Download Use of gene profiling to describe a niche for dendritic cell

Immunology and Cell Biology (2007) 85, 567–570
& 2007 Australasian Society for Immunology Inc. All rights reserved 0818-9641/07 $30.00
www.nature.com/icb
SHORT COMMUNICATION
Use of gene profiling to describe a niche for dendritic
cell development
Geneviève Despars1,3, Terence J O’Neill2 and Helen C O’Neill1
Gene profiling provides a multitude of data on individual gene expression. The view is expressed here that unreplicated data can
be used in a descriptive way to compare cell populations in terms of their lineage characteristics and function. In these studies,
the aim is to provide a snapshot of gene expression or its absence as a reflection of cell lineage or type, rather than gain a
reliable expression measure for all genes expressed. The data set used in this analysis represents gene expression in the splenic
stroma STX3 supportive of dendritic cell hematopoiesis and the lymph node stroma 2RL22, which is non-supportive. These were
obtained by hybridization of Affymetrix U74Av2 genechips. The use of P-value selection to identify genes with a high probability
of differential expression has been used effectively to detect differentially expressed genes. Genes that relate to a niche
environment for hematopoiesis have been selected for further study to make predictions about the cell types of supportive
stroma.
Immunology and Cell Biology (2007) 85, 567–570; doi:10.1038/sj.icb.7100080; published online 29 May 2007
The microenvironment required for development of dendritic cells
(DC) is still poorly understood, owing to a lack of appropriate in vitro
systems to study differentiation. Long-term cultures (LTC) of spleenproducing DC have been promising in this regard.1 These cultures
comprise an adherent stromal cell layer required for DC development
and a suspension fraction of progenitors and immature DC.2,3
Production of known growth factors regulating DC development
in vitro was found to be negative.1 Cell–cell contact is required
between DC progenitors and adherent stromal cells to drive DC
development, but no adhesion molecules have yet been identified.3,4
Very little is therefore known about the potential regulators expressed by
stromal cell components of spleen that support DC production in LTC.
The need to identify stromal regulators of DC hematopoiesis has
prompted an investigation of the genome-wide expression patterns of
two functionally different stroma. The STX3 splenic stroma and the
2RL22 lymph node stroma support and do not support, respectively,
DC development from bone-marrow-derived progenitors.5 A major
challenge of transcriptome analysis using microarrays is the analysis of
signal values and the retrieval of specific data sets. This paper details
the procedures for retrieval of probe sets identifying genes expressed
with high certainty in STX3 but not expressed in 2RL22.
RESULTS AND DISCUSSION
Transcriptome analysis on two functionally distinct stroma was
performed with the aim of identifying new regulators involved in
DC development in vitro. Since the aim was to ‘fish’ for genes
expressed with high certainty in STX3 but not expressed in 2RL22,
we performed computational analysis to identify these genes from
unreplicated Affymetrix genechip experiment. We compare here the
standard method of Affymetrix Microarray Suite 5.0 software package
(MAS 5.0) with another method based on P-value for retrieving
a short list of genes of interest. This computational subtraction
method could be relevant for gene expression studies that are not
easily replicated and that sometimes involve small numbers of stem
cells or progenitors. These studies usually involve difficult and costly
procedures for cell isolation.
Comparative transcriptome analysis is based on the hypothesis that
genes with similar expression patterns are related mechanistically, and
functionally associated with a given biological process. The STX3
splenic stroma is an in vitro niche model for early DC hematopoiesis.
To identify genes specifically expressed in STX3, computational
subtractions were performed between 2RL22 and STX3. This differential analysis will allow identification of genes potentially involved in
early DC hematopoiesis. Analysis of signal value distribution revealed
that the vast majority of genes are commonly expressed at similar
levels in both stroma. The common data set of 2RL22 and STX3
probably reflects housekeeping genes related to cell metabolism and
division. The general aim of the Affymetrix genechip experiment was
to retrieve a list of genes with high confidence in their specific
expression in STX3. Detection calls ascribed by MAS 5.0 were first
used to retrieve specific data sets. Detection calls correspond to
‘present’ for detection P-value o0.04, ‘marginal’ for detection
1School of Biochemistry and Molecular Biology, The Australian National University, Canberra, Australia and 2School of Finance and Applied Statistics, The Australian National
University, Canberra, Australia
3Current address: Institut de Génomique Fonctionnelle de Lyon, IFR128 Gerland Lyon Sud, Université Lyon 1, CNRS, INRA, Ecole Normale Supérieure de Lyon, France.
Correspondence: Professor HC O’Neill, School of Biochemistry and Molecular Biology, The Australian National University, Bldg 41 Linnaeus Way, Canberra, ACT 0200, Australia.
E-mail: [email protected]
Received 15 February 2007; revised 29 April 2007; accepted 4 May 2007; published online 29 May 2007
Gene profiling to describe a niche for DC development
G Despars et al
568
Figure 1 Comparison of specific data sets retrieved on detection calls and
detection P-values. (a) Signal plot of specific data sets retrieved according to
indicated selection criterion. For detection call, the 2RL22-specific data set
was selected by 2RL22: presence and STX3: absence or marginal. The
STX3-specific data set included probe sets with both STX3: presence and
2RL22: absence or marginal. Retrieval on stringent detection P-value was
based on detection P-value p0.005 for presence and detection P-value
X0.1 for absence. Corresponding probe sets were plotted for 2RL22 signal
value against STX3 signal value. Contours represent 1, 5, 10, 15, 20 and
25 probe set limits for the left panel and 1, 2, 3, 4 and 5 probe set limits
for the right panel. The linear regression of 1 is shown. (b) Detection P-value
plot of specific data sets retrieved on detection call and detection P-value.
The standard linear regression of 0.04 is shown.
P-value between 0.04 and 0.06 and ‘absent’ for detection P-value
40.06. This method of retrieval gives poor discrimination of specific
expression, displaying overlap between specific data sets and many
probe sets on the linear regression of 1 (Figure 1a).
In MAS 5.0, the detection P-value is calculated with the one-sided
Wilcoxon Signed Rank Test and ascribes discrete values. For a given
probe set, a low STX3 detection P-value is associated with a high
certainty of signal value. This approach limits the number of false
positives, also described as the Bonferroni Correction. A 2RL22
detection P-value greater than the MAS 5.0 default value of 0.06 for
absence should remove false negatives, for example probe sets
considered as absent but which are not. To identify probe sets specifically expressed by STX3 but not 2RL22, several detection P-values
below 0.04 were set for STX3, along with a constant 2RL22 detection
P-value above 0.1. The same approach was taken to retrieve a 2RL22specific data set. Specific data sets retrieved on stringent detection
P-values p0.005 for presence and detection P-value X0.1 showed no
overlap in terms of signal value (Figure 1a). Probe sets of these specific
data sets were annotated and clustered into functional categories. The
pattern of functional categorization was different between STX3 and
2RL22 data sets, suggesting that selection on detection P-value retrieves
genes related to distinct biological functions (data not shown).
Immunology and Cell Biology
STX3-specific data sets retrieved on standard and stringent detection
P-values were further analyzed in terms of detection P-values and signal
(Figure 1b). The nature of the one-sided Wilcoxon Signed Rank Test
was revealed by alignment of discrete values rather than random
dispersion. A data set of 673 probe sets was extracted using standard
detection P-value, which is relatively large and complicates the choice of
potentially interesting genes. Discrimination between STX3 and 2RL22
signal values was also poor, with a mean signal value of 113.3±10.0 for
STX3 and 36.3±2.1 for 2RL22. The size of data sets decreased as the
STX3 detection P-value decreased from 0.03 to 0.005, ranging from 317
to 165 probe sets. Analysis of signal values indicated good discrimination between STX3 and 2RL22 with the mean STX3 signal being
B7-fold to B15-fold higher than the mean 2RL22 signal. These were
significantly different (Z-test). Selection on the basis of stringent
detection P-value also generated data sets with higher STX3 signal
values compared with selection based on detection calls. Selection with
stringent detection P-value can efficiently discriminate specifically
expressed probe sets and enrich for highly expressed probe sets.
Since specific gene expression was thought to be related to the DC
supportive function of STX3, we were interested in identifying genes
that are potential regulators of the microenvironment for DC development. A DC niche could be regulated by secreted factors as well
as cell surface molecules. Probe sets clustered under chemokines/
cytokines, growth factors, matrix remodeling, extracellular matrix/cell
adhesion, surface proteins, receptors, signaling molecules, transcriptional regulation and development were considered to be of interest.
Selection on the basis of detection P-value resulted in enrichment of
functionally relevant categories. These categories altogether represent
56% of probe sets of the specific data set retrieved on detection call
(STX3 P-value o0.04; 2RL22 P-value 40.04) compared with 66.9% of
the STX3-specific data set retrieved on the basis of detection P-value
(STX3 P-value p0.005 and 2RL22X0.1). This was reflected by decreasing numbers of probe sets related to metabolism across data sets, rather
than by an increase in the absolute number of probe sets within relevant
categories (data not shown). Analysis of genes in the categories
‘receptor’ and ‘extracellular matrix’ showed that selection on detection
P-value led to removal of probe sets with an STX3:2RL22 signal ratio of
B1, as well as with a higher mean signal value (Figure 2). These results
indicate that selection on detection P-value eliminated genes related to
common metabolic functions, as well as genes of little interest in terms
of signal discrimination between STX3 and 2RL22 within the categories
of interest. Expression of a number of genes in STX3 but not 2RL22 was
confirmed using RT-PCR.6 In the ‘receptor’ category, these genes
included Acvrl1, Ms4s4d and Thfrsf9, and in the ‘extracelluular matrix
category’, these included Col18a1, Mcam and Cd34. With the expression
also of Flt1 but not Cd31 or Vwf (von Willebrand factor), STX3 appears
to represent an immature endothelial cell. Altogether, retrieval of genes
on the basis of detection P-value was successful in identifying the
phenotype of STX3 stromal cells that provide hematopoietic support
function for DC development. Rather than a comprehensive map of
gene expression, this method provided a biological snapshot of gene
expression in STX3. In contrast, 2RL22 showed expression of genes for
extracellular matrix proteins (Col1a1, Col2a1, Col3a1, Col5a1, Col5a2,
P4ha1 and P4ha2) expressed by fibroblasts.
This report compares various approaches for extraction of differentially expressed genes from unreplicated Affymetrix data sets. Our
aim was to detect a subset of genes with high certainty of differential
expression rather than to detect all the genes that are potentially
differentially expressed. The method for extraction of data was based
on P-value selection to remove data outside the bounds of statistical
significance. This procedure was compared with the default detection
Gene profiling to describe a niche for DC development
G Despars et al
569
METHODS
Cell lines
Derivation of the splenic stroma STX3 and the lymph node stroma 2RL22 has
been described previously.1–3 Both were derived from LTC established from
B10.A(2R) mice. The two stroma differ in cell morphology. STX3 comprises
a mix of endothelial cells and fibroblasts, while 2RL22 contains mainly
fibroblast-like cells.5 Stromal cells were cultured as described previously6 and
maintained by scraping attached cells for passage into a new flask.
Microarray analysis of gene expression
Total RNA extraction was performed using Trizol (Invitrogen Life Technologies,
Mount Waverley, VIC, Australia). Synthesis of cDNA involved the use of
T7-(dT)24 primers and SuperScript II according to the manufacturer’s instructions (Invitrogen Life Technologies). This was followed by second strand
synthesis with DNA polymerase 1 (Promega, Annandale, NSW, Australia).
In vitro transcription and biotin labeling were performed by Dr Kaiman Peng
(Biomolecular Resources Facility, Australian National University) using the
BioArray High-Yield RNA Transcript Labeling Kit (Affymetrix, Santa Clara,
CA, USA). Labeled cRNA was fragmented and hybridized to Test 3 chips
(Affymetrix), before hybridization to murine genome U74Av2 Genechips
(Affymetrix). Hybridization involved 0.05 mg/ml biotin-labelled cRNA in hybridization buffer (100 mM MES, 1 N [Na+], 20 mM EDTA, 0.01% Tween 20)
supplemented with 0.1 mg/ml herring sperm DNA and 0.5 mg/ml acetylated
bovine serum albumin for 16 h. Washing and staining with streptavidinphycoerythrin were performed on the fluidics station according to the
manufacturer’s instructions (Affymetrix). The quality of the two arrays was
confirmed on the basis of low background (STX3: 55.76; 2RL22: 58.56), high
overall % genes expressed indicative of high-quality RNA (STX3: 52.8%;
2RL22: 50.5%), low noise (RawQ) (STX3: 2.300; 2RL22: 2.210) and signal
intensity ratios of 3¢/5¢ probe sets for housekeeping genes B1.0 (b-actin:
STX3¼1.08, 2RL22¼1.29; GAPDH: STX3¼1.02, 2RL22¼1.01).
Data mining for retrieval of specific data sets
Figure 2 Comparison of the categories ‘receptor’ and ‘extracellular matrix’
for STX3-specific data sets. Dot plots of the receptor (a) and extracellular
matrix (b) categories of STX3-specific data sets retrieved on detection call
and detection P-value. The linear regression of 1 is shown.
call associated with the commonly used software package MAS 5.0.
Selection on the basis of P-value was retained as the most discriminating method for retrieving probe sets specifically expressed by STX3 but
not 2RL22. This method was considered the most appropriate, given
the size of the data set and expression levels. Using P-value selection,
probe sets were retrieved with higher expression levels in STX3
compared with 2RL22, indicated by differences in the mean and
median signal values. Several of these genes were further verified by
RT-PCR, which confirmed differential gene expression between
STX3 and 2RL22.6 Furthermore, Affymetrix genechip results have
been shown to correlate well with quantitative real-time PCR data.7
Specifically expressed genes are expressed at low levels and represent a
relatively small proportion of genes expressed by each stroma. This
suggests that a small set of genes expressed at low level may be
responsible for cell specificity. Hence, definition of the specific
microenvironment for early DC hematopoiesis could depend on
expression of a limited number of relevant genes.
Other computational approaches have involved lower confidence
bound calculation and P-value. However, the lower confidence bound
calculation depends on a large number of arrays.8 The statistical
significance (P-value) of gene expression has also been described for
microarray experiments that have a series of samples.9 Most importantly, the P-value selection method used here can be readily applied
to non-replicated data sets derived from limited numbers of rare
cells like stem cells.
Scanned images of murine genome U74Av2 genechips (Affymetrix) hybridized
with labelled cRNA prepared from STX3 or 2RL22 RNA were processed using
Affymetrix Microarray Suite 5.0 software (MAS 5.0). This analysis generated a
text file of probe set entries, P-values and signal values for STX3 and 2RL22, as
well as partial annotation of probe sets representing individual genes. In MAS
5.0, the P-value is calculated using the distribution of the test statistic, in this
case, using one-sided Wilcoxon Signed Rank Test to calculate the detection
P-value. The P-value is the probability under the null hypothesis that the test
statistic is as extreme as the observed value. Microsoft Excel or scripts written
using interactive data language (IDL) software (http://www.ittvis.com) were
used to calculate mean and median signal values for data sets. Data sets were
made for probe sets according to a range of different selection criteria. Probe
sets within each data set were manually annotated with gene name and the
Gene Ontology categories of ‘Biological Process’, ‘Cellular Component’ and
‘Molecular Function’, available on the websites of Mouse Genome Informatics
(www.informatics.jax.org), Affymetrix database (www.affymetrix.com) and
NCBI database (http://www.ncbi.nlm.nih.gov). The Conserved Domain database10 was used in a few cases to predict gene function. Statistical analysis
was performed on signal values of data sets using the Z-test, chosen because
of the skewed distribution of signal values.
ACKNOWLEDGEMENTS
This work was supported by funding from the Australian National University
to HO and TO. GD was supported by a PhD scholarship from the Fonds de la
Recherche en Santé du Québec.
1 Ni K, O’Neill HC. Long-term stromal cultures produce dendritic-like cells.
Br J Haematol 1997; 97: 710–725.
2 Ni K, O’Neill HC. Spleen stromal cells support haemapoiesis and in vitro growth of
dendritic cells from bone marrow. Br J Haematol 1999; 105: 58–67.
Immunology and Cell Biology
Gene profiling to describe a niche for DC development
G Despars et al
570
3 Wilson HL, Ni K, O’Neill HC. Identification of progenitor cells in long-term spleen
stromal cultures that produce immature dendritic cells. Proc Natl Acad Sci USA 2000;
97: 4784–4789.
4 Wilson HL, Ni K, O’Neill HC. Proliferation of dendritic cell progenitors in long
term culture is not dependent on granulocyte macrophage-colony stimulating factor.
Exp Hematol 2000; 28: 193–202.
5 Ni K, O’Neill HC. Hemopoiesis in long-term stroma-dependent cultures from lymphoid
tissues: production of cells with myeloid/dendritic characteristics. In vitro Cell Dev Biol
Anim 1998; 34: 298–307.
6 Despars G, Ni K, Bouchard A, O’Neill TJ, O’Neill HC. Molecular definition of an in vitro
niche for dendritic cell development. Exp Hematol 2004; 32: 1182–1193.
Immunology and Cell Biology
7 de Reyniès A, Geromin D, Cayuela JM, Petel F, Dessen P, Sigaux F et al. Comparison of
the latest commercial short and long oligonucleotide microarray technologies. BMC
Genomics 2006; 15: 51.
8 Li C, Wong WH. Model-based analysis of oligonucleotide arrays : model validation,
design issues and standard error application. Genome Biol 2001; 2, research
0032.1–0032.11.
9 Sasik R, Calvo E, Corbeil J. Statistical analysis of high-density oligonucleotide arrays:
a multiplicative noise model. Bioinformatics 2002; 18: 1633–1640.
10 Marchler-Bauer A, Anderson JB, DeWeese-Scott C, Fedorova ND, Geer LY, He S et al.
CDD: a curated Entrez database of conserved domain alignments. Nucleic Acids Res
2003; 31: 383–387.