* Laboratory of Molecular Immunology,
Pulmonary Division,
Department of Medicine, Brigham & Women’s Hospital, Harvard Medical School, Boston, Massachusetts
Correspondence: David L. Perkins, Brigham & Women’s Hospital, PBB-170, 75 Francis St., Boston, MA 02115. E-mail: dperkins{at}rics.bwh.harvard.edu
 TOP ABSTRACT INTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
, MIP-2, CD8,
IP-10, and RANTES), each with a unique profile of transcriptional
expression, that reproduce the differentiation of states of rejection
in dendrograms. Using histology and immunohistochemistry, we correlated
differential regulation of CD4 and CD8 at the levels of mRNA and
protein. Our strategy of data reduction successfully decreased the
number of genes to nine, which are sufficient to differentiate distinct
states of rejection in our experimental protocol.
Key Words: chemokines • cytokines • transplantation • self-organizing maps
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
The advantage of the gene-chip approach is the possibility of serendipitous discoveries and the numerous data points accumulated; however, many of these data points involve genes that have expression profiles that are negative, unchanged, or not relevant to the biological question. An alternative strategy to profiling the complete transcriptome is to analyze groups of candidate genes selected on the basis of prior knowledge of relevant function. The advantages of focusing on a group of candidate genes are the decreased amount of data analysis and lower cost. A recent study that analyzed the expression of more than 5000 genes in breast tumors with microarrays differentiated BRCA1+, BRCA2+, and nonfamilial tumors based on a subset of only 51 genes [4 ]. In the current study, we investigated a panel of 60 candidate genes that includes inflammatory molecules, cytokines, chemokines, chemokine receptors, and cell-surface markers that previously have been shown to be important mediators of inflammation and immunity.
To differentiate allograft rejection in terms of kinetic development and response to a treatment protocol that blocks CD40 ligand (CD40L) costimulation, we used two types of algorithms. First, the dissimilarity of each rejection profile was compared by a clustering algorithm and was visualized in hierarchical dendrograms. Second, we grouped genes with similar patterns of expression by self-organizing maps (SOM) into subsets that were regulated coordinately. Cluster is an agglomerative hierarchical algorithm that calculates the dissimilarity among experimental groups based on summation of Pearson correlation coeffecients, and it has been applied to the analysis of gene expression profiles detected with microarrays [6 ]. SOM are generated by an iterative nonparametric algorithm, which calculates euclidean distance, and has been applied to engineering problems, interpretation of electroencephalograms, economic data, as well as to gene-expression profiles [7 , 8 ]. In our study, the combination of these algorithms defined a molecular basis for allograft rejection that differentiated multiple experimental groups including different time points and treatment protocols.
Our experiments examined the global effects of inhibition of an important pathway in allograft rejection, the CD40/CD40L-costimulatory pathway. As previously shown, allograft survival in murine and primate models is prolonged markedly by CD40L blockade [9 10 11 ]. We integrated the profiles of expression of the 60 candidate genes during allograft rejection by cluster analysis into a single dendrogram that differentiated a molecular basis for the delayed rejection following CD40L blockade compared with untreated acute rejection. In addition, cluster analysis clearly distinguished the kinetic stages of the rejection process into an early and late phase, which correlated with the expression of different subsets of genes. Further, confirming the robustness of the SOM algorithm for data analysis, we demonstrate that the selection of only nine genes, one from each subset defined by SOM, is sufficient to recapitulate the differentiation of the experimental groups in dendrograms produced by cluster analysis. Thus, our strategy of data reduction successfully decreased the number of genes, which are necessary to differentiate distinct states of rejection in our experimental protocol, to nine.
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
Mice
Eight- to 12-wk-old male mice, including BALB/cByJ (BALB/c;
H-2d), C57BL/6J (B6; H-2b),
C57BL/6J-Rag1tm1Mom (B6-Rag KO; H-2b; JAX, Bar
Harbor, ME), and BALB/c-AnNTac-Rag2tm1N12 (BALB/c-Rag KO;
H-2d; Taconic, Germantown, NY), were used as donor and
recipients in the transplant experiments. Experimental groups include
syngeneic (B6
B6), alymphoid (B6-RAGBALB/c-RAG), and allogeneic
(B6BALB/c). In the studies of CD40L blockade, allogeneic recipients
were treated with 250 µg monoclonal antibody (mAb) MR1
[13
] on days 0 and 1 after transplantation. Mice are
maintained in vented racks with constant temperature and humidity in
our animal facility under virus antibody-free conditions.
Ribonuclease protection assay (RPA)
Chemokine, chemokine receptor, and CD marker expression were
analyzed by RPA. Briefly, total RNA was isolated from hearts using
RNAzol and analyzed using the RiboQuant Multi-Probe RPA system
(Pharmingen, San Diego, CA). RNA (15 µg) was used per hybridization
and RNase reaction with the templates mCK1 [interleukin (IL)-4, IL-5,
IL-10, IL-13, IL-15, IL-9, IL-2, IL-3, and interferon-
(IFN-)],
mCK2 [IL-12p35, IL-12p40, IL-10, IL-1, IL-1ß, IL-1RA, IL-18,
IL-6, IFN-, and migration inhibitory factor (MIF)], mCK3 [LT,
LTß, tumor necrosis factor (TNF-), IL-6, IFN-, IFN-ß,
transforming growth factor (TGF)-ß1, TGF-ß2, TGF-ß3, and MIF],
mCK-5 [MIF, RANTES (regulated on activation, normal T expressed and
secreted), eotaxin, macrophage-inflammatory protein
(MIP)-1ß, MIP-1, MIP-2, IP-10, and monocyte chemoattractant
protein-1 (MCP-1)], mCR-5 (CCR1, 1ß, 4, 5, and 2), a custom template
[inducible nitric oxide synthase (iNOS), CXCR2, 3, 4, 5, CCR6, 8,
and 8ß], and mCD-1 [T-cell receptor (TCR)-
, TCR-, CD3
,
CD4, CD8, CD8ß, CD19, F4/80, and CD45; Pharmingen, San Diego,
CA]. The IP-10 template detects the C57BL/6 allele
[14
]. The protocol was modified by labeling probes with
35S. After hybridization with the 35S-labeled
probes, the samples were treated with RNase and purified according to
the manufacturer’s protocol. The protected probes were electrophoresed
on a denaturing 5% polyacrylamide gel. The gels were exposed in a
Molecular Dynamics phosphorimager. The identity of each protected
fragment was established by analyzing its migration distance against a
standard curve of the migration distance versus the log nucleotide
length for each undigested probe. Samples were normalized to the
housekeeping gene, glyceraldehyde 3-phosphate dehydrogenase (GAPDH).
Protected bands were quantitated by densitometry analysis using
ImageQuant software (Molecular Dynamics, Sunnyvale, CA). From 18
experimental groups, a total of 1080 values were calculated as means
from 1708 RPA results and were subjected to cluster analysis.
Immunohistochemistry
Graft hearts from transplant recipient or control mice were
embedded in ornithine carbamyltransferase compound, snap-frozen in
liquid nitrogen, and cut into 4-µ sections. After mounting onto glass
slides, tissue was fixed in acetone for 10 min and blocked with normal
mouse serum diluted 1:10. Primary antibodies were rat anti-CD8 (20
µg/mL; Pharmingen, LaJolla, CA ), anti-CD4 (20 µg/mL; Pharmingen,
LaJolla, CA ), and anti-F4/80 (5 µg/mL; Serotec, Raleigh, NC). Slides
were incubated in the primary antibody overnight at 4°C, followed by
a biotinylated mouse anti-rat secondary antibody (1:500; Jackson, West
Grove, PA) for 2 h at 4°C, a methanol and hydrogen peroxide wash
(10:1), and a 30-min incubation in streptavidin horseradish peroxidase
(1:1000; Jackson). The tissue was stained with 3'-diaminobenzidine
tetrahydrochloride (DAB) substrate and counterstained with methyl green
(Sigma Chemical Co., St. Louis, MO). Images were captured using a Leica
DMLB microscope interfaced with Leica Q500IW image analysis software.
Statistics
Graft survival data were calculated as mean ±
SD. Survival data were based on six to eight transplants
per group. Comparisons among groups were performed using a
nonparametric Mann-Whitney test, and differences were considered
significant at P < 0.05. Differential expression of
mRNA determined by RPA on days 0, 1, 3, 5, and 7 in the syngeneic,
allogeneic, and MR1-treated recipients and at the time of rejection
(days 32–33) in the MR1-treated group was analyzed by two-factor
analysis of variance (ANOVA) with the variables gene expression and
experimental group. Statistical significance of variances was
calculated for P < 0.05 by F-test.
Cluster analysis
Cluster analysis was performed using Cluster and TreeView
[6
] (courtesy of M. Eisen, Lawrence Livermore Radiation
Laboratory, Berkeley, CA) and GeneCluster (courtesy of Whitehead
Institute for Biomedical Institute, Cambridge, MA) software. RPA values
were calculated as percent of GAPDH and analyzed by Cluster without
normalization using the hierarchical clustering algorithm with complete
linkage clustering. Briefly, dissimilarity is determined by calculation
of the Pearson correlation coefficient among each series of values from
each experimental group [6
]. After processing, the
dendrogram was visualized by TreeView. SOM were generated by
GeneCluster without normalization using a 3 x 3 geometry of nine
seed maps [8
]. Nine maps were selected empirically to
eliminate clusters with few genes or large standard deviations. The
centroids and standard deviations of the groupings were analyzed using
50 epochs. Additional epochs did not alter the gene clusters of the
maps.
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
B6) recipients (Table 1
). These grafts express a complete major histocompatibility complex
(MHC) mismatch and generate a robust alloimmune response that results
in rejection determined by the cessation of palpable heartbeats at
approximately day 8 following transplantation. The complete panel of 60
candidate genes was analyzed by RPA at days 1, 3, 5, and 7 following
transplantation. The results were quantitated by densitometry and
analyzed by Cluster software (Fig. 1A
). Comparisons among experimental groups are based on an
agglomerative algorithm that calculates the dissimilarity among each
series of values of gene expression. These results clearly segregate
the late (days 5 and 7) and early (days 1 and 3) profiles of the
responses in the allogeneic group. As expected, the BALB/c and B6
untransplanted control samples from untransplanted hearts are nearly
identical. In addition, the control samples are more similar to the
early, than the late, allogeneic groups. |
View this table: [in a new window] |
Table 1. Experimental Groups (CD40L Blockade Was Produced)
|
 View larger version (21K): [in a new window] |
Figure 1. Dendrogram of kinetic analysis of rejection profile in transplant
recipients. The level of expression of a panel of 60 candidate genes
including inflammatory molecules, cytokines, chemokines, chemokine
receptors, and cell-surface markers was determined at days 1, 3, 5, and
7 following transplantation by RPA of graft hearts and was analyzed by
Cluster and Tree software [6
] in allogeneic (a),
syngeneic (b), and alymphoid (c) graft recipients, which are described
in Table 1
. Controls are untransplanted BALB/c and B6 hearts. The
degree of dissimilarity is proportional to the total length of the
horizontal axis among groups.
|
B6) protocol with two control groups:
a syngeneic (B6B6) protocol that contains a complete array of
functional lymphoid and inflammatory cells (however, because of the
absence of alloantigen, no adaptive immune response should be
stimulated) and an alymphoid (BALB/c-RAG KOB6-RAG KO) protocol that
lacks functional T and B lymphocytes because of deletion of the
recombinase-activating gene (RAG) in the donor and recipient mice but
contains all other lineages of inflammatory cells including natural
killer (NK) cells (Table 1) . Kinetic analysis of the syngeneic groups
shows that the day 1 group is most dissimilar from the other groups
including days 3, 5, and 7, which show only a minor degree of
difference (Fig. 1b)
. Similarly, analysis of the alymphoid groups also
shows that the day 1 group is most dissimilar (Fig. 1c)
. Taken
together, these results demonstrate that in all three experimental
groups, the profile differs markedly during the early phase, in
particular at day 1. In contrast, the day 7 profile differs only in the
allogeneic protocol. Because the early day 1 changes occur in the
syngeneic group in the absence of alloantigen and in the alymphoid
group without functional lymphocytes, we conclude that the early
response does not require an adaptive immune response.
Comparison of the innate and adaptive groups
To directly analyze the response contributed by the adaptive and
nonadaptive immune responses, we compared the allogeneic and alymphoid
groups by cluster analysis (Fig. 2A
). The resulting dendrogram shows the marked dissimilarity of the
allogeneic day 7 group, with a lesser degree of dissimilarity between
the allogeneic and alymphoid groups on day 1. As expected, the BALB/c
and B6 control samples were nearly identical and only slightly
different from the alymphoid day 7 group, suggesting that the injury
and inflammation that occurred during the early phase in the alymphoid
group had subsided by day 7. In the transplantation model, the
recipient retains its unmanipulated, native heart in addition to the
heterologous transplanted heart; thus, the native heart provides a
control for systemic perioperative stress and nonspecific inflammatory
responses. Our analysis shows minimal differences between control
untransplanted and native hearts from the allogeneic-BALB/cB6 group
on day 1 or 7, indicating that our panel of genes was not up-regulated
in response to systemic stress or nonspecific injury. Similar analyses
did not detect differences in the native hearts from the syngeneic and
alymphoid groups (unpublished results). Cluster analysis of the
allogeneic and syngeneic groups showed similar results (syngeneic
results not shown).
 View larger version (20K): [in a new window] |
Figure 2. Dendrogram of rejection profile in allograft recipients treated with a
costimulatory blockade of CD40L. The level of expression of a panel of
60 candidate genes was determined by RPA and analyzed by Cluster and
Tree software. (a) Includes graft hearts from allogeneic and alymphoid
and native hearts from allogeneic recipients harvested at 1 and 7 days
following transplantation; (b) includes recipients treated with MR1 mAb
to block costimulation at days 0 and 1, harvested at days 1 and 7 and
time of rejection (two representative recipients with graft rejection
on days 33 and 32 are shown), and allogeneic and alymphoid recipients
harvested at days 1 and 7. Controls are untransplanted BALB/c and B6
hearts. The degree of dissimilarity is proportional to the total length
of the horizontal axis among groups.
|
Coordinate expression of genes analyzed by dendrograms and SOM
In addition to our comparison of experimental groups, we also
generated dendrograms of the candidate genes (Fig. 3A
, left). A visual inspection of the color-coded expression levels
shows subsets of genes with related expression patterns. For example,
there is a subset with an increased pattern of expression in the
MR1-treated grafts at day 32 at the time of rejection (Group A) and a
second subset with increased expression in the untreated allogeneic
group (Group B; Fig. 3a
, right). These differential expression profiles
form the basis for the dendrogram of the experimental groups (Fig. 3a ,
top). We also analyzed gene expression by ANOVA for two variables
(Table 2
). These results show decreased expression of multiple genes during
MR1 treatment. For example, the chemokines MIP-1, MIP-1ß, IP-10,
and MCP-1 are all significantly decreased by MR1 treatment. ANOVA also
detected marked changes in expression in the MR1-treated recipients at
the time of rejection (day 32–33); for example, IFN-, LTß,
TNF-, and MIP-2 were all significantly decreased. Analysis of the
syngeneic and alymphoid recipients was also similar, because we did not
detect significant changes in gene expression from native hearts of the
syngeneic, alymphoid, or allogeneic recipients (not shown).
 View larger version (28K): [in a new window] |
Figure 3. Dendrogram of expression levels of candidate genes. The level of
expression of a panel of 60 candidate genes was analyzed by RPA. (a) A
dendrogram was generated by Cluster analysis comparing the expression
profiles of individual genes (left). Group A (increased expression in
two representative recipients treated with MR1 mAb with graft rejection
on days 33 and 32) and Group B (increased expression at the time of
rejection in the allogeneic group) are expanded with individual genes
listed (right), and the corresponding dendrogram of the experimental
groups was generated (top). Pseudocolor key of gene expression: blue
(no increase); yellow (color intensity correlates with level of
increased expression). (b) The same panel of genes was analyzed with
GeneCluster [16
] using SOM and separated into nine
distinct maps, which are shown in boxes and numbered as Maps 0–8. In
each box, the open circles show the mean level of gene expression, and
the flanking lines indicate the standard deviations of the expression
for each map. For each map, the circles from left to right represent
expression for genes from BALB/c control (gray), B6 control (white),
allogeneic day 1 (maroon), allogeneic day 7 (yellow), syngeneic day 1
(purple), syngeneic day 7 (green), alymphoid day 1 (red), alymphoid day
7 (black), MR1-treated day 1 (chartreuse), MR1-treated day 7 (blue),
MR1-treated day 33 (orange), and MR1-treated day 32 (turquoise),
respectively. The y-axis represents relative expression
calculated for each SOM. Individual maps from 0 to 8 include a total of
23, 8, 7, 4, 2, 2, 7, 3, and 4 genes, respectively (see Table 3
). The
number of maps was determined empirically to minimize the standard
deviations and maximize the number of genes per
map.
|
|
View this table: [in a new window] |
Table 2. Analysis of Variance of Gene Expression
|
Although the grafts in the MR1-treated and untreated groups are
ultimately rejected (but with different kinetics), an analysis of the
composition of the SOM suggests distinct mechanisms of rejection
(Table 3
). For example, genes that are highly expressed by T cells
including TCR-, CD3, CD8, and CD8ß are highly up-regulated
in the untreated allogeneic group at day 7 but are only modestly
up-regulated in the MR1-treated groups at day 7 or day 33 at the time
of rejection (see Map 6). In contrast, genes highly up-regulated in the
MR1-treated group at the time of rejection but only modestly
up-regulated in the other groups include F4/80, CCR1, CCR2, CCR5,
MCP-1, TGF-ß1, and TGF-ß3.
|
View this table: [in a new window] |
Table 3. Clusters of Genes Selected by Self-organizing Maps
|
, MIP-2, CD8,
IP-10, and RANTES from Maps 0–8, respectively. Using this panel of
nine genes selected by a heuristic process for maximal proximity to
centroid values, cluster analysis produced a dendrogram that
differentiated the rejecting and nonrejecting groups with a similar
pattern of hierarchical clustering as the complete panel of 60 genes
(Fig. 4
). These results demonstrate the power of SOM to segregate groups
of genes with similar patterns of expression into maps that can be used
to select small numbers or even a single gene from each subset to
provide diagnostic power.
 View larger version (24K): [in a new window] |
Figure 4. Dendrogram of rejection profile in allograft recipients treated with
costimulatory blockade of CD40L based on a single gene selected from
each SOM. A single gene was selected from each of the nine subsets of
genes defined by the SOM from Table 3
based on proximity to the mean
value of expression. The nine selected genes were then analyzed by
Cluster and Tree software to produce the dendrogram of gene expression
from graft hearts of MR1-treated, allogeneic, and alymphoid recipients
at days 1 and 7 and MR1-treated samples at the time of rejection (days
33 and 32). Controls are untransplanted BALB/c and B6 hearts. The
degree of dissimilarity is proportional to the total length of the
horizontal axis among groups.
|
 View larger version (158K): [in a new window] |
Figure 5. Immunohistochemistry of CD4 and CD8. Expression was visualized by
indirect immunohistochemistry of CD4 in untransplanted control (a),
allogeneic graft heart day 7 (b), MR1-treated graft heart day 7 (c),
and MR1-treated graft heart day 33 (d), and of CD8 in untransplanted
BALB/c control (e), allogeneic graft heart day 7 (f), MR1-treated graft
heart day 7 (g), and MR1-treated graft heart day 33 (h).
|
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
In contrast, acute rejection in the untreated grafts showed increased
levels of T-cell markers including TCR and CD3. Because CD8
and CD8ß, but not CD4, were up-regulated, the cells are most likely
CD8 T cells, which was confirmed by immunohistochemistry. In addition,
several chemokines including RANTES, IP-10, and Ltn were up-regulated
in the untreated, acutely rejecting grafts. Thus, our study formulates
an approach using clustering algorithms to analyze a panel of candidate
genes to determine the molecular basis of the allogeneic immune
response in an in vivo model of vascularized, solid organ
transplantation.
Currently, a clinical diagnosis of allograft rejection is often based on histological analysis of biopsy specimens. For example, pathologic evaluation of kidney-allograft rejection is based on standardized Banff criteria that score intimal arteritis and tubulitis [22 ]. However, despite standardized, quantitative criteria of acute rejection in clinical biopsies, it is apparent that there are different mechanisms of rejection based on different clinical courses and varying responses to antirejection therapy that are not differentiated by histology. Some studies using polymerase chain reaction technologies have analyzed small subsets of genes that are up-regulated in rejecting allografts [23 , 24 ], suggesting the possibility of identifying a molecular profile of rejection. Although the regulation of protein levels can differ from mRNA expression [25 ], previous studies have shown that the analysis of the level of expression of a panel of mRNAs can be adequate to differentiate biological processes [26 27 28 29 ]. Our study suggests that future clinical studies may be able to define clinical states, such as acute rejection, based on small panels of up-regulated genes, which include a representative gene from each subset determined by SOM.
Dendrograms analyzing the kinetics of rejection in the untreated allogeneic recipients clearly segregated the response into early (days 1 and 3) and late (days 5 and 7) phases. One subset of genes was up-regulated only during the late phase and only in rejecting grafts (not in the syngeneic or alymphoid controls). In addition to markers of CD8 T cells, the most highly up-regulated gene in this subset was RANTES. Thus, robust up-regulation of genes in this map was only observed in the presence of an allogeneic-adaptive immune response. In contrast, most genes up-regulated during the early phase had increased expression in the syngeneic and alymphoid, as well as the allogeneic, recipients. Thus, up-regulation of genes in this map occurred independently of adaptive immunity, presumably triggered by injury or innate immunity. Our study did not identify qualitative differences in gene expression between the syngeneic and alymphoid recipients, indicating that NK cells were not required for the induction of the genes in our panel. This does not exclude the possibility that NK cells were activated in some experiments or that other genes not analyzed in our study were up-regulated by NK cells. Nevertheless, these results suggest that stimuli, such as ischemia/reperfusion or stress, were pivotal in up-regulating gene expression. The most highly up-regulated member of this map was MIP-2, a chemokine that has been shown previously to be up-regulated by ischemic injury [30 , 31 ].
Our results determined unique profiles of gene expression in different
phases and types of allograft rejection, suggesting that our analysis
could detect different molecular bases of rejection. By sorting the
patterns of gene expression into maps using SOM, our analysis
demonstrated that the different stages and types of rejection could be
differentiated with a subset of only nine genes (CD4, CCR3, CCR5,
LTß, MIP-1, MIP-2, CD8, IP-10, and RANTES). Our study focused
on mRNA expression using RPA as a result of the technical feasibility
of analyzing greater numbers of genes. Using immunohistochemistry, we
confirmed correlations between mRNA and protein expression of CD4 and
CD8. In addition, correlation between mRNA expression and function
(kinetics of graft rejection) suggests changes in expression of
functional proteins for at least a portion of our parameters. Future
studies using gene-chip technology could greatly expand the database of
gene expression and determine the number and specific subset of genes
necessary to optimize the diagnostic criteria. However, one of the
intriguing aspects of our study was the demonstration that a small
number of genes (only nine in our study), if appropriately selected by
SOM, were sufficient to distinguish different forms of rejection. The
ability to reduce the amount of data necessary to differentiate
rejection states using molecular profiles of rejection has important
practical implications for the diagnosis of graft rejection and the
development of individualized, therapeutic regimens.
Received June 18, 2001; revised September 21, 2001; accepted October 15, 2001.
|
TOPABSTRACTINTRODUCTIONMATERIALS AND METHODSRESULTSDISCUSSIONREFERENCES |
|---|
This article has been cited by other articles:
 |
|
 |
|
  B. P.-L. Lee, E. Mansfield, S.-C. Hsieh, T. Hernandez-Boussard, W. Chen, C. W. Thomson, M. S. Ford, S. E. Bosinger, S. Der, Z.-x. Zhang, et al. Expression Profiling of Murine Double-Negative Regulatory T Cells Suggest Mechanisms for Prolonged Cardiac Allograft Survival J. Immunol., April 15, 2005; 174(8): 4535 - 4544. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 V. J. Gimino, J. D. Lande, T. R. Berryman, R. A. King, and M. I. Hertz Gene Expression Profiling of Bronchoalveolar Lavage Cells in Acute Lung Rejection Am. J. Respir. Crit. Care Med., November 15, 2003; 168(10): 1237 - 1242. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
