for all the studied genes in all the pooled samples between mRNA and protein expression values was conducted to assess the correlation characteristics across both genomic and group levels. In this correlation, the x, y variables, correlation coefficient, and P value are , , r1, and P1, respectively.
2 ���On a genomic scale of individual samples, correlation for all the studied genes was first calculated for each pooled sample then an average correlation coefficient was obtained. The variables and parameters in this correlation analysis are , , r2j and P2j , respectively. The average correlation coefficient of r2j is denoted by .
3��� To investigate the relation between the correlation of average expression values in the study group and the average correlation (computed above) among different individuals, we obtained the average mRNA and protein expression values of all samples, then carried out the correlation using these average values. , , r3, and P3 represent the variables and parameters in this analysis.
4��� To explore whether different genes behave differently on their mRNA-protein expressions at the group level, for each gene, we calculated the correlation of its mRNA and protein expression values across the pooled samples. The variables and parameters for this analysis are , , r4i, and P4i.
5��� To assess the mRNA-protein correlations among different biological categories, we conducted expression correlation analyses for different gene clusters in different biological categories. We first classified the 71 genes according to gene ontology (GO) principles using GFINDer software [14], then carried out the expression correlation for each gene cluster across all the pooled samples. , , r5m, and P5m are the variables and parameters involved (m, the mth biological category, m=1, 2, , 12;� jm, the jth gene in the mth biological category, jm ranges from 1 to the total number of genes in the mth biological category).
6��� To find whether the expression correlation is dependent on expression values per se, correlation analysis was carried out between the mRNA-protein correlation coefficients and average mRNA or protein expression values of all samples. In this correlation, x, y, correlation coefficient, and P value are r2j, , r6x/r6y, and P6x/P6y, respectively.
Results
The characteristics of the studied samples are shown in Table 1. We selected 105 protein spots on 2-DE gels for quantification and identification. Some proteins migrated to more than one spot (commonly seen in proteomics, possibly due to differential protein processing or modification) and abundances for such proteins were calculated by integrating the intensities of the different spots. In total, protein quantities of 71 genes were determined (Table 2). The range of average protein expression values after normalization was from 12580.07 (ENO1) to 161743.13 (PARK7), whereas that of average mRNA expression values was from 3.45 (PKLR) to 5578.65 (ACTB). All of the mRNA expression data was submitted to the National Center for Biotechnology Information's Gene Expression Omnibus database (http://www.ncbi.nlm.nih.gov/geo/, accession No. GSE7158).
Correlation for all genes in
all samples
mRNA expression abundance was highly significantly positively correlated with the protein expression level for 71 genes in all the samples (r1=0.235, P1<0.0001). Fig. 1(A) shows that increasing mRNA expression is generally accompanied by increasing protein expression, but with relatively large noise. Gene expression variation at the mRNA level accounts for approximately 5.52% of the protein expression variation.
Correlation for all genes in
each pooled sample
Positive mRNA-protein correlation was observed for all genes in each pooled sample, with r2j (j=1, 2, , 10) ranging from 0.109 to 0.443 and the percentage of protein variation explained by mRNA variation ranging from 1.19 to 19.62 (Table 3). Half of the correlations displayed statistical significance (r>0.234, P<0.05), and another two (j=5, r25=0.217; j=6, r26=0.230) showed marginal significance (P25=0.069 and P26=0.070). The� was 0.270 with a standard deviation of 0.111.
Correlation between average
mRNA and average protein expression values
Significant correlation was detected using the average mRNA and protein expression values of each of the 71 genes across the human genome (r3=0.296, P3<0.01) [Fig. 1(B)]. The r3 was almost equal to the .
Correlation across all the
samples for each gene
The r4i (i=1, 2, �, 71) ranged from -0.497 to 0.892. Although relatively large correlation coefficients were achieved, only five genes (ATIC, SLA2, TPI1, PRDX3, and RSU1) showed significant positive correlations (Table 2), and two of the negative correlations were significant.
Correlation for each gene
category in all samples
For each GO principle (biological process, cellular component, and molecular function), the corresponding gene categories at level one were studied. Table 4 shows the correlation results for each GO category with the percentage of genes over 5% (the proportion of the number of genes belonging to the GO category to that of the parent ontology). There are overlaps among different categories due to some genes with multiple functions.
The correlation strength for most of the categories turned out to be
similar to the whole gene set.� However,
genes belonging to regulation (n=16, GO: 0050789, r53=0.099, P53=0.213) and development (n=10,
GO: 0050896, r54=0.163, P54=0.105)
of biological processes had non-significant and lowest mRNA-protein expression
correlations. However, genes belonging to the extracellular region (GO:
0005576, n=15, r58=0.643, P58<0.0001) of cellular component and signal transducer activity
(GO: 0004871, n=9, r512=0.476, P512<0.0001) of molecular function showed significant and highest
correlation (Table 4). Interestingly, the percentage of protein
expression variation explained by mRNA expression variation varied from 0.98%
to 41.35% for different categories studied here.
mRNA-protein correlation
strength is independent of mRNA or protein expression abundance per se
No significant relationship was observed for mRNA, r6x=0.034, P6x>0.05, or for protein, r6y=0.082, P6y>0.05 (Fig. 2).
Discussion
In this study, we investigated the mRNA and protein expression levels of 71 genes in human CMCs and carried out comprehensive correlation analyses on different levels and aspects of mRNA-protein expression. This is the first expression correlation study in a specific type of normal human cell. We found that, overall, gene expression at the mRNA level is generally correlated with that at the protein level for all the genes studied, reflected by the findings in the whole studied samples (the all-against-all correlation and correlations for GO categories) and in individual subjects (at least 50% of the subjects showed significant positive correlations). Interestingly, we also found that mRNA-protein expression correlation largely varies for different genes and different gene biological categories. These findings confirmed previous results in human normal liver cells [5], lung adenocarcinomas [6], and in yeast [4] on a genomic scale. All of these studies found an overall significant positive correlation between mRNA and protein expression. However, this study and others also suggested that the correlation strength varies among genes, and only a subset of the studied mRNAs had a significant positive correlation with the protein expression.
At the genomic level, interestingly, we found positive correlations in the whole study group (r1=0.235, P1<0.0001) [Fig. 1(A)], individual samples (r2j=0.109-0.443) (Table 3), and the average expression level (r3=0.296, P3<0.01) [Fig. 1(B)]. Although r2j displayed variation among different samples, the average value (=0.270) showed consistency with the other two coefficients (r1=0.235 and r3=0.296). Those overall similar positive correlations suggest that there is a general approximate concordance between the mRNA and the protein expression quantities on the genomic scale, both for the group and the individuals. However, alarmingly, the mRNA variation could only account for less than 9% of the protein variation in all the three types of correlation analyses, which might indicate that, at the genomic level, mRNA expression poorly predicts the corresponding protein expression level. Our results also suggest that mRNA-protein expression correlation might also vary in different individuals. Whether this initial finding holds, especially in relation to the health status of individual subjects, might deserve further investigation.
For individual genes across all the samples, however, only five genes (Table 2) showed significant positive mRNA-protein correlations. For these five genes, expression at the mRNA level could explain as much as 69.22% of the expression variation at the protein level. These five genes were ATIC, SLA2, TPI1, PRDX3, and RSU1. They are involved in important molecular functions such as protein binding, enzyme activities, and signal transduction. This finding suggests that some biologically critical genes possibly have highly correlated mRNA-protein expression and those key high correlations might be important for the whole expression system for critical organism function or health.
However, lack of significant correlation findings for other genes could be due to technological and biological reasons, as well as the limitation of the experimental design of this study. First, currently available technological methods are not perfectly accurate in either mRNA or protein quantifications. These factors potentially influence the detection of biologically truly significant correlations. Second, complicated biological processes, such as transcriptional splicing, post-transcriptional splicing, translational modifications, translational regulation, and protein complex formation, might affect the relative quantities of mRNA and protein of various genes to various degrees. Third, different biological or experimental mRNA and protein degradation rates might affect the mRNA and protein correlations. Fourth, different mRNA secondary structures might result in different protein translation efficiencies. Finally, the pooling scheme resulting in a relatively small sample size (n=10) might limit the power to detect significant mRNA-protein correlations. As pointed out in "Materials and Methods", and shown analytically in Appendix I, missing information regarding mRNA expression for some subjects, as indicated in Table 1, renders the correlations computed in this study the lower bounds (approximately two-thirds) of true correlations. These factors considered, it suffices to say that the correlations found in this study are lower bounds of true biological correlations.�
We found two genes with significant negative expression correlation. This might be due to artifacts due to the technical noise, and/or statistical multiple testing, or a novel mRNA to protein regulation mechanism (e.g., negative feedback) that might need to be explored.
Genes belonging to the same functional category might share similar stringency or regulatory mechanisms for the expressions of mRNA and corresponding proteins. This could underlie the findings here that the correlations vary by different gene category. Functionally similar categories of genes/proteins might contribute importantly to the degree of correlation between mRNA and protein expression [4]. Motivated by this hypothesis, we grouped the genes into 12 categories according to three GO principles (biological process, cellular component, and physical biology) and carried out expression correlations for genes in each category across all the samples. All the 12 categories showed positive correlations, and categories of biological process showed relatively low correlations compared to other categories. The non-significant correlation for "regulation" and "development" categories of biological process (Table 4) might suggest that genes involved in regulation and development of a biological process undergo relatively less stringent translational regulation with more discordant expression patterns at both the mRNA and protein levels, in the specific cell type under study, CMCs. This finding might be surprising given that, intuitively, genes for "regulation" and "development" categories of biological process should be under stringent regulation for translation. This result could be related to the roles of these genes in CMCs and they might be under different translation regulation mechanisms in other cell types that could be more relevant to growth and development.
Some studied genes were classified into more than one functional category, which might mask the correlation difference of different categories. The difference would be more apparent if more genes were involved in the study. However, the percentage of protein variation explained by mRNA still showed a relatively large range (from 0.98% to 41.35%) for different categories. Most of the categories showed significant mRNA-protein correlations, substantiating that gene expression at transcription and translation levels were correlated globally. However, the large difference among the percentage of protein expression accounted for by mRNA could suggest that there is relatively large expression divergence between the transcription and translation levels for genes in different biological categories.
In normal human CMCs, our correlation results in both individual genes and different categories showed big variations. We speculate that in different developmental stages or disease conditions, the expression correlations for specific genes or biological categories might be more discordant, for example, some genes or gene categories have significant and high correlations and other genes or gene categories showed non-significant or low correlations, all perhaps related to the specific functions of the genes in specific cell types. It is conceivable that the same gene or gene category has very different mRNA-protein correlations in different cell types. Therefore, in gene expression studies for development or diseases, the developmental or disease gene expression measurement at the mRNA level using methods such as real-time reverse transcription-polymerase chain reaction might not be sufficient and protein level studies using, for example, Western blot analysis, are necessary in cells that are related to the development or disease etiology under study.
Studies suggested that mRNA or protein expression abundance per se is probably a factor influencing the mRNA-protein expression correlation [4,6]. However, we found that the mRNA-protein expression correlation coefficients are independent of the mRNA or protein abundances, results that are not in agreement with the findings in yeast [4] but are consistent with those from human lung adenocarcinomas [6]. Whether this discrepancy represents the species or cell type differences under study is not clear.
Proteins are the major direct executors of life processes and might reflect gene function more directly than mRNA. In addition, some messengers are transcribed but not translated, thus the number of mRNA copies does not necessarily reflect the number of functional protein molecules. However, mRNA expression is still informative to protein expression, given the study here and those reported earlier. Therefore, both mRNA and protein expression studies are important and complementary to each other. In combination, they should provide us with a more detailed and comprehensive understanding of gene function. The positive correlation between mRNA and protein expression as observed in human CMCs (in the current study) and in human cancer tissues [6,7], as well as in yeast [3,4], are biologically significant and worth further investigation. Currently, a challenge facing scientists is to develop and implement methodologies to integrate and analyze large-scale datasets at different levels of gene/protein expression to unravel the underlying regulatory processes or complex metabolic networks in vivo. Ideker et al reported on an integrated approach to build, test, and refine a model of a cellular pathway, in which perturbations to critical pathway components are analyzed using DNA microarray, quantitative proteomics, and databases of known physical interactions [15]. In addition, the models jointly analyzing transcriptomic and proteomic data of Desulfovibrio vulgaris [16] correctly predicted an abundance of undetected proteins in some cases. With better understanding of the relationship between mRNA and protein expression, we can optimize combining analyses and provide additional insights into the metabolic mechanisms underlying complex biological systems [17,18].
In summary, this is the first multilevel expression correlation study in normal human cells, CMCs. CMCs are an important cell type in the human immune system. They play an important role in many physiological and pathological conditions, such as cancer, inflammation, infection, and bone metabolism. Our findings indicated that gene expression at the mRNA level is generally informative but not predictive for that at the protein level. Thus, functional genomics approaches at both mRNA and protein levels are necessary and complementary for human complex disease studies.
References
�
1�� Celis JE, Kruhoffer M, Gromova I, Frederiksen
C, Ostergaard M, Thykjaer T, Gromov P et al. Gene expression profiling:
monitoring transcription and translation products using DNA microarrays and
proteomics. FEBS Lett 2000, 480: 2-16
2�� Gygi SP, Rochon Y, Franza BR, Aebersold R.
Correlation between protein and mRNA abundance in yeast. Mol Cell Biol 1999,
19: 1720-1730
3�� Futcher B, Latter GI, Monardo P, McLaughlin
CS, Garrels JI. A sampling of the yeast proteome. Mol Cell Biol 1999, 19: 7357-7368
4�� Greenbaum D, Colangelo C, Williams K,
Gerstein M. Comparing protein abundance and mRNA expression levels on a genomic
scale. Genome Biol 2003, 4: 117
5�� Anderson L, Seilhamer J. A comparison of
selected mRNA and protein abundances in human liver. Electrophoresis 1997, 18:
533-537
6�� Chen G, Gharib TG, Huang CC, Taylor JM, Misek
DE, Kardia SL, Giordano TJ et al. Discordant protein and mRNA expression
in lung adenocarcinomas. Mol Cell Proteomics 2002, 1: 304-313
7�� Lichtinghagen R, Musholt PB, Lein M, Romer A,
Rudolph B, Kristiansen G, Hauptmann S et al. Different mRNA and protein
expression of matrix metalloproteinases 2 and 9 and tissue inhibitor of
metalloproteinases 1 in benign and malignant prostate tissue. Eur Urol 2002,
42: 398-406
8�� Tan LJ, Lei SF, Chen XD, Liu MY, Guo YF, Xu
H, Sun X et al. Establishment of peak bone mineral density in Southern
Chinese males and its comparisons with other males from different regions of
China. J Bone Miner Res 2007, 25: 114-121
9�� Deng HW, Xu FH, Huang QY, Shen H, Deng H,
Conway T, Liu YJ et al. A whole-genome linkage scan suggests several
genomic regions potentially containing quantitative trait loci for
osteoporosis. J Clin Endocrinol Metab 2002, 87: 5151-5159
10� Deng FY, Liu YZ, Xiao P, Jiang C, Li LM, Wu S,
Chen Y et al. In vivo proteomics analysis of circulating
monocytes in Chinese females underlying osteoporosis. Proteomics 2008, in Press
11� Peng X, Wood CL, Blalock EM, Chen KC,
Landfield PW, Stromberg AJ. Statistical implications of pooling RNA samples for
microarray experiments. BMC Bioinformatics 2003, 4: 26
12� Perkins DN, Pappin DJ, Creasy DM, Cottrell JS.
Probability-based protein identification by searching sequence databases using
mass spectrometry data. Electrophoresis 1999, 20: 3551-3567
13� Rosner B. Fundamentals of Biostatistics. 5th
edn. Pacific Grove: Duxbury Press 2005
14� Masseroli M, Galati O, Pinciroli F. GFINDer:
genetic disease and phenotype location statistical analysis and mining of
dynamically annotated gene lists. Nucleic Acids Res 2005, 33: W717-W723
15 Ideker T, Thorsson V, Ranish JA, Christmas R,
Buhler J, Eng JK, Bumgarner R et al. Integrated genomic and proteomic
analyses of a systematically perturbed metabolic network. Science 2001, 292:
929-934
16� Nie L, Wu G, Brockman FJ, Zhang W. Integrated
analysis of transcriptomic and proteomic data of Desulfovibrio vulgaris:
Zero-inflated Poisson regression models to predict abundance of undetected
proteins. Bioinformatics 2006, 22: 1641-1647
17� Lee PS, Shaw LB, Choe LH, Mehra A,
Hatzimanikatis V, Lee KH. Insights into the relation between mRNA and protein
expression patterns: II. Experimental observations in Escherichia coli.
Biotechnol Bioeng 2003, 84: 834-841
18� Mehra A, Lee KH, Hatzimanikatis V. Insights
into the relation between mRNA and protein expression patterns: I. Theoretical
consideration. Biotechnol Bioeng 2003, 84: 822-833