Technical article

Gene expression changes in an in vitro model of

chondroinduction: a comparison of two methods

KAREN E. YATES, PhD; JULIE GLOWACKI, PhD

There are many useful technologies to describe patterns of gene expression that occur during tissue repair and
regeneration. Results from different methods used in one experimental setting are not often compared. In this case
study of chondrogenesis, we compare two methods to identify differentially expressed genes, representational
difference analysis and targeted macroarray analysis, as a model for investigating genes that may be relevant to
tissue repair. We sought to identify genes whose expression was altered when human dermal fibroblasts were
cultured in a three-dimensional, porous collagen sponge with the chondroinductive agent, demineralized bone.
Both representational difference analysis and macroarray experiments revealed several functional families of genes
as up-regulated or down-regulated in chondroinduced fibroblasts. An advantage of representational difference
analysis is that altered expression of specific mRNA transcripts can be revealed. In this example, representational
difference analysis uncovered the up-regulation of a specific transcript of Wnt5a in fibroblasts cultured with
demineralized bone. Representational difference analysis is limited, however, as there can be false negatives for
genes not readily amplified by polymerase chain reaction. We conclude that small arrays containing functional
classes of genes can be used to ask specific, hypothesis-driven questions at minimal cost. It may be prudent,
however, to use more than one method to survey differences in gene expression in order to validate and expand
findings. (WOUND REP REG 2003;11:386–392)

Many useful technologies are available to describe the ACTB b-Actin

complex patterns of gene expression that occur during the
course of tissue repair and regeneration (reviews in1–3).
Methods such as microarray yield large, complex datasets
that need to be reduced to useful information. Other
methods based upon subtractive hybridization offer great-
er opportunity for gene discovery. Those methods can
readily identify thousands of differences between two
states, conditions, or sources, but the challenge is to direct
nonspecific ‘‘data mining’’ with testable hypotheses. Those
techniques have shown potential for increasing our
understanding of pathological processes and for advancing
From the Department of Orthopedic Surgery, Brigham
and Women’s Hospital, Harvard Medical
School, Boston Massachusetts.
Reprint requests: Karen E. Yates, PhD, Orthopedic Re-
search, Brigham and Women’s Hospital, 75
Francis Street, Boston, MA 02115. Fax: (617) 732-
6705; Email: kyates@partners.org.
Copyright
2003 by the Wound Healing Society.
ISSN: 1067-1927 $15.00 + 0
API Average pixel intensity
cDNA-RDA Complementary DNA-representational
difference analysis
DBP Demineralized bone powder
hDF Human dermal fibroblast
IGF-BP3 Insulin-like growth factor-binding
protein 3
PCR Polymerase chain reaction
scientific discovery in disease diagnostics and drug
development. Some important information has been
obtained from recent applications of microarray technol-
ogies to problems of interest in wound healing.4–10 In this
case study of chondrogenesis, we compare two approa-
ches for gene expression analysis as a model for identifying
genes that may be relevant to tissue repair.
Factors that influence investigators’ choice of a
method for gene expression studies include cost, amount
of starting material, and the hypothesis being tested.
Laboratories may invest heavily and develop expertise in
one technique. The practical aspects of analyzing large

386
WOUND REPAIR AND REGENERATION
VOL. 11, NO. 5
amounts of data may also be restricting. Results from
different methods are not often compared.
Our group developed highly porous three-dimensional
(3D) collagen sponge culture systems for analysis of cell
differentiation and histogenesis.11 The lattice of thin
collagen fibers in these sponges defines pores ranging
from 120 to 200 lm in diameter.11,12 This architecture
maximizes 3D cellular interactions and promotes histo-
genesis. We have shown in vitro histogenesis of oral
mucosa,13 bone,14,15 cartilage,12,16 and induced cartilage11
with these sponges, which are biocompatible and have
potential for in vivo use.17
We sought to identify genes whose expression was
altered when human dermal fibroblasts (hDFs) were
cultured with a chondroinductive agent, demineralized
bone powder (DBP). Human DFs cultured with DBP
in collagen sponges for 7 days are induced to express
cartilage-specific matrix genes18,19 and matrix compo-
nents.16 We proposed that changes in important genes
would occur prior to expression of the chondrocyte
phenotype. After 3 days of culture in DBP/collagen spon-
ges, hDFs adhere to the particles of DBP, but cartilage-like
matrix is not yet detectable.19 Therefore, 3 days represents
an early time point of cellular interaction with DBP.
The purpose of the current study was to compare
results from two methods that were used to identify
differentially expressed genes in one experimental setting.
We used both discovery-driven and hypothesis-driven
approaches to identify differentially expressed genes in
hDFs cultured in DBP/collagen or control collagen sponges
for 3 days. cDNA representational difference analysis
(cDNA-RDA) is a polymerase chain reaction (PCR)-based
method of subtractive hybridization.20 Differentially
expressed genes are amplified via short oligonucleotide
primers annealed to cDNAs. The PCR products are
subcloned and sequenced to identify the differentially
expressed genes. Therefore, cDNA-RDA is a discovery-
driven, ‘‘nonspecific’’ approach, in that differentially
expressed genes may be found without a priori knowledge
of specific DNA sequences. We also used a hypothe-
sis-driven, ‘‘gene-targeted’’ approach focused on signal
transduction genes that could regulate postnatal chondro-
induction in this in vitro system. We used a commercially
available nylon cDNA macroarray to compare expression
of signal transduction genes in hDFs cultured in DBP/
collagen and control collagen sponges. This analysis
compares the results from these two methods.
MATERIALS AND METHODS
Skin explants were obtained from discarded materials
(neonatal foreskins) with institutional approval from the
YATES AND GLOWACKI 387
Brigham and Women’s Hospital Human Committee.
Primary hDF cultures were established by outgrowth from
minced tissue (1-mm3 pieces) and were expanded in vitro to
passage #12 before use. DBP was prepared from rat long
bones as previously described.21 Porous 3D collagen
sponges were prepared from pepsin-digested bovine type
I collagen.11 Bilaminate DBP/collagen sponges (8-mm
diameter) were prepared by packing 3 mg of DBP between
two thin (0.7-mm) layers of porous collagen. Control
sponges consisted of a single layer of collagen (1.5-mm
thick). Suspensions of hDFs were seeded onto DBP/
collagen and control collagen sponges (106 cells in 50 ll
per sponge) and cultured as described16,19 for 3 days.
RNA isolation
Total RNA was prepared from pooled DBP/collagen
(n ¼ 10) and control collagen sponges (n ¼ 10) by
homogenization in TRIzol reagent (Invitrogen, Carlsbad,
CA) according to the manufacturer¢s instructions.19 The
RNAs were examined by electrophoresis on agarose-
formaldehyde gels to ensure that they were of similar
quality.
cDNA representational difference analysis
RDA was performed as previously described.22 In brief,
poly A+ mRNA was purified from 100 lg total RNA and
reverse-transcribed into oligo dT-primed cDNA. The cDNA
was digested with Dpn II endonuclease. Oligonucleotide
linkers were annealed to the Dpn II-digested cDNA and
representations were generated by PCR. Pools of cDNAs
representing differentially expressed genes (up-regulated
or down-regulated) were identified by a series of subtrac-
tive hybridization and PCR amplification steps.22 Differ-
ence products were subcloned with the Topo Cloning Kit
(Invitrogen), sequenced, and genes identified by BLAST
Sequence Similarity Searches of the GenBank database
(http://www.ncbi.nlm.nih.gov/blast/). A match was defined
as a clone that contained an insert with ‡90% identity to a
GenBank sequence with 100 or more contiguous base
pairs. Previously, we described 88 clones identified by RDA
as up-regulated in chondroinduced hDFs on day 3.19 We
also reported connective tissue matrix genes that we found
among an additional 172 up-regulated clones, as well as 111
down-regulated clones.23 Gene expression levels in chon-
droinduced and control hDFs were measured by semi-
quantitative RT-PCR or by Northern hybridization.19, 23
Nylon cDNA macroarray analysis
GEArrayTM Q series Human Signal Transduction Pathway
Finder nylon macroarrays were purchased (catalog #HS-
008, SuperArray, Inc., Bethesda, MD), in which cDNA
features 300–600 bp in length are spotted in quadruplicate

388 YATES AND GLOWACKI
in a 1-mm diameter ‘‘tetrad.’’ There are 96 experimental and
six control features per array. The controls are plasmid
(pUC18), blank (no DNA), and four potential normalization
features (glyceraldehye-3-phosphate dehydrogenase, pept-
idylprolyl isomerase A, ribosomal protein L13a, and b-actin.
Duplicate hybridizations were performed for each
pooled RNA. Aliquots of 3 lg of total RNA were reverse-
transcribed into [32P]-labeled cDNA with reagents provided
by the manufacturer. Each labeled cDNA was hybridized
overnight to an array, which was washed and exposed to
X-ray film. Initial autoradiographs showed a wide range of
expression that prevented accurate assessment of all
experimental features from a single exposure time.
Therefore, multiple exposures were performed (1, 2, 5,
and 18 hours). Digital images were acquired from autora-
diographs with an Epson 1200s Scanner with transparency
adapter. Average pixel intensity (API) was measured in a
circle 10 pixels in diameter centered within each feature
tetrad.
GEArrayAnalyzerTM version 1.2 software was used to
calculate gene expression levels for each feature. The
usable range of raw data values (API) was between the
minimum detection due to background on the autoradio-
graphs (API ¼ 8000) and the maximum or plateau reading
(API ¼ 65,000). Raw data values that fell within those
limits were accepted for analysis. Raw data values that
exceeded those limits were discarded and reevaluated on
another autoradiograph with a different duration of
exposure and a different normalization gene. Raw data
values less than or equal to background levels on an
18-hour film exposure were classified as ‘‘absent calls.’’
Features that were located adjacent to highly expressed
genes were excluded from analysis because of ‘‘bleeding’’
of the strong signals.
The raw data were transformed by background
subtraction (API of the pUC18 feature) and normalization
for each exposure. Measurements of gene expression
levels were made only when a signal was clearly visible by
eye on the autoradiographs. Autoradiographs were inspec-
ted visually to select the most appropriate normalization
reference for measurement of each feature (e.g., similar
signal intensity). Another requirement for selection of a
normalization reference was that its raw data value be
within the usable range. Gene expression levels were
calculated as a ratio of feature to reference and averaged.
Normalized gene expression values were ratified by visual
inspection of the autoradiographs.
Northern hybridization
Total RNA was subjected to electrophoresis through 1%
agarose gels (10 lg per lane) and was transferred onto a
positively charged nylon membrane (Roche Molecular
WOUND REPAIR AND REGENERATION
SEPTEMBER–OCTOBER 2003
Biochemicals). A probe corresponding to a portion of the
Wnt5a 3¢ untranslated region (GenBank accession No.
L20861, bases 3,204–3718) was prepared by PCR ampli-
fication, purification by a QIAquick PCR Purification Kit
(Qiagen, Valencia, CA) and labeling with [32P]-dCTP using
the High Prime Random Priming Kit (Roche Applied
Science, Indianapolis, IN). Hybridization, washing, and
detection steps were as described.19,23 Autoradiographs
were scanned with an Epson 1200s Scanner with a
transparency adapter and the images were analyzed with
Scion Image software (Scion Corporation, Frederick,
MD). Gene expression levels were normalized to 18S
RNA (18S rRNA oligonucleotide, Ambion, Inc., Austin
TX).
RESULTS
We used cDNA-RDA to identify differentially expressed
genes in hDFs exposed to DBP for 3 days. As expected for
a PCR-based method, there was redundancy among the
pools of differentially expressed genes: 172 up-regulated
clones corresponded to 82 different GenBank sequences,
and 111 down-regulated clones corresponded to
72 GenBank sequences. Clones that corresponded to
fibronectin were numerous (19 clones) among the up-
regulated genes. Three clones corresponding to portions of
fibronectin were also found among the pool of down-
regulated genes. Up-regulation or down-regulation for
selected genes revealed by cDNA-RDA was confirmed by
semiquantitative RT-PCR or Northern hybridization. The
magnitude of differences in chondroinduced hDFs ranged
from 0.6 to 6.5-fold.
Northern hybridization showed that cDNA-RDA detec-
ted differences in expression of specific mRNA isoforms.
For example, one of the up-regulated clones (No. 93A)
corresponded to the human Wnt5a 3¢ untranslated
region (GenBank accession #L20861, bases 3,204–3718;
Figure 1A). Northern hybridization with this clone as a
probe showed expression of four different Wnt5a mRNAs
at equivalent levels in chondroinduced and control hDFs
on day 3, whereas a specific, 9.0-kb transcript was found
only in chondroinduced hDFs (Figure 1, B and C).
Identification of differentially expressed genes by
nylon cDNA macroarray
We measured expression of 96 ‘‘signal transduction’’ genes
in control and chondroinduced hDFs with a commercial
nylon macroarray. The same RNA samples were used for
this analysis as were used for the representational
difference analysis. Fibronectin and Wnt1 were the most
highly expressed genes in control hDFs, and insulin-like
growth factor binding protein 3 (IGF-BP3) and b-actin
WOUND REPAIR AND REGENERATION
VOL. 11, NO. 5 YATES AND GLOWACKI 389
(A) evaluate expression levels for all of the genes on the array;
Wnt5a 17 genes were excluded because their cDNAs were
adjacent to highly expressed genes.
It was possible to compare expression levels for 15
Clone 93A genes.24 Four genes (cyclin-dependent kinase inhibitor 1A,
(500 bp) EGR-1, fibronectin, IGF-BP3) were up-regulated in hDFs
cultured in DBP/collagen sponges for 3 days (range of 1.9-
(B) Col DBP to 11.9-fold greater expression than in hDFs cultured in
control collagen sponges). An unexpected finding was that
-9.0 kb expression of b-actin, a feature intended as a normalization
-8.2 kb control, was also increased (1.9-fold) in chondroinduced
-6.8 kb
hDFs. Eight genes (CD5, cyclin-dependent kinase inhibitor
-4.5 kb 2A, engrailed-1, fatty acid synthase, hepatocyte nuclear
factor 3 beta, patched homolog 1, platelet/endothelial cell
adhesion molecule 1, wnt 1) were down-regulated (range
of 40–70% of control). Three genes (heat shock proteins,
-1.9 kb 27 and 90, quinone oxidoreductase homolog) were
expressed equally in control and chondroinduced hDFs.

18S RNA DISCUSSION

This post-hoc analysis shows the relative advantages of
(C) two methods of gene expression analysis from one
10 experimental setting. Each of these methods has potential
for advancing scientific discovery in tissue formation and
Gene Expression

1
0.1
0.01
1.9 4.6 6.8 8.2
Transcript Size (kb)
FIGURE 1. Expression of Wnt5a mRNA in hDFs cultured in DBP/
collagen (DBP) and control sponges (Col) for 3 days. (A) Sche-
matic of the 4.1-kb Wnt5a mRNA sequence from GenBank
(Accession No. L20861). The open bar indicates the Wnt5a
protein-coding sequence, and the black lines indicate untranslat-
ed regions. The location of clone 93A, a sequence found among
the pool of up-regulated genes identified in the RDA experiment, is
indicated by a closed bar. (B) Northern hybridization that used
clone 93A as probe. (C) Gene expression levels were measured for
each Wnt5a mRNA shown in panel B and were normalized to 18S
RNA.
were the most highly expressed genes in chondroinduced
hDFs.
Thirty-two percent (31/96) of the genes on the array
were present calls. Of those, 16 genes were expressed in
either chondroinduced or control hDFs at levels too low
for quantification or comparison. It was not possible to
repair. Our experience illustrates the value of using two
methods to obtain unique information as well as confirm-
atory results. Use of such complementary approaches can
be applied to comparisons of gene expression in, for
example, normal and keloid scars, acute and chronic
wounds, partial- and full-thickness grafts, or young and old
9 repair tissue.
In this study, we compared two methods to identify
Collagen
differentially expressed genes in hDFs cultured with DBP.
DBP/Collagen We used cDNA-RDA for discovery-driven studies of gene
expression during chondroinduction.19 That work estab-
lished that several functional classes of genes, including
signal transduction genes, are altered by DBP.19,23 For this
study, we used a macroarray for a set of signal transduc-
tion genes because the regulatory mechanisms activated in
hDFs by DBP are not well known. Small arrays containing
functional classes of genes can be used to ask specific,
hypothesis-driven questions at minimal cost. Small cDNA
arrays therefore may be more feasible than high-density
arrays for targeted or pilot studies.
Our experience with the practical aspects of two
methods to identify differentially expressed genes, cDNA-
RDA and cDNA macroarray, is summarized in Table 1. We
used the recommended amount of total RNA for cDNA-
RDA, but this technique has been successful with nano-
gram quantities of poly A + mRNA.25 cDNA-RDA has been
used with small numbers of primary cells or tissues26–28 and

390 YATES AND GLOWACKI
Table 1. Comparison of features of cDNA-RDA and cDNA
macroarray methods to identify differentially expressed genes
in hDFs chondroinduced with demineralized bone
Method
Attribute RDA Macroarray
Amount of total RNA used 100 lg 3 lg
Sample Preparation cDNA synthesis cDNA synthesis
PCR amplification required Yes No
Time to complete 3 weeks 3 days
experimental protocol
Identification of genes DNA sequencing Manufacturer’s
key
Time to complete Days Days
data analysis
Output List of genes List of genes
Library of clones
is appropriate for limited sources of RNA. Both RDA and
macroarray analysis require cDNA synthesis, and RDA also
requires PCR amplification of cDNAs. The amount of time
required to complete the protocol was shorter for the array
experiment, particularly in the post-experimental analysis.
The necessity of cloning and sequencing difference
products increased the time and expense of identifying
genes, but that cost was balanced by the benefit of having a
library of clones for further studies. The findings from both
methods support our hypothesis that chondrogenesis is
initiated by changes (both up and down) in more than one
cellular signaling pathway. Some functional classes of
genes (such as growth factors and matrix proteins) were
identified as changed by both methods. IGF-BP3 was
identified as up-regulated both by cDNA-RDA and on the
signal transduction array. There were no genes identified
as up-regulated by one method and down-regulated by the
other.
Subsequent studies using cDNA arrays with different
genes identified four additional up-regulated genes that
were also revealed in the cDNA-RDA experiment.24 An
unexpected finding in the cDNA-RDA experiment was that
fibronectin was present in both the up-regulated and down-
regulated pools of clones. This was explained as cross-
contamination of the pools. Therefore, fibronectin was
initially excluded from the curated cDNA-RDA results. The
macroarray experiment showed fibronectin was the most
abundantly expressed gene on the array in control hDFs,
and was expressed at even greater levels in chondroin-
duced hDFs. Given the high levels of gene expression, it is
likely that subtraction of fibronectin was incomplete
during cDNA-RDA. Use of the two methods resolved the
apparently internally inconsistent findings for fibronectin
by cDNA-RDA.
The cDNA-RDA experiment uncovered up-regulation
of the Wnt5a 3¢ untranslated region. Northern hybridization
with that sequence as probe showed that only one of five
WOUND REPAIR AND REGENERATION
SEPTEMBER–OCTOBER 2003
different transcripts was uniquely expressed in hDFs
exposed to DBP. Expression of Wnt5a in human tissues
appears to be complex. In one study, five different Wnt5a
transcripts (approximate sizes 9.2, 8.6, 6.8, 5.2, and 4.6 kb)
were shown to be expressed in tissue-specific patterns in
human tissues and cell lines.29 The differences between
transcripts appear to be mainly in the 3¢ untranslated
regions, which can be 7kb or longer.30 There are also at
least two transcription initiation sites and two polyadeny-
lation signals in the human Wnt5a gene.30 The finding that a
specific Wnt5a mRNA transcript was up-regulated in
chondroinduced hDFs suggests that the transcript is of
particular significance in the process of chondroinduction
and should be investigated further for its function.
The ability of cDNA-RDA and methods like it to identify
changes in gene expression of differentially processed
mRNAs is a significant advantage. The importance of 3¢
untranslated regions has only recently been understood.
These sequences encode information for subcellular local-
ization of the mRNA and translated protein, regulation of
translation, and message stability (reviews in31,32). Muta-
tions in 3¢ untranslated regions can contribute to human
disease (review in 33). The average length of human
transcripts is estimated at approximately 1000 bp (range
of 21–8555 bp).34 Clearly, much information relevant to
cell biology may be lost if untranslated regions of mRNA
are not included in analysis of differentially expressed
genes.
The number of altered genes identified in the macro-
array experiment was small compared to the number
shown by cDNA-RDA. Greater numbers of altered genes
can be identified with high-density nylon or glass slide
arrays. Even with high-density arrays, the percentage of
altered genes may be very low. Two percent of genes (167/
8734) were found to be altered in an ear punch model of
wound repair,6 and 5% of genes (64/1176) were found to be
altered after laser ablation of mouse cornea.8 In a study of
punch biopsy healing in normal human skin, 3% of genes
(124/4000) were altered after 30 minutes, and 8% (310/
4000) were altered after 60 minutes.5 In comparison, we
found altered expression in hDFs cultured with DBP for
12% of genes (12/97) on a signal transduction array.
Some investigators have used cDNA-RDA or other
subtractive hybridization-based methods to identify differ-
entially expressed genes, and then prepared that library
of clones for ‘‘custom-made’’ DNA arrays.9,28,35 For such
applications, one or two rounds of RDA are performed with
low molar ratios (1 : 100 or 1 : 500) of DNA for subtractive
hybridization. Genes that are not differentially expressed
are therefore included in the difference product and can
serve as controls on custom-made arrays. The number of
genes that are subsequently confirmed to be altered by
WOUND REPAIR AND REGENERATION
VOL. 11, NO. 5
these custom arrays vary from 90%9,28 to 30%35 and is
probably related to the experimental system and the extent
of subtraction.
Another approach is to use difference products as
probes for DNA arrays in order to increase the sensitivity
of detection.28,36,37 Both low-and high-abundance sequences
are identified when difference products are used as
probes.37 Sequences that have the greatest differences in
expression, however, are preferentially enriched.28,37
This report shows how RDA and arrays used in
combination may provide a more complete survey of
differences in gene expression. Small but significant
changes in specific mRNAs, particularly within 3¢ untrans-
lated regions, may go undetected in cDNA arrays. Abun-
dantly expressed genes could be excluded as ‘‘false
positives’’ in cDNA-RDA. We conclude that it may be
prudent to use more than one method to survey gene
expression profiles in a biological system in order to
validate and expand findings.
ACKNOWLEDGMENTS
We thank Shuanhu Zhou and Rebecca MacLean for their
assistance with the experiments. This work was supported
by grants from the Aircast Foundation (#SA-501, KEY), and
the National Institutes of Health (#AR044873, JG).
REFERENCES
1. Wallrapp C, Muller-Pillasch F, Micha A, Wenger C, Geng M, Solinas-
Toldo S, Lichter P, Frohme M, Hoheisel JD, Adler G, Gress TM.
Strategies for the detection of disease genes in pancreatic cancer. Ann
NY Acad Sci 1999;880:122–46.
2. Liang P. A decade of differential display. Biotechniques 2002;33:
338–46.
3. Ye SQ, Lavoie T, Usher DC, Zhang LQ. Microarray, SAGE and their
applications to cardiovascular diseases. Cell Res 2002;12:105–15.
4. Tsou R, Cole JK, Nathens AB, Isik FF, Heimbach DM, Engrav LH,
Gibran NS. Analysis of hypertrophic and normal scar gene expression
with cDNA microarrays. J Burn Care Rehabil 2000;21:541–50.
5. Cole J, Tsou R, Wallace K, Gibran N, Isik F. Early gene expression
profile of human skin to injury using high-density cDNA microarrays.
Wound Rep Reg 2001;9:360–70.
6. Li X, Mohan S, Gu W, Baylink DJ. Analysis of gene expression in the
wound repair/regeneration process. Mamm Genome 2001;12:52–9.
7. Tsukahara T, Tsujino S, Arahata K. CDNA microarray analysis of gene
expression in fibroblasts of patients with X-linked Emery-Dreifuss
muscular dystrophy. Muscle Nerve 2002;25:898–901.
8. Cao Z, Wu HK, Bruce A, Wollenberg K, Panjwani N. Detection of
differentially expressed genes in healing mouse corneas, using cDNA
microarrays. Invest Ophthalmol Vis Sci 2002;43:2897–904.
9. Hadjiargyrou M, Lombardo F, Zhao S, Ahrens W, Joo J, Ahn H, Jurman
M, White DW, Rubin CT. Transcriptional profiling of bone regener-
ation. Insight into the molecular complexity of wound repair. J Biol
Chem 2002;33:30177–82.
10. Samulewicz SJ, Seitz A, Clark L, Heber-Katz E. Expression of prea-
dipocyte factor-1 (Pref-1), a delta-like protein, in healing mouse ears.
Wound Rep Reg 2002;10:215–21.
YATES AND GLOWACKI 391
11. Mizuno S, Glowacki J. Three-dimensional composite of demineralized
bone powder and collagen for in vitro analysis of chondroinduction of
human dermal fibroblasts. Biomaterials 1996;17:1819–25.
12. Allemann F, Mizuno S, Eid K, Yates KE, Zaleske D, Glowacki J.
Effects of hyaluronan on engineered articular cartilage ECM gene
expression in 3-dimensional collagen scaffolds. J Biomed Mat Res
2001;55:13–9.
13. Navarro FA, Mizuno S, Huertas JC, Glowacki J, Orgill DP. Perfusion of
medium improves growth of human oral neomucosal tissue con-
structs. Wound Rep Reg 2001;9:507–12.
14. Mueller SM, Mizuno S, Gerstenfeld LC, Glowacki J. Medium per-
fusion enhances osteogenesis by murine osteosarcoma cells in
three-dimensional collagen sponges. J Bone Mineral Res 1999;14:
2118–26.
15. Mueller SM, Glowacki J. Age-related decline in the osteogenic
potential of human bone marrow cells cultured in three-dimensional
collagen sponges. J Cell Biochem 2001;82:583–90.
16. Mizuno S, Glowacki J. Chondroinduction of human dermal fibroblasts
by demineralized bone in three-dimensional culture. Exp Cell Res
1996;227:89–97.
17. Glowacki J, Yates KE, Warden S, Allemann F, Peretti G, Strongin G,
MacLean R, Zaleske D. Engineering a biological joint. Ann NY Acad
Sci 2002; 961:126–9.
18. Glowacki J, Yates K, Little G, Mizuno S. Induced chondroblastic dif-
ferentiation of human dermal fibroblasts by three-dimensional culture
with demineralized bone matrix. Mat Sci Eng C 1998;6:199–203.
19. Yates KE, Mizuno S, Glowacki J. Early shifts in gene expression
during chondroinduction of human dermal fibroblasts. Exp Cell Res
2001;265:203–11.
20. Hubank M, Schatz D. cDNA representational difference analysis:
a sensitive and flexible method for identification of differentially
expressed genes. Meth Enzymol 1999;303:325–49.
21. Glowacki J. Cellular reactions to bone-derived material. Clin Orthop
Rel Res 1996;324:47–54.
22. Yates KE, Forbes RL, Glowacki J. New chondrocyte genes found by
representational difference analysis of chondroinduced human
fibroblasts. Cells Tiss Org 2003. [In press.]
23. Yates KE, Glowacki J. Altered expression of connective tissue genes
in postnatal chondroinduced human dermal fibroblasts. Conn Tiss
Res 2003. [In press.]
24. Yates KE, Zhou S, Glowacki J. Genes that influence cell differentiation
are altered by demineralized bone in vitro. J Bone Min Res 2003. [In
press.]
25. O’Neill MJ, Sinclair AH. Isolation of rare transcripts by representa-
tional difference analysis. Nuc Acids Res 1997 25:2681–2.
26. Chu CC, Paul WE. Fig1, an interleukin 4-induced mouse B cell gene
isolated by cDNA representational difference analysis. Proc Natl Acad
Sci U S A 1997;94:2507–12.
27. Simons A, Janssen IM, Suijkerbuijk RF, Veth RPH, Pruszczynski M,
Hulsbergen-van de Kaa CA, du Manoir S, Geurts van Kessel A. Isolation
of osteosarcoma-associated amplified DNA sequences using repre-
sentational difference analysis. Genes Chrom Cancer 1997;20:196–
200.
28. Welford SM, Gregg J, Chen E, Garrison D, Sorenson PH, Denny CT,
Nelson SF. Detection of differentially expressed genes in primary
tumor tissues using representational differences analysis coupled to
microarray hybridization. Nuc Acids Res 1998;26:3059–65.
29. Iozzo RV, Eichstetter I, Danielson KG. Aberrant expression of
the growth factor Wnt-5A in human malignancy. Cancer Res
1995;55:3495–9.
30. Danielson KG, Pillarisetti J, Cohen IR, Sholehvar B, Huebner K, Ng L-J,
Nicholls JM, Cheah KSE, Iozzo RV. Characterization of the complete
genomic structure of the human WNT-5A gene, functional analysis of
its promoter, chromosomal mapping, and expression in early human
1 embryogenesis. J Biol Chem 1995;270:31225–34.
31. Grzybowska EA, Wilczynska A, Siedlecki JA. Regulatory functions of
3¢ UTRs. Biochem Biophys Res Comm 2001;288:291–5.

392 YATES AND GLOWACKI
32. Stebbings H. Cytoskeleton-dependent transport and localization of
mRNA. Int Rev Cytol 2001;211:1–31.
33. Conne B, Stutz A, Vassalli J-D. The 3¢ untranslated region of messenger
RNA: a molecular ‘‘hotspot’’ for pathology? Nat Med 2001;6:637–41.
34. Pesole G, Mignone F, Gissi C, Grillo G, Licciulli F, Liuni S. Structural
and functional features of eukaryotic mRNA untranslated regions.
Gene 2001;276:73–81.
35. Dillner K, Kindblom J, Flores-Morales A, Pang S-T, Törnell J, Wennbo
H, Norstedt G. Molecular characterization of prostate hyperplasia in
WOUND REPAIR AND REGENERATION
SEPTEMBER–OCTOBER 2003
prolactin-transgenic mice by using cDNA representational difference
analysis. Prostate 2002;52:139–49.
36. Geng M, Wallrapp C, Müller-Pillasch F, Frohme M, Hoheisel JD, Gress
TM. Isolation of differentially expressed genes by combining repre-
sentational difference analysis (RDA) and cDNA library arrays.
Biotechniques 1998;5:434–8.
37. Kim S, Zeller K, Dang CV, Sandgren EP, Lee LA. A strategy to identify
differentially expressed genes using representational difference ana-
lysis and cDNA arrays. Anal Biochem 2001;288:141–8.