Spatial patterns of gene expression in the olfactory bulb.
ABSTRACT How olfactory sensory neurons converge on spatially invariant glomeruli in the olfactory bulb is largely unknown. In one model, olfactory sensory neurons interact with spatially restricted guidance cues in the bulb that orient and guide them to their target. Identifying differentially expressed molecules in the olfactory bulb has been extremely difficult, however, hindering a molecular analysis of convergence. Here, we describe several such genes that have been identified in a screen that compiled microarray data to create a three-dimensional model of gene expression within the mouse olfactory bulb. The expression patterns of these identified genes form the basis of a nascent spatial map of differential gene expression in the bulb.
- [Show abstract] [Hide abstract]
ABSTRACT: Neural crest development is orchestrated by a complex and still poorly understood gene regulatory network. Premigratory neural crest is induced at the lateral border of the neural plate by the combined action of signaling molecules and transcription factors such as AP2, Gbx2, Pax3 and Zic1. Among them, Pax3 and Zic1 are both necessary and sufficient to trigger a complete neural crest developmental program. However, their gene targets in the neural crest regulatory network remain unknown. Here, through a transcriptome analysis of frog microdissected neural border, we identified an extended gene signature for the premigratory neural crest, and we defined novel potential members of the regulatory network. This signature includes 34 novel genes, as well as 44 known genes expressed at the neural border. Using another microarray analysis which combined Pax3 and Zic1 gain-of-function and protein translation blockade, we uncovered 25 Pax3 and Zic1 direct targets within this signature. We demonstrated that the neural border specifiers Pax3 and Zic1 are direct upstream regulators of neural crest specifiers Snail1/2, Foxd3, Twist1, and Tfap2b. In addition, they may modulate the transcriptional output of multiple signaling pathways involved in neural crest development (Wnt, Retinoic Acid) through the induction of key pathway regulators (Axin2 and Cyp26c1). We also found that Pax3 could maintain its own expression through a positive autoregulatory feedback loop. These hierarchical inductions, feedback loops, and pathway modulations provide novel tools to understand the neural crest induction network.Developmental Biology 12/2013; · 3.87 Impact Factor
- [Show abstract] [Hide abstract]
ABSTRACT: The human olfactory bulb displays high morphologic dynamics changing its volume with olfactory function, which has been explained by active neurogenetic processes. Discussion continues whether the human olfactory bulb hosts a continuous turnover of neurons. We analyzed the transcriptome via RNA quantification of adult human olfactory bulbs and intersected the set of expressed transcriptomic genes with independently available proteomic expression data. To obtain a functional genomic perspective, this intersection was analyzed for higher-level organization of gene products into biological pathways established in the gene ontology database. We report that a fifth of genes expressed in adult human olfactory bulbs serve functions of nervous system or neuron development, half of them functionally converging to axonogenesis but no other non-neurogenetic biological processes. Other genes were expectedly involved in signal transmission and response to chemical stimuli. This provides a novel, functional genomics perspective supporting the existence of neurogenesis in the adult human olfactory bulb.Brain Structure and Function 08/2013; · 7.84 Impact Factor
- [Show abstract] [Hide abstract]
ABSTRACT: A challenge in gene expression studies is the reliable identification of differentially expressed genes. In many high-throughput studies, genes are accepted as differentially expressed only if they satisfy simultaneously a p value criterion and a fold change criterion. A statistical method, TREAT, has been developed for microarray data to assess formally if fold changes are significantly higher than a predefined threshold. We have recently applied the NanoString digital platform to study expression of mouse odorant receptor genes, which form with 1,200 members the largest gene family in the mouse genome. Our objectives are, on these data, to decrease false discoveries when formally assessing the genes relative to a fold change threshold, and to provide a guided selection in the choice of this threshold. Statistical tests have been developed for microarray data to identify genes that are differentially expressed relative to a fold change threshold. Here we report that another approach, which we refer to as tTREAT, is more appropriate for our NanoString data, where false discoveries lead to costly and time-consuming follow-up experiments. Methods that we refer to as tTREAT2 and the running fold change model improve the performance of the statistical tests by protecting or selecting the fold change threshold more objectively. We show the benefits on simulated and real data. Gene-wise statistical analyses of gene expression data, for which the significance relative to a fold change threshold is important, give reproducible and reliable results on NanoString data of mouse odorant receptor genes. Because it can be difficult to set in advance a fold change threshold that is meaningful for the available data, we developed methods that enable a better choice (thus reducing false discoveries and/or missed genes) or avoid this choice altogether. This set of tools may be useful for the analysis of other types of gene expression data.BMC Bioinformatics 02/2014; 15(1):39. · 3.02 Impact Factor
Spatial patterns of gene expression in the
David M. Lin*†‡, Yee Hwa Yang‡§, Jonathan A. Scolnick*, Lisa J. Brunet*, Heather Marsh†, Vivian Peng*,
Yasushi Okazaki¶, Yoshihide Hayashizaki¶, Terence P. Speed§?, and John Ngai*,**
*Department of Molecular and Cell Biology, Functional Genomics Laboratory, Helen Wills Neuroscience Institute, and§Department of Statistics, University
of California, Berkeley, CA 94720;¶RIKEN Genomic Sciences Center, Genome Exploration Research Group, Yokohama, Kanagawa 230-0045, Japan;?Division
of Genetics and Bioinformatics, The Walter and Eliza Hall Institute, Melbourne, Victoria 3050, Australia; and†Department of Biomedical Sciences, Cornell
University, Ithaca, NY 14853
Communicated by Richard Axel, Columbia University, New York, NY, July 9, 2004 (received for review November 15, 2003)
How olfactory sensory neurons converge on spatially invariant
glomeruli in the olfactory bulb is largely unknown. In one model,
olfactory sensory neurons interact with spatially restricted guid-
ance cues in the bulb that orient and guide them to their target.
Identifying differentially expressed molecules in the olfactory bulb
has been extremely difficult, however, hindering a molecular
analysis of convergence. Here, we describe several such genes that
have been identified in a screen that compiled microarray data to
create a three-dimensional model of gene expression within the
mouse olfactory bulb. The expression patterns of these identified
genes form the basis of a nascent spatial map of differential gene
expression in the bulb.
neurons (OSNs) to specific targets in the olfactory bulb. OSNs
expressing a given odorant receptor are distributed throughout
large zones in the olfactory epithelium but ultimately converge
on common targets in the bulb, which are called glomeruli.
Remarkably, these glomerular positions are spatially invariant
from animal to animal, consistent with the presence of a
hard-wired map of connectivity in the peripheral olfactory
How are OSNs able to identify their glomerular targets? The
stereotypic nature of this innervation suggests that regions of the
bulb, perhaps the glomeruli, are molecularly distinct. These
spatially distributed cues would then function to guide OSNs
expressing complementary receptors. In this model, the collec-
tive expression patterns of these guidance cues would form a
spatial map within the bulb, molecularly distinguishing potential
paths and substrates encountered by OSNs. However, although
it has become increasingly clear that a number of molecules have
been identified that are differentially expressed by OSNs [for
e.g., along the path that OSNs traverse to the bulb (2–4) and by
ensheathing cells (5–7)], efforts to identify spatially restricted
molecules within the bulb itself have been generally unsuccessful
(8, 9), greatly impeding our understanding of how olfactory
axons select their partner glomeruli. Indeed, the most plausible
models of olfactory development assume that spatially restricted
cues exist within the olfactory bulb to guide the sensory axons to
their targets (1, 10). Yet, despite this widely held assumption, it
has been remarkably difficult to identify such guidance and
Does a spatial map of differential gene expression exist within
the bulb? We decided to address this issue directly by employing
a molecular screen using microarrays to reconstruct, in three
dimensions, patterns of gene expression within the mouse ol-
factory bulb. This statistical representation of the bulb was then
scanned for genes with differential expression patterns, which
were validated by RNA in situ hybridization. A number of genes
known to be involved in neural development and pattern for-
mation were shown to be expressed in restricted patterns,
consistent with their possible roles in imparting spatial informa-
tion within the developing olfactory bulb.
n the olfactory system, a major challenge has been to identify
the mechanisms responsible for guiding olfactory sensory
Materials and Methods
RNA Purification and Amplification. Olfactory bulbs were dissected
from postnatal day 0 CD-1 mice, and slices were obtained by
manual dissection, which was done approximately in thirds along
each principal axis. For each aspect [anterior (A), dorsal (D),
lateral (L), medial (M), posterior (P), and ventral (V)], samples
from ?10 mice were pooled and total RNA was isolated by using
the TRIzol reagent. We amplified 3 ?g of RNA once by using
an optimized T7 RNA polymerase amplification protocol
(D.M.L., P. Luu, T. Serafini, and J.N., unpublished data),
incorporating an oligo(dT)-T7 primer (5?-ATCGATTCGA-
ATAGGGAGACCACAT21-3?) to generate cDNA. Amplified
RNA was then obtained in an in vitro transcription reaction (1?
transcription buffer?2 mM each NTP?1 unit of RNAsin?100
units of T7 RNA polymerase), and 5 ?g of amplified RNA was
used per labeling reaction.
Microarrays. We generated microarrays by using the RIKEN 19K
full-length mouse cDNA set (11). Microarrays were hybridized
with Cy3- and Cy5-labeled cDNAs by using protocols essentially
as described by the Microarrays.org public protocols distributor
Normalization and MA Plot. Raw data from each scanned slide
were processed by SPOT (12), with foreground seeds set at a 5 ?
5 pixels square. Within- and between-slide normalization were
performed on all data (13). An MA plot (14) was used to
represent the data, where M ? log2(red intensity?green inten-
sity) and A ? log2?R?G.
Estimation of Contrasts. For each gene, a linear model was used to
estimate each of the 15 contrasts [AL, DM, VP, etc.; for details,
published as supporting information on the PNAS web site]. For
visualization purposes, we estimated the contrast of a single
effect vs. an average of all six [e.g., a ˜ ? a ?1⁄6(a ? p ? d ? v
? m ? l)], in effect, recreating the pooled bulb reference (a ?
p ? d ? v ? m ? l) in silico. Note that this estimate does not
represent the absolute expression ‘‘profile’’ but rather the rela-
tive expression between a portion of the bulb to the pooled
Two-Stage Cluster Analysis. The top 100 genes from each contrast
were taken (a total of 614 unique genes), and hierarchical
clustering of these genes was performed by building a dendro-
Abbreviations: OSN, olfactory sensory neuron; A, anterior; D, dorsal; L, lateral; M, medial;
P, posterior; V, ventral; ptd2, prostaglandin D2 synthase; IGF-2, insulin-like growth factor
2; pcV?2, procollagen V?2.
‡D.M.L. and Y.H.Y. contributed equally to this work.
**To whom correspondence should be addressed. E-mail: firstname.lastname@example.org.
© 2004 by The National Academy of Sciences of the USA
August 24, 2004 ?
vol. 101 ?
gram based on a modified Mahalanobis distance and Ward
agglomeration (see Supporting Materials and Methods). The
metric used to measure the similarity between the bulb expres-
sion profile for gene x and gene y is defined as ?xy? (x ? y)?
(X?X)(x ? y), where X is the design matrix of our experiment. We
considered all 614 clusters that consist of more than two genes
in the tree, measured the heterogeneity h by calculating the 15
SDs across the cluster for each of the pairwise effects, and took
the largest. By choosing a score of s ? 0.45 and taking all
maximal disjoint clusters with h ? s, we obtained 16 clusters. For
each cluster, we ranked genes based on their Mahalanobis
distance between the bulb expression profile of the gene and the
origin, which is written as ?x? (x)?(X?X)(x). Genes with a high
?xscore were selected for further verification. To further un-
derstand the relationship between these clusters, we carried out
a second hierarchical clustering by using the average profiles of
the disjoint clusters that were obtained as described above. A
dendrogram based on correlation metric and Ward agglomera-
tion was then built. A complete list of the 614 genes can be found
in Table 1, which is published as supporting information on the
PNAS web site.
Spatial Map. Patterns obtained from in situ hybridization exper-
iments were traced and digitized by using the NEUROLUCIDA
program. Up to 18 sections, taken from the entire AP axis of the
bulb, were used per reconstruction. The location of NT3 ex-
pressing cells was estimated from individual slides.
In Situ Hybridization. In situ hybridizations were performed as
described in ref. 15 by using 20-?m-thick coronal fresh-frozen
sections of olfactory bulbs from postnatal day 0 mice. Expression
patterns of candidate genes were verified with digoxigenin
Given the historical difficulty in determining where within the
bulb differentially expressed molecules might exist, we decided
to sample cell types from throughout the bulb by comparing
relatively large domains (i.e., the A, P, D, V, M, and L aspects)
(Fig. 1a). Because the bulb is composed of a series of concentric
layers, these aspects will each contain all of the different cell
types and layers within this structure.
Gene expression within the olfactory bulb fragments was
assayed by using DNA microarrays, and spatial patterns were
identified from these data by using statistical approaches. Pre-
vious attempts that used microarrays to compare grossly differ-
ent regions of the brain against one another (e.g., bulb vs.
amygdala) identified a surprisingly limited number of gene-
expression differences (16, 17). By comparison, our intended
approach relies on recreating patterns of expression that exist
within a tissue of the brain, comparing highly similar subregions
that are composed of morphologically equivalent populations of
cells and layers.
possible comparisons among each of the six aspects (Fig. 1b).
Next, we considered how to integrate this information to deter-
mine the three-dimensional expression pattern for each gene on
the array. In essence, we wished to produce a ‘‘bulb profile’’ of
expression for each gene that could then be surveyed to identify
those patterns that indicate differential expression. To generate
three-dimensional profiles, data from multiple hybridizations
must be combined in a manner that maintains the spatial nature
of each aspect used in the hybridization. Therefore, we expanded
on a previous approach (14) and used fixed effects linear models
to analyze our data (see Materials and Methods and Supporting
Materials and Methods).
For each of the 15 possible pairwise comparisons (e.g., A vs.
L), robust multiple regression was used to combine all measure-
ments of gene expression (see Materials and Methods). Measure-
ments obtained from direct measurements (e.g., A3L) and
indirect measurements (e.g., A3D3L) were weighted and
combined to produce a contrast value for that comparison. Each
gene would, therefore, be associated with 15 contrast values.
Thus, a gene may possess an AL contrast estimate of zero
(indicating, on average, equal expression in a comparison of the
A and L aspects) but a positive PM contrast estimate, indicating
relatively higher P expression as compared with M expression.
Together, the 15 contrast values for each gene represent for each
gene a profile of expression within the bulb. Fig. 1c shows such
a bulb profile for gene no. 15,228, which corresponds to jagged,
a ligand for Notch (18).
For ease of visualization, we simplified these 15 contrast
profiles and reduced each to a six-contrast representation, in
which each effect (a, p, d, v, m, and l) is compared with a
computed in silico pooled reference to generate an average
estimate (a ˜, p ˜, d˜, v ˜, m ˜, and l˜) of overall expression (see Materials
and Methods and Fig. 2a). In this simplified six-way profile,
positive or negative values indicate relatively increased or de-
creased expression, respectively, for a given effect as compared
with the computed pooled reference. Thus, it is more apparent
that jagged is expressed relatively strongly in the D and P aspects
of the bulb and comparatively evenly expressed across the other
four aspects (Fig. 2 a and b). These relative increases in
expression do not preclude expression in other regions of the
bulb because our predictions depend on the aggregate total of
mRNA that is present within each fragment used in these
Clustering Bulb Profiles. Our linear model analysis produced
?19,000 bulb profiles, with each profile comprising 15 contrasts
spatially restricted expression patterns, we considered different
approaches. One possibility would be simply to rank genes
according to their contrast values. A weakness of such an
approach, however, is that genes with certain profiles may
dominate other genes showing more subtle, but potentially
interesting, patterns. We, therefore, chose to cluster genes with
similar profiles to identify patterns that are prominently repre-
sented in our data.
In this approach, it is difficult to determine a priori how many
clusters (or ‘‘groups’’) should be represented within our data. To
dissected individually into three slices along each axis: AP (Left), DV (Middle),
and ML (Right). Slices shown in red represent the six aspects that were used in
this analysis. (b) All pairwise comparisons between the six aspects were made.
This design permits gene expression to be measured directly (e.g., AD) and
indirectly (APD, AVD, etc.). By convention, the sample to which the arrow
points was labeled with Cy5. (c) For each gene (e.g., jagged, gene no. 15,228),
15 contrasts (MD, LD, etc.) are estimated and assigned values (average, esti-
mated contrast value). By convention, the first letter in each contrast deter-
mines the sign of the y axis value (e.g., in the MD contrast for jagged, the
positive MD value indicates that M has relatively higher expression than D).
Lin et al.PNAS ?
August 24, 2004 ?
vol. 101 ?
no. 34 ?
produce an experimentally manageable number, we generated
16 clusters by using a two-stage hierarchical clustering procedure
(see Materials and Methods and Supporting Materials and Meth-
ods); the resulting dendrogram is shown in Fig. 2c, and the list
of 614 genes, sorted by cluster, is given in Table 1. It is important
to note that the number of clusters and the membership or rank
of an individual gene within a cluster can be changed without
affecting the profile of these genes. Thus, hierarchical clustering
is used here to organize the data and not necessarily to infer
mechanistic relationships between genes within a cluster.
The individual profiles of all genes contained within a cluster
were averaged to generate corresponding six-way cluster pro-
files, shown below the dendrogram as a red-and-green profile
map, where red, green, and black correspond to increased,
decreased, and unchanged expression, respectively, as compared
with the in silico pooled reference (Fig. 2d). We performed a
randomization analysis to test the null hypothesis that the data
are what we would expect from random data without structure).
The null hypothesis was convincingly rejected in an analysis
comprising 500 sets of randomized data, each containing all
?19,000 genes (P ? 0.01; see Supporting Materials and Methods).
An examination of the six-way cluster profiles reveals inter-
esting differences among the various patterns, which we orga-
nized into four general categories. The first category (category
1) includes groups whose six-way cluster profiles indicate dif-
ferences in expression along any one axis of the bulb (groups 2,
5, and 12). For example, the cluster profile of group 2 indicates
reduced expression in the A bulb and high relative expression in
the P bulb (low A, high P). This type of pattern, in which a bias
is observed from one pole of the bulb relative to the opposite
pole, is observed also in group 5 (low A, high P) and group 12
(low D, high V). Category 2 includes groups with cluster profiles
showing differential expression within a single aspect of the bulb
(groups 3, 9, and 14–16). For example, the cluster profile of
group 3 shows high relative expression in the P and D aspects
(high P and D) but minimal differences in the opposing A or V
aspects, respectively. This pattern holds as well for group 9 (low
A or M, no P or L changes), and groups 14–16 (all high M).
Category 3 includes most of the groups (groups 1, 4, 6–8, 11, and
13), and it comprises cluster profiles that show dramatic and
correlated changes along at least two axes of the bulb. For
example, in group 1, there is elevated expression in both the A
and P aspects (high A, high P) and reduced M and L expression
(low M, low L). In group 11, expression is high in both the D and
V aspects of the bulb (high D, high V) but is reduced in the M
and L bulb (low M, low L). Such patterns of high expression
along one axis but low expression along another axis suggest a
discontinuity of expression within the bulb consistent with
layer-specific, but not three-dimensionally restricted expression.
Finally, category 4 contains a single member (group 10), which
exhibits minimal differences between each aspect as compared
with the computed whole-bulb average.
Validation. The qualitative categorization of cluster profiles pro-
vided a preliminary filter for genes showing potentially inter-
esting expression patterns. Next, we used in situ hybridization to
validate differentially expressed genes within these clusters,
focusing on category 1 (groups 2, 5, and 12) and category 2
(groups 3, 14, and 15). Genes from categories 3 and 4 were not
considered further because the nature of their predicted profiles
the olfactory bulb. We disregarded group 16 (category 2)
because it is composed predominantly of hemoglobin genes,
which is likely an artifact of our dissection technique.
To select genes for in situ hybridizations, we typically assessed
the individual profiles of the 10 top-ranked genes from each
cluster and selected three to four genes showing the most
promising differences (i.e., of a magnitude or pattern likely to be
detected in an in situ hybridization). In some cases, however,
some genes (e.g., NT3) were chosen based on their known
function in the nervous system and not on their final ranking
within a group. We successfully identified nine genes that indeed
are differentially expressed within the olfactory bulb.
Category 1. Group 2. The cluster profile of group 2 (Fig. 2d), which
contains three genes (Table 1), suggests differential expression
along both the AP and DV axes. The individual profile (Fig. 3a)
of jagged indicates relatively high D and P expression as com-
pared with the remainder of the bulb. To determine the actual
pattern of expression, we hybridized digoxigenin-labeled jagged
probes to tissue sections of the olfactory bulb (Fig. 3b). As a
control, we used the glutamate receptor (subunit 1), which labels
cells in the periglomerular, mitral, and granule layers of the bulb
(Fig. 3 c and d). For jagged, we detected significant reactivity in
the accessory olfactory bulb, located on the DP surface of the
main olfactory bulb. Weaker signal could be detected in other
layers (data not shown). Thus, the in situ hybridization results for
jagged are fully consistent with the expression pattern predicted
from our three-dimensional statistical profile. Adhesion protein
RA175C, another member of group 2, is expressed in a similar
pattern, and is present in the accessory olfactory bulb and other
cell layers of the main olfactory bulb although at higher levels in
the mitral layer than jagged (see Figs. 5–9, which are published
as supporting information on the PNAS web site). Its pattern of
the top-ranked member of this group (an EST) displayed only
weak mitral expression and no obvious restriction in expression
(data not shown).
Group 5. Of the top three genes tested from group 5, two are
expressed in restricted patterns in the bulb. Prostaglandin D2
synthase (ptd2) is predicted to have relatively high P expression
and reduced L expression (Fig. 3e). In situ hybridization shows
expression of ptd2 in the leptomeninges of the bulb. This
can be simplified to produce an in silico estimate of gene expression along
each axis relative to the pooled reference. This permits visualization of the
higher D and P expression (positive average value) as compared with the
pooled reference. (b) The predicted jagged expression pattern can be super-
imposed on a model of the olfactory bulb, which is shown as a cube. On
average, higher D and P expression is predicted for this gene within the bulb.
(c) We generated 16 clusters, and the corresponding dendrogram shows the
relationship among the clusters. The individual bulb profiles of all genes
which is shown below the dendrogram. Red, green, and black indicate in-
creased, decreased, and unchanged relative expression, respectively, as com-
cluster 2 shown as both a red-and-green map and as a bar diagram. Compare
with the individual profile of jagged, which is shown in a.
Hierarchical clusting of bulb expression profiles. (a) The 15 contrasts
www.pnas.org?cgi?doi?10.1073?pnas.0404872101Lin et al.
expression is not uniform, however, and is absent from the AVL
Fig. 8), consistent with the prediction that was made from the
DNA microarray analysis. Insulin-like growth factor 2 (IGF-2)
exhibited a similar predicted pattern and was also validated by in
situ hybridization (Fig. 3 g and h; see also Fig. 9). The top-ranked
clone in this cluster (ZX00001L15) possessed strong mitral layer
expression but no obvious differential expression in the bulb
(data not shown).
Group 12. Four genes were tested from group 12. The top-ranked
genes in this cluster, clones ZA00006A12 (EST, similar to dynein
heavy chain) and ZX00032G15 (keratin complex 2, basic, gene 1),
showed weak or no signal in the bulb (data not shown). However,
the fourth- and fifth-ranked genes in this cluster, NAD(P)H
menadione oxidoreductase 1 (MO 1) and procollagen V?2
(pcV?2), were found to be expressed in similarly restricted
patterns in the bulb. For both of these genes, expression is biased
to the V portion of the bulb in a layer just external to the
glomeruli, corresponding possibly to a subset of periglomerular
cells (for pcV?2, see Fig. 3 i and j, and for MO 1, see Fig. 7 C and
six-way profiles, which predicted high V, low D expression.
In summary, of the 10 category 1 genes tested, 6 are spatially
restricted in expression in the bulb (jagged, RA175C, ptd2, IGF-2,
pcV?2, and MO 1), 2 are not restricted in their expression, and
2 gave a weak or not visible signal.
right and D is up. (a) Individual six-way profile for jagged. (b) In situ hybridization of jagged probe to a caudal olfactory bulb section. Note strong reactivity in
the accessory olfactory bulb (arrow) on the D–caudal surface of the main olfactory bulb. (c) Control hybridization with a glutamate receptor probe (GluR1) to
a caudal section. (d) Control hybridization of GluR1 probe to a more rostral bulb section. Strong reactivity is observed in the periglomerular (arrowhead), mitral
(white arrow), and granule (black arrow) cell layers. (e) Individual six-way profile for ptd2. (f) In situ hybridization using a ptd2 probe reveals a strong signal in
the leptomeninges (arrow). Note the absence of VL signal in this section (arrowheads). (g) Individual six-way profile for IGF-2. (h) In situ hybridization using an
(arrowheads). (i) Individual six-way profile for pcV?2.1. (j) Expression of pcV?2 is located immediately external to the glomerular layer (arrow). Note higher
expression on M and V aspects of the bulb as compared with the D and L aspects. (k) Individual six-way profile for cadherin-11. (l) Cadherin-11 probe labels cells
in the mitral layer on the M aspect of the olfactory bulb (arrows). Signal is also detected in the granule layer on the L aspect of the bulb (arrowheads). (m)
expression in a subset of mitral cells (arrows). Scale bar indicates 300 ?m.
In situ hybridization patterns of differentially expressed genes in the bulb. All tissue sections shown are coronal sections, oriented so that M is to the
Lin et al.PNAS ?
August 24, 2004 ?
vol. 101 ?
no. 34 ?
Category 2. We next tested genes belonging to clusters belonging
to category 2 (groups 3, 14, and 15). These groups exhibited
patterns predicting up-regulation in one aspect of the bulb but
no complementary differences along the same axis in the
opposite aspect (e.g., high P, unchanged A).
Group 3. We selected the first two clones in group 3, beta-site APP
cleaving enzyme, which gave no signal, and clone ZX00035C13
(an EST), which showed uniform expression in the mitral cell
layer (data not shown).
Group 14. In this cluster, neither ERK3 protein kinase nor clone
ZX00028A22 (an EST), the 3rd- and 4th-ranked genes, exhibit
restricted expression patterns. However, group 14 contains the
cadherin-11 gene (19), a member of the cadherin superfamily,
whose functions have been widely studied in the nervous system
(20). The bulb profile for cadherin-11, ranked 5th in this cluster,
predicts it to be relatively highly expressed along the AP axis,
weakly along the DV axis, and somewhat higher in the M aspect
cell layer of the bulb, particularly along the M aspect. Surpris-
ingly, a subset of granule neurons (or possibly cells belonging to
the A olfactory nucleus) along the L aspect of the bulb also
labeled with this probe (Fig. 3l; compare with Fig. 3c, which
shows the unrestricted labeling of all granule cells by using the
glutamate receptor probe). This restriction of cadherin-11 was
not predicted by its six-way bulb profile (Fig. 3k), probably
because the L bias within the granule cell layer is overshadowed
by a complementary M bias in the mitral cell layer. A similar
pattern of protein expression has been published for ephrin A2
(Elf-1 in Table 1), the top-ranked gene in this cluster (21);
however, we could not confirm this pattern by in situ hybridiza-
tion using a probe for ephrin A2.
Group 15. Five clones were tested from group 15. The first two,
ZX00001F10 (an EST) and ZX00032I10 (an EST) showed no
restriction, and the 5th-ranked gene, ZX00001F09 (EST) gave
no detectable signal. However, the 13th-ranked gene, vitronectin,
is expressed in the extraglomerular layer (similar to pcV?2 and
MO 1), with a three-dimensional restriction to the M and V
aspects of the bulb (Fig. 3n), consistent with the predicted
pattern (Fig. 3m). Cluster 15 also contains the neurotrophin-3
(NT3) gene, which is predicted to have relatively high M
expression and lower D and V expression (Fig. 3o). Previous
studies have been unable to determine where within the bulb
NT3 is expressed (22, 23). We were unable to detect any signal
by using digoxigenin-labeled probes. However, by using33P-
labeled probes, we detected hybridization within a subset of
mitral cells located along the M surface of the bulb (Fig. 3p;
compare with Fig. 3c). This band of NT3 positive cells extends
more ventrally within this layer at more rostral positions (see
In summary, of the 11 genes tested from category 2, 3 are
differentially expressed within the bulb (cadherin-11, vitronectin,
and NT3), 6 are uniformly expressed, and 2 gave no detectable
A Nascent Spatial Map of Gene Expression in the Olfactory Bulb. The
results of the in situ hybridization patterns paint a surprising and
unexpected picture of differential gene expression in the olfac-
torybulb. JaggedandAdhesionproteinRA175C arepreferentially
expressed in the accessory olfactory bulb; cadherin-11 labels a
subpopulation of granule neurons; ptd2 and IGF-2 are absent in
large areas of the leptomeninges; NT3 labels a small proportion
of previously undistinguished mitral cells; and vitronectin,
pcV?2, and MO 1 are expressed in the periglomerular region. To
obtain a more global view of these spatial distributions, we
reconstructed from multiple tissue sections the three-
dimensional expression patterns for a subset of genes. Major
olfactory bulb (Fig. 4). The resulting map reveals how different
regions of the bulb can be distinguished from one another based
on their profiles of gene expression. For example, the M aspect
of the olfactory bulb, which expresses both ptd2 and NT3, is
molecularly distinct from the L portion of the AV surface, which
appears to express neither gene.
Spatial maps of differential gene expression can be found
throughout development. Such maps are particularly evident
from studies on pattern formation, where complex spatial pat-
terns have been identified in the Drosophila embryo (24), the
mammalian telencephalon (25), and in the spinal cord (26).
Similarly, spatial maps are thought to be used in the formation
although guidance cues have been identified, a comprehensive
spatial map of these cues has yet to be produced. Such a map
would begin to address how these multiple signals distinguish
among potential targets by identifying overlapping patterns of
expression. The function of these individual genes could then be
tested for their role in this process (28, 29).
The identification of a spatial map in the olfactory system and
its potential function has been particularly difficult to address.
Although differentially expressed cues have been found in the
periphery, only a limited number have been described in the
bulb. In this article, we identify molecules with restricted pat-
terns of expression in the bulb by using a statistical approach for
DNA microarray-based expression profiling, and demonstrate
the existence of a spatial map in this structure. The nascent map
formed by these patterns allows us to begin to describe, on a
global level, how regions of the bulb can be differentiated from
one another at the molecular level. It should be noted, however,
that our analysis provides only a ‘‘snapshot’’ of olfactory bulb
development. Thus, it is possible that the patterns of gene
expression we observe here are a result of differential develop-
mental maturation across the olfactory bulb, and not reflective
of spatial restrictions per se. Indeed, it has been reported that
glomeruli form in a spatiotemporal wave, with rostral glomeruli
maturing before caudal glomeruli (30). However, several of the
patterns identified here describe gene-expression differences
across axes of the bulb orthogonal to the spatiotemporal axis,
indicating that such restrictions are not a secondary result of
temporal waves of cellular maturation, but instead represent
bona fide asymmetries within this tissue.
Despite the relatively large size of the tissue fragments used
in our analysis, our estimate in general successfully predicted the
location of expression patterns associated with individual genes.
bulb. Major domains of expression were retained for each gene to simplify
presentation. In this representation, A is toward the lower-right corner. Dark
blue, outline of olfactory bulb and cortex; yellow, ptd2 expression; light blue;
jagged expression; red, cadherin-11 expression; purple, NT3 expression.
www.pnas.org?cgi?doi?10.1073?pnas.0404872101Lin et al.
Note that in all of our validated cases, expected average differ-
ences in any one aspect of the bulb as compared with the pooled
reference are ?2-fold (Fig. 3), underscoring the ability of our
approach to identify subtle changes in gene expression within the
bulb and distinguish them from background noise. By clustering
genes with similar patterns together, other, weaker patterns that
may potentially be buried within the noise could be identified
and considered for further analysis.
Several of the genes that were uncovered in this screen (jagged,
IGF-2, cadherin-11, NT3, pcV?2, and vitronectin) have clearly
been implicated in synapse formation and connectivity (18, 19,
22, 31–33). These genes, as well as other members of the families
that they belong to, are excellent candidates for a more detailed
analysis of how they might affect development and target
selection within the bulb. For example, mutants in cadherin-11
have been obtained by employing transgenic techniques (19). In
these mice, the olfactory bulbs are reduced significantly in size
in convergence patterns. Thus, the genes uncovered in this work
could possibly impact many aspects of olfactory bulb develop-
ment and function.
The identification of differentially expressed genes in the
olfactory system is a fundamental and necessary first step toward
defining the mechanisms underlying the patterning of sensory
input in the olfactory bulb. We anticipate that iterations of this
nascent spatial map will incorporate progressively smaller olfac-
tory bulb fragments, allowing finer resolution of gene expression
patterns within the three-dimensional space of this structure.
Ultimately, this information should provide a rational basis for
multiple genetic alterations and interpretation of the subsequent
We thank K. van Fossen for technical assistance; J. Winer for use of the
for help in generation of the microarrays. This work was supported by
grants from the National Institutes of Health (to J.N. and T.P.S.) and by
funds from the Department of Molecular Cell Biology and the Helen
Wills Neuroscience Institute. D.M.L. was supported by the Whitehall
Foundation, the National Science Foundation, the Sloan Foundation,
and the Beckman Young Investigator Program. The 19K RIKEN cDNA
set was supported in part by the Special Coordination Funds for
Promoting Science and Technology and a research grant from the
Science and Technology Agency of the Japanese government (to Y.H.).
1. Key, B. & St. John, J. (2002) Chem. Senses 27, 245–260.
2. Cutforth, T., Moring, L., Mendelsohn, M., Nemes, A., Shah, N. M., Kim, M. M.,
Frisen, J. & Axel, R. (2003) Cell 114, 311–322.
3. Norlin, E. M., Alenius, M., Gussing, F., Hagglund, M., Vedin, V. & Bohm, S.
(2001) Mol. Cell. Neurosci. 18, 283–295.
4. Yoshihara, Y., Kawasaki, M., Tamada, A., Fujita, H., Hayashi, H., Kagamiyama,
H. & Mori, K. (1997) J. Neurosci. 17, 5830–5842.
5. Gong, Q. & Shipley, M. T. (1996) J. Comp. Neurol. 366, 1–14.
6. Crandall, J. E., Dibble, C., Butler, D., Pays, L., Ahmad, N., Kostek, C., Puschel,
A. W. & Schwarting, G. A. (2000) J. Neurobiol. 45, 195–206.
7. Schwarting, G. A., Kostek, C., Ahmad, N., Dibble, C., Pays, L. & Puschel, A. W.
(2000) J. Neurosci. 20, 7691–7697.
8. Marillat, V., Cases, O., Nguyen-Ba-Charvet, K. T., Tessier-Lavigne, M., Sotelo,
C. & Chedotal, A. (2002) J. Comp. Neurol. 442, 130–155.
9. Otaki, J. M. & Firestein, S. (1999) NeuroReport 10, 2677–2680.
10. Gogos, J. A., Osborne, J., Nemes, A., Mendelsohn, M. & Axel, R. (2000) Cell
11. Miki, R., Kadota, K., Bono, H., Mizuno, Y., Tomaru, Y., Carninci, P., Itoh, M.,
Shibata, K., Kawai, J., Konno, H., et al. (2001) Proc. Natl. Acad. Sci. USA 98,
12. Yang, Y. H., Buckley, M. J., Dudoit, S. & Speed, T. P. (2002) J. Comput. Graph.
Stat. 11, 108–136.
13. Yang, Y. H., Dudoit, S., Luu, P., Lin, D. M., Peng, V., Ngai, J. & Speed, T. P.
(2002) Nucleic Acids Res. 30, e15.
14. Dudoit, S., Yang, Y. H., Callow, M. J. & Speed, T. P. (2002) Statistica Sinica
15. Speca, D. J., Lin, D. M., Sorensen, P. W., Isacoff, E. Y., Ngai, J. & Dittman,
A. H. (1999) Neuron 23, 487–498.
16. Sandberg, R., Yasuda, R., Pankratz, D. G., Carter, T. A., Del Rio, J. A.,
Wodicka, L., Mayford, M., Lockhart, D. J. & Barlow, C. (2000) Proc. Natl.
Acad. Sci. USA 97, 11038–11043.
17. Zirlinger, M., Kreiman, G. & Anderson, D. J. (2001) Proc. Natl. Acad. Sci. USA
18. Lindsell, C. E., Shawber, C. J., Boulter, J. & Weinmaster, G. (1995) Cell 80,
19. Manabe, T., Togashi, H., Uchida, N., Suzuki, S. C., Hayakawa, Y., Yamamoto,
M., Yoda, H., Miyakawa, T., Takeichi, M. & Chisaka, O. (2000) Mol. Cell.
Neurosci. 15, 534–546.
20. Yagi, T. & Takeichi, M. (2000) Genes Dev. 14, 1169–1180.
22. Nef, S., Lush, M. E., Shipman, T. E. & Parada, L. F. (2001) Dev. Biol. 234, 80–92.
23. Guthrie, K. M. & Gall, C. M. (1991) J. Comp. Neurol. 313, 95–102.
24. St Johnston, D. & Nusslein-Volhard, C. (1992) Cell 68, 201–219.
25. Wilson, S. W. & Rubenstein, J. L. (2000) Neuron 28, 641–651.
26. Price, S. R., De Marco Garcia, N. V., Ranscht, B. & Jessell, T. M. (2002) Cell
27. Sperry, R. W. (1963) Proc. Natl. Acad. Sci. USA 50, 703–710.
28. Winberg, M. L., Mitchell, K. J. & Goodman, C. S. (1998) Cell 93, 581–591.
29. Stein, E. & Tessier-Lavigne, M. (2001) Science 291, 1928–1938.
30. Bailey, M. S., Puche, A. C. & Shipley, M. T. (1999) J. Comp. Neurol. 415,
31. Chernousov, M. A. & Carey, D. J. (2000) Histol. Histopathol. 15, 593–601.
32. Holland, S. J., Peles, E., Pawson, T. & Schlessinger, J. (1998) Curr. Opin.
Neurobiol. 8, 117–127.
33. Schvartz, I., Seger, D. & Shaltiel, S. (1999) Int. J. Biochem. Cell Biol. 31,
Lin et al.PNAS ?
August 24, 2004 ?
vol. 101 ?
no. 34 ?