Research Collection Habilitation Thesis Transcriptomic and proteomic approaches to the discovery of new markers of pathology Author(s): Elia, Giuliano Publication Date: 2005 Permanent Link: https://doi.org/10.3929/ethz-a-005151527 Rights / License: In Copyright - Non-Commercial Use Permitted This page was generated automatically upon download from the ETH Zurich Research Collection . For more information please consult the Terms of use . ETH Library
249
Embed
Rights / License: Research Collection In Copyright - Non ...28533/eth-28533-01.pdfCancer chemotherapy relies on the expectation that anti-cancer drugs will preferentially kill rapidly
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Research Collection
Habilitation Thesis
Transcriptomic and proteomic approaches to the discovery ofnew markers of pathology
Transcriptomic and Proteomic Approaches to the Discovery of
New Markers of Pathology
A Habilitation Thesis
Presented by
Dr. Giuliano Elia
Institute of Pharmaceutical Sciences Department of Chemistry and Applied Biosciences
Swiss Federal Institute of Technology Zurich
Zurich, February 2005
Index
Page 1. Abstract 5 2. Introduction 8
2.1. Antibody based tumor targeting 8 2.2. The tumor environment and tumor-associated antigens 10 2.3. Angiogenesis and tumor angiogenesis 13 2.4. Markers of angiogenesis 18
2.4.1. EDB domain of fibronectin 18 2.4.2. Integrins αvβ3 and αvβ5 19 2.4.3. Prostate-specific membrane antigen (PSMA) 20 2.4.4. Endoglin (CD105) 20 2.4.5. VEGF and VEGF-receptor complex 21 2.4.6. CD44 21 2.4.7. Phosphatidyl serine phospholipids 22 2.4.8. Large isoform of tenascin C 22 2.4.9. Magic Roundabout 23 2.4.10. Components of the coagulation cascade 23
2.5. Transcriptomic methods for target identification 26
2.5.1. Evolution of transcriptomics 26 2.5.2. Application of transcriptomics methods to the
identification of new targets for tumor therapy 30
2.6. Proteomic methods for target identification 32 2.6.1. Evolution of proteomics: gel-based methods 32
2.6.1.1. Two-dimensional polyacrylamide gel electrophoresis 32 2.6.1.2. Combination of chromatography and 1D SDS-PAGE 36
3.3. Identification and relative quantification of membrane
proteins by surface biotinylation and two-dimensional peptide mapping 109
3.3.1. Surface biotinylation and two-dimensional peptide mapping for the proteomic study of membrane proteins 111
3.3.2. Detection of tryptic peptides derived from a BSA-spike added to an HEK membrane protein extract 115
3.3.3. Two-dimensional peptide mapping of membrane proteins of HUVEC cell cultures 117
3.3.4. Comparison of membrane protein expression in HUVEC cells exposed to hypoxia and normoxia 118
3.3.5. Discussion 121
3.4. In vivo protein biotinylation for identification of
organ–specific antigens accessible from the vasculature. 125 3.4.1. Terminal perfusion and in vivo biotinylation 127 3.4.2. Histochemical analysis of biotinylated structures
in organ and tumor sections 129 3.4.3. Purification and identification of biotinylated proteins
by proteomic techniques 131 3.4.4. Validation of some candidate marker proteins identified 135 3.4.5. Discussion 138
4. Conclusions and Outlook 147
4.1. Conclusions 147 4.2. Ex vivo protein biotinylation of surgical specimens
from tumor-bearing patients 148 5. Materials and Methods 151
5.4. Mass spectrometry 160 5.4.1. µLC-MS/MS 160 5.4.2. MALDI-TOF/TOF MS 161
5.5. In vitro biotinylation of surface proteins for 2D peptide mapping 162 5.5.1. Biotinylation of standard proteins 162 5.5.2. Biotinylation of cell surface proteins 163 5.5.3. Isolation of biotinylated proteins 163 5.5.4. Digestion of the eluted proteins 164
5.6. Two-dimensional peptide mapping 165 5.6.1. Reversed-phase HPLC 165 5.6.2. Programming of the Spectational software 165 5.6.3. Processing of MALDI-TOF spectra 166
5.7. In vivo biotinylation of proteins accessible from the bloodstream 166 5.7.1. Animal experiments 166 5.7.2. Terminal perfusion of animals and in vivo biotinylation 167 5.7.3. Preparation of protein extracts from organ homogenates 168 5.7.4. Biotinylated protein purification 168 5.7.5. On-resin tryptic digestion of biotinylated proteins 169
5.8. Target validation methods 169 5.8.1. Histochemical analysis of biotinylated structures in
organ and tumor sections 169 5.8.2. Western Blot analysis 170 5.8.3. Immunofluorescence experiments 172 5.8.4. Indirect Immunofluorescence 173
6. Acknowledgements 174 7. References 176 8. Appendix A 229 9. Appendix B 234
4
1. Abstract
The identification of new molecular markers of pathology, to be used as target for the
specific delivery of diagnostic or therapeutic agents, is possibly the most important
goal of modern pharmacological research. In this thesis, I will first describe results
obtained by transcriptomic and proteomic high-throughput methods, applied to simple
in vitro model systems, aiming at the identification of new, candidate targets for
cancer therapy.
We have performed a broad-range analysis of the patterns of gene expression in
normal human dermal fibroblasts at two different pH values (in the presence and in
the absence of serum), with the aim of getting a deeper insight in the regulation of the
transcriptional program as a response to a pH change. Using the Affymetrix gene chip
system, we found that the expression of 2,068 genes (out of 12,565) was modulated
by more than two-fold at 24, 48 or 72 hours after the shift of the culture medium pH
to a more acidic value, stanniocalcin 1 being a remarkable example of strongly up-
regulated gene. Genes displaying a modulated pattern of expression included, among
others, cell cycle regulators (consistent with the observation that acidic pH abolishes
the growth of fibroblasts in culture) and relevant extracellular matrix (ECM)
components. Extracellular matrix protein 2, a protein with a restricted pattern of
expression in adult human tissues, was found to be remarkably over expressed,
consequent to serum starvation.
Aiming at getting a detailed insight into the oxygen-dependent regulation of the
transcriptional program of vascular endothelial cells, we have performed a broad-
range transcriptomic analysis, using the Affymetrix HG-U133A Gene Chips, of
mRNA expression levels in human umbilical cord vein endothelial cells (HUVEC),
5
exposed in vitro to hypoxia for different time periods. The transcriptomic analysis
was complemented by a semi-quantitative RT-PCR analysis of mRNA levels and
alternative splicing for some selected extracellular matrix protein genes, and by a
proteomic analysis (using 2D-PAGE and tandem mass spectrometry for protein
separation and identification) of hypoxic and normoxic HUVEC whole cell lysates
and sub-cellular fractions. Our analysis confirmed previous findings on genes whose
expression is regulated by oxygen concentration, but also identified new genes (e.g.,
The profiling of complex proteomes by LC-MS/MS is complicated by the very large
number of redundant peptides. Theoretically, one unique peptide would be sufficient
to identify unambiguously each parent protein. If such unique peptides could be
isolated, the complexity of the samples could be reduced by one or two orders of
magnitude considered that tryptic digestion generates several dozen peptides per
protein (Zhang, Yan et al. 2004). In 2002, the group of Joël Vandekerckhove
introduced a peptide-based protein identification technique termed “combined
fractional diagonal chromatography” (COFRADIC™) which allows the selective
enrichment of peptides containing unique (N-terminal) or rare (Cys, Met) amino acids
(Gevaert, Van Damme et al. 2002). COFRADIC™ separates analytes in two
sequential RP-HPLC runs. Between the two runs, fractions collected from the first
dimension are chemically modified in a way that alters the retention time of the
peptides containing the target amino acid. Each fraction is then analyzed in the second
dimension using the same chromatographic conditions as in the first. Consequently,
all modified peptides exhibit altered retention times and are thus known to contain the
targeted amino acid whereas the remaining peptides elute from the column with the
same retention time as in the first dimension. The modified peptides can then be
specifically collected for further LC-MS/MS analysis. Gevaert and colleagues
successfully applied COFRADIC™ to the analysis of methionine-containing peptides
40
in a total unfractionated cell lysate obtained from E.coli K12 cells (Gevaert, Van
Damme et al. 2002). More than 800 proteins were identified including abundant, rare,
large, small, acidic, basic and hydrophobic proteins. In another experiment, the same
group analyzed N-terminal peptides isolated from both a cytosolic and membrane-
cytoskeleton fraction of human thrombocytes (Gevaert, Goethals et al. 2003). Free
amino groups were first blocked by acetylation and then digested with trypsin. After
RP-HPLC of the generated peptides, internal peptides were blocked using 2,4,6-
trinitrobenzenesulfonic acid; the blocked internal peptides displayed a strong
hydrophobic shift and therefore segregated from the unaltered N-terminal peptides
during a second identical separation step. More recently, Gevaert et al. specifically
isolated cysteine-containing peptides in samples obtained from human platelets and
enriched human plasma (Gevaert, Ghesquiere et al. 2004).
The COFRADIC™ technique features a high dynamic range (the simultaneous
detection of proteins present in ratios of 1:10,000 is possible) coupled with the ability
of isolating large numbers of membrane proteins. However, it remains to be seen
whether COFRADIC™ can be applied to perform a relative quantification in two
closely related protein samples. This goal may be facilitated by the stable
incorporation of 18O at newly formed carboxy termini that are generated by
proteolytic cleavage between the two COFRADIC™ runs (Gevaert, Ghesquiere et al.
2004).
2.6.2.3. Isotope-coded affinity tags (ICAT)
In 1999, Aebersold and coworkers introduced the concept of isotope-coded affinity
tags (“ICAT”) for the stable non-radioactive isotopic labeling of proteins, compatible
with protein identification and relative quantification in different biological specimens
41
(Gygi, Rist et al. 1999). In its original implementation, the ICAT technology consists
in the biotinylation of cysteine residues in proteins with reactive derivatives of biotin,
carrying a linker arm with hydrogen or deuterium atoms. These “light” and “heavy”
biotin derivatives serve a dual purpose. First, they allow reducing the complexity of
tryptic peptides to be analyzed in a gel-free mass-spectrometry experiments (they can
be purified on affinity resins; only few tryptic peptides in a protein contain a cysteine
residue). Second, the labeling of peptides from two different samples with a light or
heavy tag allows the use of LC-MS/MS methodologies for the relative comparison of
protein abundance in the two samples. The relative protein abundance is in fact
reflected in the relative intensity of the mass spectrometry signals of the
corresponding biotinylated peptides, which are separated in the m/z axis by the
number of Daltons corresponding to the number of atoms that are either hydrogen or
deuterium in the biotin derivative tag. A number of modified ICAT implementations
have been developed in the last few years. They include the use of ICAT for the
relative quantization of spots in 2D gels (Smolka, Zhou et al. 2002), the use of solid-
phase isotope tagging (Zhou, Ranish et al. 2002) and the use of special chemical
procedures for the analysis of post-translational modifications, such as glycosylation
and phosphorylation (Zhou, Watts et al. 2001).
The ICAT technology has recently been used for the quantitative profiling of
differentiation-induced microsomal proteins. The method was used to identify and
determine the ratios of abundance of each of 491 proteins contained in the
microsomal fractions of naïve and in vitro-differentiated human myeloid leukemia
cells (Han, Eng et al. 2001). In this study, the authors recognize that a subset of
proteins that lack cysteine residues, very low abundant proteins and very hydrophobic
proteins would not be analyzed with this technique.
42
The substitution of a thiol-reactive ICAT reagent with a similar ICAT reagent,
capable of reaction with primary amino groups is not straightforward. The higher
number of lysine residues in proteins (compared to cysteines) may lead to incomplete
chemical coupling, and to a distribution of patterns of (incomplete) lysine labeling,
thus complicating the down-stream LC-MS/MS analysis.
2.6.2.4. iTRAQ™ Reagents
At the 6th Siena Proteomic Meeting “From genome to proteome”, in September 2004,
Applied Biosystems reported about the development of a multiple set of four isobaric
iTRAQ™ Reagents 114, 115, 116 and 117 which are amine specific and yield labeled
peptides which are identical in mass and hence also identical in single MS mode, but
which produce strong, diagnostic, low-mass MS/MS signature ions allowing for
quantization of up to four different samples simultaneously. iTRAQ™ Reagents
consist of a reporter group, a balance group, and a peptide reactive group. The peptide
reactive group covalently links an iTRAQ™ Reagent isobaric tag with each lysine
side chain and N-terminus group of a peptide, labeling all peptides in a given sample
digest. The balance group ensures that an iTRAQ™ Reagent-labeled peptide displays
the same mass, whether bound to iTRAQ™ reagent 114, 115, 116, or 117. During
MS/MS, the isobaric tag cleaves and because of fragmentation, there is neutral loss of
the balance group. The iTRAQ™ reporter groups are generated, displaying diagnostic
ions in the low-mass region between m/z of 114 – 117.
With reagents that label amines instead of thiols, the overall protein and proteome
coverage is improved, while posttranslational modification (PTM) information is
retained. Specific proteins of interest can be quantified in absolute terms by including
labeled internal standard peptides representative for the protein of interest. In contrast
43
to ICAT reagents, which label cysteines prior to protein digestion, iTRAQ™
Reagents react with peptides derived from proteolytically digested protein samples.
Labeling peptides instead of proteins features several advantages: first, peptides are
more soluble than proteins. It is therefore likely, that peptides representative for large
hydrophobic proteins (e.g. membrane proteins), which would escape analysis in a
protein labeling experiment are detected in a peptide labeling experiment. Second,
quantitative labeling is more feasible with peptides than with proteins, since the target
amino groups are easier accessible in peptides compared to proteins.
iTRAQ™ appears to be an interesting alternative to ICAT, however, its potential in
terms of dynamic range and detection of “difficult” proteins (e.g. membrane proteins)
remains to be evaluated.
A conceptually similar approach for the accurate quantification of peptides and
proteins has been published by Thompson and colleagues in 2003 (Thompson,
Schafer et al. 2003).
2.6.2.5. Two-dimensional peptide mapping
Stimulated by the work of Schrader, Schulz-Knappe and colleagues (Schulz-Knappe,
Zucht et al. 2001; Tammen, Hess et al. 2002) which have routinely used the
orthogonal combination of chromatography and matrix assisted laser desorption
ionization-time of flight mass spectrometry (MALDI-TOF MS) for the relative
quantization of peptides in biological fluids (e.g., sera and cerebrospinal fluids), we
have set up a method for the simultaneous recovery, separation, identification and
relative quantization of membrane proteins isolated from cultured mammalian cells
termed two-dimensional peptide mapping (2D-PM).
44
In the course of 2D-PM, cells are biotinylated with a biotin reagent reactive for
primary amino groups followed by the isolation of biotinylated proteins on
streptavidin. The purified biotinylated proteins are either eluted from streptavidin and
digested with trypsin or directly digested on the streptavidin-coated resin. Resulting
peptides are separated by microcapillary HPLC on a C18 column, followed by the
analysis of the eluted fractions with matrix assisted laser desorption ionization mass
spectrometry (MALDI-TOF MS). The resulting data is used to establish a two-
dimensional peptide map (2D peptide map) reporting the HPLC fractions on the y-
axis and the m/z ratios of the measured peptides on the x-axis. The mass peaks signal
intensities are displayed by means of a grayscale. The grayscale is standardized to an
internal standard peptide, which is added to each chromatographic fraction in a known
amount prior to the MALDI-TOF MS experiment. The 2D peptide map allows the
immediate appreciation of the entire mass spectrometric profile of the
chromatographic fractions, but allows also a comparison analysis of the same HPLC
fraction derived from two different samples (e.g., control and treatment). Interesting
fractions are used for subsequent tandem mass spectrometry measurements in order to
identify differentially expressed proteins.
It is almost certain that modern mass spectrometric procedures and instrumentation
[(such as the use of Fourier transform ion cyclotron (FT-ICR) and MALDI-TOF time
of flight (MALDI-TOF-TOF) spectrometers (Aebersold and Mann 2003)], which
offer unprecedented resolution and sensitivity, will contribute to the increased use of
mass spectrometry-based methods for gel-free proteomic analysis. However, the study
of membrane proteins will also require improved methodologies for the chemical
modification and recovery of peptides from these low abundant, hydrophobic
proteins.
45
2.6.3. Ligand based methods
Traditionally, the first markers of angiogenesis have been discovered either by limited
proteolysis of purified protein preparations (Zardi, Carnemolla et al. 1987), or by
animal immunization with biological samples derived from tumors, followed by an
extensive immunohistochemical analysis of the resulting hybridomas (Liu, Moy et al.
1997; Chang, O'Keefe et al. 1999). The introduction of recombinant antibody
technologies, and in particular of antibody phage technology (Winter, Griffiths et al.
1994), has greatly facilitated the production of good-quality monoclonal antibodies
without immunization. These technologies are particularly efficient when pure antigen
preparations are available (Viti, Nilsson et al. 2000), but antibodies have also been
generated from phage libraries against “difficult” antigens (Hoogenboom, Lutgerink
et al. 1999).
It is difficult to imagine that antibody-based chips may facilitate the study of the
relative abundance of membrane proteins in different biological specimens (Elia,
Silacci et al. 2002). Nonetheless, if larger public-domain collections of monoclonal
antibodies become available in the future, it should be possible to use fluorescence-
activated cell sorters (FACS) and/or immunohistochemistry with tissue arrays
(Schraml, Kononen et al. 1999) for the relative quantization of membrane proteins in
different cells/tissues.
Ruoslahti, Pasqualini and co-workers have pioneered the in vivo biopanning of
peptide phage libraries, in an attempt to identify binding specificities against different
vascular addresses in different tissues and/or tumors (Pasqualini and Ruoslahti 1996;
Rajotte, Arap et al. 1998). Among others, peptides specific to integrins and to CD13
were identified with this procedure. However, the real potential of this technology
46
remains to be demonstrated, considering that the use of peptides on tissue sections is
often less efficient than the use of antibodies (which normally display a higher affinity
for the antigen), and in the absence of quantitative biodistribution studies and clinical
studies with purified preparations of the vascular-targeting peptides.
2.6.4. Application of proteomic methods to the identification of new targets for tumor
therapy
As already mentioned above, a number of studies have been carried out by proteomic
profiling in order to discover novel targets for therapy of cancer. While a
comprehensive review of the studies carried out would be beyond the scope of this
section, a couple of examples of proteomic investigations comparing malignant and
nonmalignant cell lines are worth mentioning.
Enhanced levels of urokinase plasminogen activator (uPA) and urokinase
plasminogen activator receptor (uPAR) are possibly the strongest indicator of poor
prognosis in mammary carcinoma (Ganesh, Sier et al. 1994; Duffy 2002). By
transfecting a highly metastatic colon carcinoma cell line with an antisense 5’uPAR
cDNA fragment, Ahmed and colleagues (Ahmed, Oliva et al. 2003) silenced the cell
surface expression of uPAR, obtaining a regression of the metastatic phenotype. A
proteomic comparison of the uPAR-silenced cell line and their wild type counterpart
allowed the identification of more that 300 proteins modulated by uPAR silencing,
some of which could represent valuable targets for metastasis inhibition.
Cravatt and colleagues (Jessani, Humphrey et al. 2004) carried out a functional
proteomic analysis, using a suite of activity-based chemical probes, on the human
breast cancer line MDA-MB-231 grown either in culture or as orthotopic xenografts
47
in the mammary fat pad of immunodeficient mice. Cells isolated from tumors
exhibited profound differences in their enzyme activity profile compared with the
parental cell line, including the dramatic posttranscriptional up-regulation of uPA and
of tissue plasminogen activator and down-regulation of the glycolytic enzyme
phosphofructokinase.
uPA and uPAR appear then to be interesting targets for molecular intervention, but
the pharmaceutical development of inhibitory molecules (either low-molecular weight
compounds or therapeutic antibodies) is still in its infancy. Strategies based on
antisense vectors, siRNA (Arens, Gandhari et al. 2005), as well as linear and cyclic
peptides (Magdolen, Burgle et al. 2001; Ploug, Ostergaard et al. 2001) have been
proposed. Furthermore, efforts have been made to raise antibodies against different
components of the urokinase complex to limit tumor progression. However, the
potential of these antibodies may not have been fully exploited because of insufficient
knowledge about the localization of the epitopes. Recently, experiments aimed at
mapping the epitopes for a series of monoclonal antibodies to uPA, directed against
either the kringle or the serine protease domain have shown that different
functionalities of the enzyme may be modulated by interfering with the appropriate
epitope (Petersen, Hansen et al. 2001).
2.7. From in vitro to in vivo model systems for the identification of vascular
targets
48
In principle, the most direct way to assess differences in protein abundance between
the tumor endothelium and the normal endothelium would consist in the in vivo
labeling of vascular structures, followed by rapid recovery and comparative proteomic
analysis of the proteins in the two samples.
The group of Jan Schnitzer has pioneered the use of colloidal silica for the in vivo
coating of vascular structures in tumors and in normal organs (Jacobson, Schnitzer et
al. 1992; Czarny, Liu et al. 2003). This physical modification allows the recovery (by
centrifugation and fractionation) of silica-coated structures (luminal cell plasma
membranes and caveolae of the endothelium), providing ideal material for proteomic
investigations, for example by immunization (McIntosh, Tan et al. 2002) or by 2D-
PAGE. Combining the colloidal silica based subcellular fractionation procedure with
subsequent differential proteomic analysis of luminal endothelial cell plasma
membranes and caveolae isolated from normal rat organs or tumors, Oh and
colleagues discovered aminopeptidase-P and annexin A1 as selective in vivo targets
for antibodies in lungs and solid tumors, respectively (Oh, Li et al. 2004).
De la Fuente et al. have described the artificial perfusion of lungs isolated from
normal and hyperoxic rats with Sulfo-NHS-LC-biotin (De La Fuente, Dawson et al.
1997). After SDS-PAGE, the biotinylated proteins were visualized using a
chemiluminescence substrate for the streptavidin-horseradish peroxidase conjugate,
outlining differences in rats exposed to hyperoxia for 48-60 hours.
Our group is using the terminal perfusion of tumor-bearing mice with sulfo-NHS-LC-
biotin solution as an avenue for the in vivo covalent modification of amine-containing
phospholipids and proteins, which are accessible to the reagent during the perfusion
(Rybak, Scheurer et al. 2004; Rybak, Ettorre et al. 2005). After anesthesia, mice are
first perfused with saline solution, to remove circulating cells, proteins and other
49
primary-amine containing compounds. Few minutes later, perfusion is continued with
an aqueous solution of sulfo-NHS-LC-biotin, followed by a primary amine (e.g., Tris
buffer) to quench unreacted ester derivatives of biotin. This methodology leads to
reliable and efficient labeling of accessible structures (mainly vascular structures) in
vivo, and is ideally suited for proteomic investigations. The resulting biotinylated
proteins (or the corresponding peptides generated by endoproteolytic cleavage) can be
purified on streptavidin-coated resins in the presence of SDS. However, the choice of
detergent and of purification protocol depends on the experimental strategy chosen for
proteomic investigations. In principle, the in vivo biotinylation method presents
several attractive features, as it allows a direct investigation of those accessible
targets, which are likely to be amenable to targeted anticancer imaging and
therapeutic strategies. As protein biotinylation lends itself not only to purification
strategies (special precautions for elution must be chosen, considering the high-
affinity interaction with streptavidin) (Rybak, Scheurer et al. 2004), but also to
biochemical analysis (e.g., by blotting or microscopic analysis with streptavidin-based
detection reagents), it is possible to monitor the efficiency of the biotinylation
reaction in various organs, prior to proteomic analysis.
50
3. Results & Discussion
3.1. Modulation of gene expression by extracellular pH variations in human
fibroblasts. A transcriptomic and proteomic study.
Physiological variations of pH values are observed during development (Baltz 1993;
Gilbert 2000), in physiological processes and in disease (e.g., tumor cells often create
an acidic extracellular environment, but display basic intracellular pH as a
consequence of increased metabolic activity) (Stubbs, McSheehy et al. 2000).
Intracellular and extracellular pH values control the proliferation of fibroblasts, their
response to growth factors and their processing of primary transcripts of extracellular
matrix components (Borsi, Allemanni et al. 1996). For example, the pattern of
alternative splicing of tenascin-C (TN-C) is tightly regulated by the pH at which
primary fibroblast cell cultures are grown. While in the pH range 6.7 – 6.9 the small
isoform of TN-C is preferentially expressed, at pH values above 7.2 the large TN-C
isoform (containing 8 extra fibronectin type-III homology repeats) becomes
predominant (Borsi, Balza et al. 1995). The large TN-C isoform is an oncofetal
marker, which is over-expressed in a variety of solid tumors and which is the target of
monoclonal antibodies being used in anticancer therapeutic strategies (Paganelli,
Bartolomei et al. 2001; Reardon, Akabani et al. 2002). Interestingly, malignantly
transformed fibroblasts predominantly express the large TN-C isoform independent of
the culture medium pH, since these cells are able to maintain a basic intracellular pH
under a broad range of culture conditions (Borsi, Allemanni et al. 1996).
Changes in pH are known to regulate gene expression in different cell types
(Petronini, Alfieri et al. 1995; Shi, Le et al. 2001). However, little is known about the
51
regulation of gene expression in fibroblasts as a function of pH fluctuations, such as
those occurring in physiological and pathological processes.
In this thesis, we have performed a comprehensive analysis of the patterns of gene
expression in normal human dermal fibroblasts (NHDF) at two different pH values (in
the presence and in the absence of serum), with the aim of getting a deeper insight in
the regulation of the transcriptional program of fibroblasts as a response to pH
change. Primary fibroblast cell cultures were chosen as model system, since these
cells adapt their intracellular pH on the basis of the pH of the culture medium (Austin
and Wray 1993; Mellergard, Ou-Yang et al. 1994; Borsi, Allemanni et al. 1996), and
since they are responsible for the synthesis of a large portion of ECM components in
health and disease. Furthermore, since our interest mainly lies in the identification of
components of the ECM, which are aberrantly expressed in the tumor environment,
we have also used 2D-PAGE and mass spectrometry to identify proteins, which are
differentially secreted by NHDF cultured at pH 6.7 and 7.5. In the past, our group (in
collaboration with the group of L. Zardi, Genova, Italy) has used components of the
modified ECM, whose expression is regulated by pH, as targets for antibody-based
anti-cancer strategies (Birchler, Neri et al. 1999; Carnemolla, Castellani et al. 1999;
Tarli, Balza et al. 1999; Viti, Tarli et al. 1999; Nilsson, Kosmehl et al. 2001;
Carnemolla, Borsi et al. 2002; Halin, Rondini et al. 2002), which are currently being
investigated in clinical studies.
3.1.1. Genome-wide analysis of mRNA levels in NHDF cultured at different pH
We have aimed at performing a genome-wide analysis of the modulation of gene
expression in fibroblasts, as a consequence of a change in the pH of the culture
medium, in the presence or absence of serum.
52
The decision of including serum starvation among the variables of our experimental
design was stimulated by the fact that acidosis and nutrient starvation are often
associated in physiological and pathological processes. Furthermore, when studying
the pattern of alternative splicing of the TN-C transcript in fibroblasts as a function of
culture pH, we had observed (as previously reported by Luciano Zardi) that the
Figure 1: Northern Blot analysis of the expression of TN-C and GAPDH in NHDF cultured at pH 6.7 or 7.5, in the presence or absence of serum. Zardi and coworkers had previously reported that modulations of the extracellular pH have an effect on the pattern of alternative splicing of transcripts coding for TN-C in NHDF in in vitro culture conditions (Borsi, Balza et al. 1995; Borsi, Allemanni et al. 1996). When cultured at pH 7.5, normal fibroblasts mainly express a “large” form of the transcript, of about 8 kb, coding for a protein isoform containing 14.5 EGF-like and 16 FN type III repeats. With increasing acidity of the culture medium, cells begin to express also a “small” form of the transcript, of about 6.8 kb, which becomes predominant at pH 6.7-6.6. pH appeared to control the pattern of splicing for cells grown both in the presence or in the absence of serum (Borsi, Allemanni et al. 1996). However, serum stimulation of quiescent fibroblasts had been reported to
result in a dynamic modulation of the patterns of alternative splicing of the primary TN-C transcript (Borsi, Balza et al. 1994). We have reproduced the findings described above, observing that the large isoform of TN-C is preferentially expressed in NHDF grown at basic pH. However, we found that the removal of serum reduced the expression of the large TN-C isoform, independent of culture pH. large TN-C isoform is preferentially expressed at basic pH (Figure 1). However, we
had also found that the removal of serum from the culture medium reduced the
expression of the large TN-C isoform, independent of the culture pH.
The analysis of the effect of pH on fibroblast gene expression in the presence or the
absence of serum was performed according to the experimental scheme depicted in
Figure 2. Fibroblasts were grown at pH 7.5 in DMEM medium containing fetal
bovine serum (FBS) until confluence. At that point, replicate flasks were washed and
the medium replaced with fresh DMEM, buffered at pH 6.7 or 7.5, in the presence or
absence of 10% FBS. After changing the medium to the two different pH values, the
53
Figure 2: Experimental design.
t the end of the incubation time, cells were harvested and used for total RNA extraction. Media collected after 72 ours incubation of NHDF at pH 6.7 or 7.5 in the absence of FBS were concentrated 800-fold by ultrafiltration and
used for 2D-PAGE analysis.
confluent cells did not change morphology, and for both conditions, cell mortality was
constantly below 5%. At different time points, total RNA was extracted and used for
the synthesis of biotinylated cRNA, in order to allow the quantization of gene
expression with the Affymetrix technology (Lipshutz, Fodor et al. 1999; Lipshutz
2000)
In a first experimental set up, we have compared the levels of transcripts in NHDF
cells grown in the presence of 10% FBS at different time points (24, 48 and 72 hours)
Replicate flasks were seeded with NHDF and cells were grown until confluence. Monolayers were washed (t = 0) and exposed for 24, 48 or 72 hours to DMEM buffered at pH 6.7 or 7.5, in the presence or absence of 10% FBS. Ah
54
Figure 3: Comparative analysis of NHDF gene expression modulation as a consequence of cell exposure to acidic medium. Triplicate, independent RNA preparations were obtained from NHDF grown at pH 6.7 for 24, 48 and 72 hours or at pH 7.5 for 24 hours, in the presence of serum. Using the Affymetrix oligonucleotide microarray technology, a measurement of expression levels for 12,565 genes (represented on the Affymetrix HG-U95A ver 2 gene chip) was carried out. Comparative analysis of gene expression levels in NHDF exposed to pH 6.7 vs. pH 7.5 was performed by GeneSpring software. Replicates were averaged, and a per gene normalization was carried out, meaning that the average values of gene expression in samples of fibroblasts grown at pH 6.7 were used to calculate (for each gene and experimental condition) ratios relative to the average of the expression levels of the same gene in samples of fibroblasts grown at pH 7.5 for 24 hours. 2,068 out of 12,565 genes resulted to be modulated by more than two-fold (up-regulation or down-regulation) in at least one of the three time points. Figure 3A presents a hierarchical clustering of such 2,068 genes, according to the method of Eisen (Eisen, Spellman et al. 1998). Genes are grouped based on the “similarity” of their expression pattern along the experimental time dimension and are shown as horizontal bars of different colors. The similarity tree is displayed in the left portion of the figure. The green-to-red color grading shown on the right-hand side of panel A represents the ratios of gene expression levels for the different time points of NHDF cells grown at pH 6.7, relative to the corresponding gene expression levels at pH 7.5. Figure 3B presents the K-means clustering (Dougherty, Barrera et al. 2002) of the same genes in 7 classes, according to the profile of up-regulation or down-regulation of gene expression as a function of time. A comprehensive list of the expression levels can be found in the supplementary information [Supplementary Table 1, available at http://www.pharma.ethz.ch/bmm/div/NHDF/].
55
after changing the pH medium to pH 6.7, relative to control cells maintained at pH
7.5.
At each time point, the absolute expression levels for the more than 12,000
represented genes were measured using the Affymetrix Micro Array Suite 2.0
software, starting from NHDF cell preparations performed in triplicate. The
comparative analysis of gene expression was performed using the Silicon Genetics
GeneSpring version 4.2.1 software, as described in the Materials and Methods
section.
2,068 out of 12,565 genes resulted to be modulated by more than two-fold (up-
regulation or down-regulation) in at least one of the three time points. Figure 3A
presents a hierarchical clustering of such 2,068 genes, automatically performed by the
GeneSpring software according to the method of Eisen (Eisen, Spellman et al. 1998).
Genes are grouped on the basis of the “similarity” of their expression pattern along
the experimental time dimension. The similarity tree is displayed in the left portion of
the figure. A green-to-red color grading represents the ratios of gene expression levels
for the different time points of NHDF cells grown at pH 6.7, relative to the
corresponding gene expression levels at pH 7.5 (i.e., ratio > 1 for enhanced gene
expression at acidic pH). Figure 3B presents the K-means clustering (Dougherty,
Barrera et al. 2002) of the same genes in 7 classes, according to the profile of up-
regulation or down-regulation of gene expression as a function of time. A
comprehensive list of the expression levels can be found in Supplementary Table 1,
[available at the website http://www.pharma.ethz.ch/bmm/div/NHDF/]. Sixty-seven
out of 2,068 genes showed a modulation of expression > 5-fold at 72 hours (the ratio
being < 0.2 or > 5).
56
In Figure 4, genes showing a modulated expression pattern were grouped into eight
4E), DNA-binding proteins/transcription factors (Fig. 4F), receptors (Fig. 4G) and
ligands (Fig. 4H).
At pH 6.7, strongly over-expressed genes of ECM components include microfibril-
associated glycoprotein 4 (Zhao, Lee et al. 1995), EFEMP-1 [a fibulin-like protein
containing five EGF-like domains (Ikegawa, Toda et al. 1996)] and UDP-GAL:β-
GLCNAC: β-1,3-galactosyl transferase polypeptide 4 (Amado, Almeida et al. 1998).
Down-regulated genes include the matrix metalloproteases 12 [macrophage elastase,
(Belaaouaj, Shipley et al. 1995)] and 3 [stromelysin 1 (Wilhelm, Collier et al. 1987)],
the elastin precursor (Rosenbloom, Bashir et al. 1991) and the prolyl 4-hydroxylase
alpha (II) subunit (Annunen, Helaakoski et al. 1997) (Figure 4A).
An overall reduction of expression of cell cycle regulator genes, like cyclins A2, B1,
B2, E2, and CDC2 (Furukawa 2002) as well as an increase in the levels of cyclin-
dependent kinase inhibitor p57KIP2 (Lee, Reynisdottir et al. 1995), correlates with
the observation that the switch to a more acidic environment promotes an arrest in
NHDF cell cycle progression (Borsi, Allemanni et al. 1996) (Figure 4B). A decrease
in expression of survival factors, such as survivin (Ambrosini, Adida et al. 1997) and
increased levels of the apoptosis-specific protein ASP (Grand, Milner et al. 1995) are
counterbalanced by the concomitant increase in the expression of HIAP-1 (Liston,
Roy et al. 1996) and of an isoform of caspase-like apoptosis regulator protein 2
(Inohara, Koseki et al. 1997) (Figure 4C). Some genes encoding for growth factor
57
Figure 4: Clustering of NHDF genes modulated at pH 6.7. A simplified gene ontology was automatically built up by the GeneSpring software and the 2,068 genes whose expression is modulated by exposure to acidic medium were clustered accordingly. In the figure, the time course (24, 48 and 72 hours) of expression of genes coding for proteins of the extracellular matrix (A), for cell-cycle (B) and apoptosis (C) regulators, for growth factor receptors (D) and growth factor receptor ligands (E), for DNA-binding proteins and transcription factors (F) or for receptors (G) and ligands (H) are shown, together with the relative Affymetrix gene identifier and the gene systematic name (GeneSpring, Silicon Genetics, Redwood City, CA, USA; http://www.sigenetics.com). Genes that are mentioned in the text are outlined in boldface. In the lower part of the figure, a green-to-red color grading represents the ratios of gene expression levels for the different time points of NHDF cells grown at pH 6.7, relative to the corresponding gene expression levels at pH 7.5. Further information about the indicated genes, as well as the sequences of oligonucleotides employed by Affymetrix for recognition of the different genes is available at the Affymetrix website (http://www.affymetrix.com/analysis/index.affx)
58
Figure 4: Clustering of NHDF genes modulated at pH 6.7 (Continued).
receptors show an increase in their expression level at acidic pH, namely a platelet-
derived growth factor receptor-like coding gene (Fujiwara, Ohata et al. 1995) and a
growth factor inducible nuclear protein N10 (Chang, Kokontis et al. 1989) (Figure
59
4D). Among the growth factors and growth factor-related genes (Figure 4E), a
markedly increased expression was observed for the insulin-like growth factor
binding proteins (IGFBP) 2 (Agarwal, Hsieh et al. 1991) and, to a lower extent, for
IGFBP-5 (Allander, Larsson et al. 1994) and the keratinocyte growth factor (Rubin,
Osada et al. 1989). A reduction in expression was observed, in contrast, for heregulin
[glial growth factor 2 (Holmes, Sliwkowski et al. 1992)], for basic fibroblast growth
factor (Abraham, Whang et al. 1986) and for the insulin-like growth factor binding
protein 3 (Ferry, Cerri et al. 1999), this last, however, being expressed at very high
absolute levels in both NHDF cultured at pH 6.7 and pH 7.5. A variety of modulated
genes can be observed among DNA-binding proteins/transcription factors (Fig. 4F),
receptors (Fig. 4G) and ligands (Fig. 4H). One of the strikingly over-expressed genes
at pH 6.7 is stanniocalcin 1 (Olsen, Cepeda et al. 1996), a calcium and phosphate
homeostasis regulating hormone (Figure 4H).
3.1.2. Modulation in the expression of ECM components in NHDF after serum
starvation and/or pH change
The Affymetrix gene chip system was also used to investigate whether the modulation
of gene expression, following a change of the culture medium pH, was influenced by
the concomitant removal of serum. Our analysis was mainly focused on ECM
components, since those which display a restricted pattern of expression in healthy
tissues but not in disease may provide excellent targets for biomolecular intervention
(Birchler, Neri et al. 1999; Carnemolla, Castellani et al. 1999; Tarli, Balza et al. 1999;
Viti, Tarli et al. 1999; Nilsson, Kosmehl et al. 2001; Carnemolla, Borsi et al. 2002;
Halin, Rondini et al. 2002). However, a complete list of modulated genes is provided
in Supplementary Table 2 [http://www.pharma.ethz.ch/bmm/div/NHDF/]. The most
60
Figure 5: Effect of change of culture medium pH, of serum starvation or of their combination on ECM components gene expression levels. Levels of expression of genes coding for components of the ECM in NHDF cultured at pH 6.7 for 24, 48 or 72 hours in the absence of serum were compared to NHDF cultured at pH 7.5 for 24 hours in the presence (A) or absence (B) of serum. Panel C shows the effect of serum starvation on NHDF grown for 24, 48 or 72 hours at pH 7.5 in absence of serum, relative to control NHDF cultures, grown at pH 7.5 for 24 hours in the presence of serum. On the right part of each panel, the Affymetrix gene identifiers and gene systematic names are shown. Genes that are mentioned in the text are outlined in boldface. In the lower part of the figure, a green-to-red color grading of gene expression ratios is included.
61
striking modulations of gene expression observed for the simultaneous acidification of
the culture pH and serum withdrawal (Figure 5A), were also observed in fibroblasts
kept at pH 7.5 but deprived of serum (Figure 5C), indicating that serum starvation
dominates these transcriptional programs. The most notable up-regulations of gene
expression in both Figure 5A and 5C were observed for microfibril-associated
glycoprotein 4 (Zhao, Lee et al. 1995), collagen type XIV α1 [undulin, (Bauer,
Dieterich et al. 1997)], matrix metalloproteinase 11 [stromelysin 3, (McDonnell and
Matrisian 1990)] and extracellular matrix protein 2 (Nishiu, Tanaka et al. 1998), while
elastin precursor (Rosenbloom, Bashir et al. 1991), matrix metalloproteases 12
[macrophage elastase, (Belaaouaj, Shipley et al. 1995)] and cartilage oligomeric
matrix protein precursor (Newton, Weremowicz et al. 1994) were consistently down-
regulated. Interestingly, collagen type XIV α1 expression was strongly down
regulated upon acidification of a serum-free medium (Figure 5B), indicating that
serum starvation and pH change had opposite effects on the control of transcriptional
activity for this gene. By contrast, expression of dermal fibroblast elastin precursor
(Uitto, Christiano et al. 1991) was up-regulated upon acidification of a serum-free
medium (Figure 5B), but was strongly down-regulated by FBS removal of a serum-
rich medium, independent of culture pH (Figure 5A and 5C).
Some genes displayed a strongly modulated gene expression upon serum starvation at
pH 7.5 (Figure 5C), which was abolished by pH change (Figure 5A), collagen XV α1
(Muragaki, Abe et al. 1994) being a notable example.
3.1.3. 2D-PAGE analysis of secreted proteins expressed by NHDF cultured at
different pH
62
In order to study the modulation in the expression of proteins secreted by fibroblasts,
at the protein level, as a consequence of the acidification of the culture medium and
serum starvation, we adopted the following experimental approach. NHDF were
grown until confluence in complete DMEM medium. Cells were then incubated for
72 hours in serum-free DMEM at two different pH values (6.7 or 7.5). The resulting
supernatant, containing the proteins secreted by the fibroblasts, was collected at the
end of incubation time and concentrated 800-fold by ultrafiltration. The proteins were
Figure 6: 2D-PAGE of proteins secreted by NHDF at pH 6.7 and at pH 7.5. Concentrated supernatants from NHDF grown for 72 hours in DMEM FBS-free at pH 6.7 (A) or at pH 7.5 (B), were subjected to 2D gel electrophoresis as described in Experimental Procedures. Two representative gels are shown, in which the IPG strips pH range was 4-7. Molecular weight markers were included and their values are indicated on the left. Red circles/ellipses on both gels indicate those proteins whose expression is apparently increased at the corresponding pH value. Blue squares/rectangles indicate those proteins that were poorly modulated by a pH change. All the spots indicated by a circle/ellipse or by a square/rectangle were excised from the gel, trypsin-digested and subjected to mass spectrometric analysis. Numbering of the spots indicates the proteins that could be identified and refers to Table 1 (see text). Asterisks mark the proteins for which an unambiguous identification could not be obtained. Spots indicated with roman numbers correspond to serum contaminants. A selected, magnified area of gels from two replicate, independent samples at pH 6.7 (C, D) or at pH 7.5 (E, F) shows the reproducibility of the experimental setup. The position of the selected area is indicated in A and B by a dotted black square. Arrows in E and F point to two relevant protein spots (lamin A and tetranectin), over expressed in NHDF grown at pH 7.5. While most spots on the gels correspond to ECM components, some spots correspond to intracellular proteins, originating from cell lysis.
63
then separated by 2D-PAGE, using both wide range (3-10) and narrow range (4-7)
IPG strips. The resulting gels were then stained, scanned and compared. Relevant
protein spots (showing either constant or modulated expression) were cut and
identified using mass spectrometry.
The patterns of proteins secreted by NHDF at pH 6.7 and at pH 7.5 are shown in
Figure. 6 A and B, respectively. Spots showing a differential pattern of expression at
the two pH values are indicated with a red circle. All the experiments were repeated at
least twice and the overall 2D-PAGE spot pattern was found to be reproducible in the
different conditions, with only minor gel-to-gel variations (Figure 6 C-F).
In total, more than 650 spots were cut out of 2D gels, trypsin-digested and measured
using an LCQDeca ion trap µLC-MS/MS mass spectrometer. More than 55% (365)
could be identified. Some of them were identified only with a single peptide, but with
a good correlation coefficient by the SEQUEST algorithm. Most of these
identifications were later confirmed by measurements of the same spot in a duplicate
gel. In general, we found rarely more than one protein per spot, with the exception of
some contamination by trypsin and keratin peptides.
Table 1 shows a list of the identified proteins, with the SwissProt accession numbers,
indicating whether their expression at the two different pH values is modulated.
For many proteins, several isoforms were identified, with small differences in
molecular weight and broad pI variations (e.g. collagen VI α1, PEDF, cathepsin B).
In spite of the fact that the 2D-PAGE analysis reflected an accumulation of protein
from transcripts produced at different time points, a number of clear differences
hybridization; (Seta, Kim et al. 2001)) and confirmed by in situ hybridization
(Budanov, Shoshani et al. 2002). Furthermore, a set of genes which are under the
73
transcriptional control of HIF-1α (the main oxygen sensor within the cells) have been
identified over the last few years (Semenza 2002).
Using the most recent Affymetrix gene chips (22,525 genes) and proteomic
techniques, we have performed a genome-wide analysis of gene expression in human
umbilical cord vein endothelial cells (HUVEC), cultured in hypoxic or normoxic
conditions. Our study confirms previous findings about hypoxic regulation of gene
expression in endothelial cells, but also identifies additional proteins and ESTs whose
expression is modulated (most often, up regulated) by a decreased oxygen
concentration. This transcriptomic analysis has been complemented by a semi-
quantitative RT-PCR analysis of patterns of expression of alternative splicing for
selected domains of extracellular matrix proteins and by a 2D-PAGE analysis of
proteins expressed by HUVEC cells, cultured in hypoxic or normoxic conditions.
3.2.1. Genome-wide analysis of gene expression modulation in HUVEC cultured in
hypoxic vs. normoxic conditions
We have performed a broad-range analysis of the modulation of gene expression in
HUVEC, as a function of time of exposure to hypoxic growth conditions, according to
the experimental scheme depicted in Figure 7. We have compared the levels of
transcripts in HUVECs grown in hypoxia at different time points (12, 24, and 48
hours), relative to control cells maintained for 48 hours in normoxic growth
conditions.
At each time point, the absolute expression levels for the more than 22,000 genes
represented in the Affymetrix Chips were measured using the Affymetrix Micro
Array Suite 2.0 software, starting from cell preparations performed in triplicate.
74
Figure 7: Experimental design. For transcriptomic experiments, replicate flasks were seeded with HUVECs and cells were grown until confluence. Monolayers were washed (t = 0) and exposed for 12, 24 or 48 hours to hypoxia (5% CO2, 2% O2) in EGM-2 medium (containing 10% FBS, pH 7.5) or exposed for 48 hours to normoxia (5% CO2, 21% O2) in the same medium. At the end of the incubation time, cells were harvested and used for total RNA extraction. For proteomic experiments, HUVECs were exposed to either hypoxia or normoxia for 48 hours, in the same conditions as described above. At the end of the incubation time, cells were harvested with a cell scraper and directly processed for 2D-PAGE analysis (whole cell lysate) or subjected to further purification procedures (cell fractionation).
The comparative analysis of gene expression was performed using the Silicon
Genetics GeneSpring version 4.2.1 software, as described in the Materials and
Methods.
3,996 out of 22,525 genes resulted to be modulated by hypoxia more than two-fold
(up-regulation or down-regulation) in at least one of the three averaged time points
(data not shown). A comprehensive list of the expression levels for these genes can be
found at http://www.pharma.ethz.ch/bmm/div/HUVEC/ as Supplementary Table 1.
75
Figure 8: Comparative analysis of HUVEC gene expression modulation as a consequence of cell exposure to hypoxia. Triplicate, independent RNA preparations were obtained from HUVEC grown for 12, 24 or 48 hours in hypoxic conditions or for 48 hours in normoxic conditions. Using the Affymetrix oligonucleotide microarray technology, a measurement of expression levels for 22,525 genes (represented on the Affymetrix HG-U133A gene chip) was carried out. Comparative analysis of gene expression levels in HUVEC exposed to hypoxia vs. normoxia was performed by GeneSpring version 4.2.1 software. Replicates were averaged, and a per gene normalization was carried out, meaning that the average values of gene expression in samples of HUVEC grown in hypoxia were used to calculate (for each gene and experimental time point) ratios relative to the average of the expression levels of the same gene in samples of HUVEC grown in normoxia for 48 hours. Sixty-five out of 22,525 genes resulted to be more than five-fold up regulated in at least one of the three time points of exposure to hypoxia (panel A). Forty-two genes resulted to be more than five-fold down regulated in at least one of the three time points of exposure to hypoxia (panel B). Figure 8A and 8B present a hierarchical clustering, according to the method of Eisen (Eisen, Spellman et al. 1998), of such 65 up-regulated and 42 down-regulated genes, respectively. Genes are grouped based on the “similarity” of their expression pattern along the experimental time dimension and are shown as horizontal bars of different colors. The similarity tree (dendrogram) is displayed in the left portion of each panel. The green-to-red color grading shown on the right-hand side of the Figure represents the ratios of gene expression levels for the different time points of HUVEC cells grown in hypoxia, relative to the corresponding gene expression levels in normoxia.
In order to reduce further the complexity of information, we performed a detailed
analysis, applying different filtering strategies. The first strategy was to focus on those
genes showing the greatest degree of modulation between the two conditions (hypoxia
and normoxia) and which were called “present” in at least one of them. Figure 8
presents a hierarchical clustering of such genes, performed by the GeneSpring
76
software according to the method of Eisen (Eisen, Spellman et al. 1998). Genes are
grouped on the basis of the “similarity” of their expression pattern along the
experimental time dimension. The similarity tree is displayed in the left portion of
each panel. A green-to-red color grading represents the ratios of gene expression
levels for the different time points of HUVEC cells grown in hypoxia, relative to the
corresponding gene expression levels in normoxic conditions. Panel A shows
hierarchical clustering of genes that resulted to be up regulated more than 5-fold (and
flagged as present) in at least one of the experimental time points (65 genes, ratio
hypoxia/normoxia > 5), while Panel B shows hierarchical clustering of genes that
resulted to be more than 5-fold down-regulated (42 genes, ratio hypoxia/normoxia
<0.2).
Fifty-two of the 65 strongly up-regulated genes were annotated genes, the rest being
ESTs and hypothetical proteins. Many of the annotated genes comprised in Fig. 8A
have already been shown to be up regulated by exposure to hypoxia, in HUVEC cells
(adrenomedullin (Ogita, Hashimoto et al. 2001), VEGFR1 (Gerber, Condorelli et al.
1997)) or in other cell model systems. Other genes were also strongly up regulated
whose expression however, to our knowledge, has never been shown to be pO2-
dependent (claudin 3, CD24, tetranectin and so on). A few genes were found to be
significantly up regulated in hypoxic growth conditions at all time points, namely the
chemokine C-X-C receptor 4, the pyruvate dehydrogenase kinase 1, the myostatin and
the taurine transporter (solute carrier family 6 member 6), many other genes were up
regulated only at late times of exposure to hypoxia (e.g., adrenomedullin, adenylate
In hypoxia, strongly over-expressed genes belonging to ECM or ECM-related
proteins (Fig. 9 B) include collagenVα1, both procollagen proline 4-hydroxylase
alpha-polypeptides 1 and 2, lysine hydroxylase 1 and 2, matrix metalloproteinase 16.
Down-regulated genes include matrix metalloproteinases 17, stromelysin,
microfibrillar-associated protein 2 and collagen XIII α1.
At early times of exposure to hypoxic condition, the modulation of apoptosis-related
genes (Fig. 9 C) in HUVEC seems to point to an overall anti-apoptotic effect of
hypoxia that, however, is lost at later times. This pattern is mirrored in the transient up
regulation of expression of cell cycle (Fig. 9 D) genes, like CDC2, CDC20, CDC25c
and CDC6, observed at 12 hours of hypoxic growth that is followed by a significant
decrease to severely down regulated levels for almost the totality of these genes.
Similar is also the behavior of a great majority of cell-cycle related (Fig. 9 E)
78
Figure 9: Clustering of HUVEC genes modulated in hypoxic conditions. A simplified gene ontology was automatically built up by the GeneSpring software and applied to the analysis of the 3,996 genes whose expression resulted to be modulated by more than two times in up- or down regulation upon exposure of HUVEC cells to hypoxia. For a complete list of the modulated genes, see Supplementary table 1 at the website http://www.pharma.ethz.ch/bmm/div/HUVEC/). In the figure, the time course (12, 24 or 48 hours) of expression of genes coding for membrane proteins (Fig. 3A), extracellular matrix proteins (Fig. 9B), apoptosis regulators (Fig. 9C), cell cycle proteins (Fig. 9D) and cell cycle regulators (Fig. 9E), receptors (Fig. 9F), ligands (Fig. 9G), and DNA-binding proteins/transcription factors (Fig. 9H) are shown, together with the relative Affymetrix gene identifier and the gene systematic name (GeneSpring, Silicon Genetics, Redwood City, CA, USA; http://www.sigenetics.com). In the lower part of the figure, a green-to-red color grading represents the ratios of gene expression levels for the different time points of HUVEC cells grown in hypoxia, relative to the corresponding gene expression levels in normoxic HUVEC. Further information about the indicated genes, as well as the sequences of oligonucleotides employed by Affymetrix for recognition of the different genes is available at the Affymetrix website (http://www.affymetrix.com/analysis/index.affx)
79
Figure 3: Clustering of HUVEC genes modulated in hypoxic conditions (Continued).
genes, including cyclins A1, A2, B1, B2, D1, E2, F, and the cyclin-dependent kinase
inhibitor p18, which are up regulated at early times of hypoxic growth, but become
severely down regulated thereafter. A notable exception is constituted by the
80
prostacyclin synthase gene, which is moderately affected by hypoxia at early times of
exposure, but becomes dramatically up regulated at 24 and 48 hours of HUVEC
incubation in hypoxic conditions.
A few genes, among those encoding for receptors (Fig. 9 F), showed a marked
increase in their expression level in hypoxic conditions, namely FLT/VEGFR1 and
the chemokine C-X-C receptor 4 (fusin). Ligands (Fig. 9 G) which resulted to be most
prominently up regulated by hypoxia included VEGF, Del-1, TGFβ-induced 68 kDa
protein, IGFBP3 and secreted frizzled-related protein 1.
A range of genes among DNA-binding proteins/transcription factors (Fig. 9 H) were
considered as “present” in the MAS 2.0 absolute analysis, but few of them showed a
significant modulation upon shift to hypoxic growth conditions. Basic leucine zipper
transcription factor 1 appeared to be significantly up regulated up to 48 hours of
incubation in hypoxia, while STAT1 was constantly down regulated throughout all
the experimental time. The level of expression of transcripts for the hypoxia-inducible
factor 1 alpha polypeptide were not significantly modulated, in agreement with the
post-transcriptional regulation of the activity of this transcriptional activator.
Finally, we focused on the expression levels of the genes whose transcriptional
regulation is known to be under the control of the HIF transcription factor complex
(Semenza 2002), and which code for proteins involved in glycolysis and metabolism,
proliferation and survival, iron homeostasis and erythropoiesis, and angiogenesis.
Table 2 summarizes the expression level for 49 of the above genes. Thirty-nine genes
(~80%) were found to be considered present in all the experimental conditions, seven
(~14%) were always flagged “absent”. Although the majority of the present, HIF-
81
Table 2. Transcriptomic analysis of expression of HIF transcription factor complex-regulated genes 1.
Gene product GenBank Accession No.
Flag Hypoxia/normoxia
12h 24h 48h Aminopeptidase A L12468 A 0.74 0.82 1.03 Adenylate kinase 3 NM_013410 P 2.43 6.61 6.35 α1B-adrenergic receptor NM_000679 A 0.92 0.62 0.72 Adrenomedullin NM_001124 P 3.34 5.26 6.43 Aldolase A NM_000034 P 1.36 1.78 1.69 Aldolase C NM_005165 P 1.64 5.85 2.58 Carbonic anhydrase 9 NM_001216 A,P 1.04 0.89 1.17 Ceruloplasmin AA191647 P 0.68 0.92 0.77 Collagen type V α1 AI130969 P 0.70 1.33 2.26 DEC1 NM_003670 P 3.34 3.84 2.03 Endocrine gland-derived VEGF AF333022 n.a. n.a. n.a. n.a. Endothelin-1 NM_001955 P 0.83 1.01 1.35 Enolase 1 NM_001428 P 1.33 1.73 1.29 Erythropoietin NM_000799 A 0.70 0.92 1.47 ETS-1 NM_005238 P 1.81 1.38 1.06 Glucose transporter 1 NM_006516 P 1.67 2.32 2.81 Glucose transporter 3 NM_006931 P 3.70 7.28 5.67 GAPDH M33197 P 1.23 1.40 1.31 Heme oxygenase 1 NM_002133 P 0.99 0.80 0.63 Hexokinase 1 NM_000188 P 1.38 1.40 0.80 Hexokinase 2 AI761561 P 2.07 2.70 1.96 Insulin-like growth factor 2 M17863 A 0.98 1.20 0.90 IGF binding protein 1 NM_000596 P 1.62 1.68 0.55 IGF binding protein 2 NM_000597 P 0.46 0.62 1.33 IGF binding protein 3 M31159 P 1.53 2.90 10.02 Intestinal trefoil factor NM_003226 P,M,A 0.92 1.42 0.94 Lactate dehydrogenase A NM_005566 P 1.33 1.81 1.59 LDL receptor related protein 1 BF304759 A 1.30 1.27 1.79 NO synthase 2 L24553 P 1.25 1.10 1.29 NIP3 U15174 P 1.81 3.43 3.11 NIX AL132665 P 1.37 2.40 2.50 p21 NM_000389 P 1.39 1.22 0.83 p35srj AF109161 P 1.05 1.25 1.18 6-phosphofructo-2-kinase NM_004566 P 0.98 1.48 1.55 Phosphofructokinase L BC006422 P 1.23 1.77 1.54 Phosphoglycerate kinase 1 NM_000291 P 1.33 1.94 1.09 Plasminogen activator inhibitor 1 NM_000602 P 1.18 1.38 1.28 Prolyl-4-hydroxylase α(I) NM_000917 P 2.09 4.41 5.58 Pyruvate kinase M NM_002654 P 1.61 2.25 1.17 Transferrin NM_001063 A 4.36 3.93 3.90 Transferrin receptor BC001188 P 0.45 0.44 0.64 Transforming growth factor β3 J03241 A 1.15 0.99 1.35 Transglutaminase 2 M98478 P 1.05 1.14 1.07 Triosephosphate isomerase M10036 P 1.27 2.36 1.82 Vascular endothelial growth factor M27281 P 2.02 5.77 11.01 VEGF receptor FLT-1 U01134 P 5.72 9.26 3.43
[1] According to: Semenza, G., Biochem Pharmacol 2002, 64, 993-998.
82
regulated genes did not show modified expression level, a number of them (17/49,
~35%) appeared to be significantly up regulated (more than two-fold) in hypoxic
growth conditions. In HUVEC, the most striking up regulation was registered for
procollagen prolyl 4-hydroxylase α1, VEGF and FLT-1/VEGFR1. Only transferrin
receptor gene appeared to be down regulated (about two-fold) in our experimental
conditions.
3.2.2. RT-PCR analysis of expression level and of alternative splicing for some
extracellular matrix protein genes.
Components of the modified extracellular matrix (e.g., oncofetal fibronectin and
Figure 10: Semi-quantitative RT-PCR analysis of transcript abundance and pre-mRNA alternative splicing for the extracellular matrix protein gene Del-1. (A) Total RNA was extracted from triplicate samples of HUVECs grown in hypoxia or normoxia for 48h, as described in Experimental Procedures. RNA concentration was determined and identical amounts (200 ng) of each template were used in a reverse transcription-polymerase chain reaction (RT-PCR), employing the primers Del-1for and Del-1rev (see Table 6 in Materials and Methods). N1, N2 and N3 indicate the three replicate normoxic samples; H1, H2 and H3 indicate the three replicate hypoxic samples. Arrows on the left point to the position of the two expected amplification products (110 and 80 bp for Del-1 major transcript and Del-1 Z20 splice variant, respectively). The position of relevant bands in a DNA ladder (M4) is indicated on the right. (B) A serial dilution of RNA samples N1 and H1 was realized and a total amount of 200, 20, 2 and 0.2 ng of each template (or no template at all) was employed in a RT-PCR reaction with either the Del-1for and Del-1rev primers (upper panel “Del-1”) or the GAPDHfor and GAPDHrev primers (lower panel “GAPDH”). Arrows on the left point to the position of the expected amplification products (110 and 80 bp for Del-1 major transcript and Del-1 Z20 splice variant, and 983 bp for the GAPDH transcript, respectively).
83
tenascin isoforms generated by alternative splicing of the primary transcript) have
been used successfully by our group (Birchler, Neri et al. 1999; Carnemolla,
Castellani et al. 1999; Tarli, Balza et al. 1999; Viti, Tarli et al. 1999; Nilsson,
Kosmehl et al. 2001; Carnemolla, Borsi et al. 2002; Halin, Rondini et al. 2002) and
Figure 11: Semi-quantitative RT-PCR analysis of transcript abundance and pre-mRNA alternative splicing for the extracellular matrix protein genes Fibronectin and Tenascin C. (A) A serial dilution of RNA samples N1 and H1 (see legend to Fig. 10B) was realized and a total amount of 10, 5, 2.5 and 0.125 ng of each template (or no template at all) was employed in a RT-PCR reaction with the following couples of primers (see Table 6 in Materials and Methods for details): the FN-ED/Bfor and FN-ED/Brev primers (panel A: “FN ED/B”); the FN-ED/Afor and FN-ED/Arev primers (panel A: “FN ED/A”); the FN1for and FN1rev primers (panel A: “FN type III 2-4”) or the GAPDHfor and GAPDHrev primers (panel A: “GAPDH”). Arrows on the right point to the position of the expected amplification products (268 bp for the fibronectin extra-domain B, 263 bp for the fibronectin extra-domain A, 999 bp for the fibronectin type III homology domains 2-4, and 983 bp for the GAPDH transcript, respectively). The position of relevant bands in a DNA ladder (M4) is shown. (B) The same serial dilution of RNA samples N1 and H1 (see legend to Fig. 10B) was realized and a total amount of 10, 5, 2.5, 0.125 and 0,062 ng of each template (or no template at all; not shown) was employed in a RT-PCR reaction with the following couples of primers (see Table 6 for details): the TNCCfor and TNCCrev primers (panel B: “TNC repeat C”); the TNCDfor and TNCDrev primers (panel B: “TNC repeat D”); or the TNCfor and TNCrev primers (panel B: “TNC FN III 6-7). Arrows on the right point to the position of the expected amplification products (269 bp for the tenascin C repeat C, 263 bp for the tenascin C repeat D, and 368 bp for the tenascin C FN-type III repeats 6-7, respectively). The position of relevant bands in a DNA ladder (M4) is shown.
84
others (Paganelli, Bartolomei et al. 2001; Reardon, Akabani et al. 2002) as targets for
ligand-based anti-cancer intervention. We confirmed the hypoxia-induced over
expression of Del-1 (Figure 3) using semi-quantitative RT-PCR (Figure 10 A and 10
B). Del-1 has been reported as an endothelial cell-specific gene, absent in all adult
tissues tested, but strongly over expressed in certain fetal tissues and in a number of
tumors (Hidai, Zupancic et al. 1998; Aoka, Johnson et al. 2002). No significant up
regulation of fibronectin isoforms containing the extra-domains EDA and EDB could
be detected by RT-PCR (Fig. 11 A). These isoforms are known to be expressed by
proliferating ECs in aggressive tumors such as glioblastoma (Castellani, Borsi et al.
2002). Similarly, no striking over expression of tenascin-C isoforms could be detected
(Fig. 11 B), consistent with the observation that these ECM components are mainly
expressed by stromal cells and cancer cells in tumors (Hindermann, Berndt et al.
1999).
3.2.3. Proteomic study
We have used 2D-PAGE to study patterns of protein expression in HUVEC exposed
to hypoxic (2 % O2) or normoxic conditions (21 % O2). Figure 12 shows comparative
2D-gels of cell lysates. The vast majority of proteins showed comparable expression
levels of different samples exposed either to hypoxic or normoxic conditions, with the
possible exceptions of the heterogeneous nuclear ribonucleoprotein K and the 60 kDa
heat shock protein mitochondrial precursor (see Figure 12, spot 11 and Table 3) and
the 60 kDa heat shock mitochondrial precursor (Fig. 12, spot 14 and Table 3), which
were up regulated in hypoxic and normoxic conditions respectively. A 2D-PAGE
reference map of proteins contained in the HUVEC cell lysate is available at the
website http://www.pharma.ethz.ch/bmm/div/HUVEC/.
85
Figure 12: 2D-PAGE of a whole cell lysate of HUVEC grown in hypoxic and normoxic conditions. Proteins contained in whole cell lysates of HUVEC cells, grown for 48 hours in normoxic (gel “Normoxia”) or hypoxic (gel “Hypoxia”) conditions, were separated on 2D-PAGE gels, as described in Experimental Procedures. Two representative gels are shown, in which the IPG strips pH range was 3-10. Molecular weight markers were included and their values are indicated on the left. Two selected areas (a and b) in which differences in protein expression patterns could be appreciated are shown, on a magnified scale and in duplicate, at the bottom of each respective 2D-gel (N1a and N2a, N1b and N2b: replicates of normoxic HUVEC samples, magnification of areas a and b; H1a and H2a, H1b and H2b: replicates of hypoxic HUVEC samples, magnification of areas a and b). All the spots indicated by a number were excised from the gel, trypsin-digested and subjected to mass spectrometric analysis. A circled number identifies a protein over-expressed in hypoxia, a number in a square indicates a protein repressed in hypoxia. The correspondence between spot numbering and protein identification is given in Table 3.
86
Table 3 . Protein identification in whole cell lysate of HUVEC grown in normoxic and hypoxic conditions.
ID
Common name
SwissProt
Accession
Sequence coverage
%
pI (theor.)
pI (exptl.)
MW (theor.)
MW (exptl.) Normoxia Hypoxia
Ratio
H/N
1 Heat shock 27 kDa protein (HSP 27)
P04792 13.6 5.98 5.60 22782 23802 0.61 ± 0.17
0.19 ± 0.00 0.31
2 Endoplasmin precursor
P14625 6.0 4.76 4.58 92468 97335 4.78 ± 0.12
7.16 ± 1.19 1.50
3 Alpha-actinin 1
Alpha-actinin 4
P12814
O43707
4.6
6.3
5.22
5.27
5.28
5.17
102974
104854
103780
104383 1.44 ± 0.27
1.77 ± 0.36 1.22
Heat shock protein HSP 90-beta
P08238 9.5 4.97 83133
4 Transitional endoplasmic reticulum ATPase
P55072 4.2 5.14
4.81
89322
88018 4.55 ± 0.37
5.00 ± 1.41 1.10
5 78 kDa Glucose-regulated protein precursor (GRP 78)
Table 3 The table lists the proteins identified in whole cell lysates of HUVEC grown in normoxic or hypoxic conditions, shown in Fig. 12. A software-aided quantization of spot intensities in normoxic and hypoxic samples, as well as the ratio of hypoxic vs. normoxic spot intensities, the experimental isoelectric point and molecular weight are provided for each identified protein. Theoretical isoelectric point and molecular weight are derived from the amino acidic sequences published under the different SwissProt accession numbers and refer to the unprocessed precursor.
Figure 13: Experimental design of the cell fractionation proteomic experiment. For the cell fractionation, triplicate, independent samples were prepared from normoxic and hypoxic HUVEC cells. The cells were detached with a scraper, swollen for 20 minutes and lysed in a hypotonic buffer. The cells were then spun for 15 minutes at 3,000 g and the resulting pellet (Pellet 1) was washed extensively with hypotonic buffer before being subjected to 2D-PAGE analysis. The supernatant of this low-speed centrifugation (Sup 1) was further spun for 2 hours at 100’000xg at 2°C to pellet the microsomal fraction (Pellet 2). This pellet was washed extensively with hypotonic buffer and analyzed by 1D SDS-PAGE (data not shown). The proteins in the ultracentrifugation supernatant (Sup 2) were precipitated with trichloroacetic acid, redissolved in 100 µl of rehydration solution and subjected to 2D-PAGE analysis (data not shown).
89
Figure 14: 2D-PAGE of the cell fractionation “Pellet 1” in HUVEC grown in hypoxic and normoxic conditions. Proteins contained in “Pellet 1” of the cell fractionation protocol (see Experimental Procedures and legend to Fig. 7) of HUVEC cells, grown for 48 hours in normoxic (gel “Normoxia”) or hypoxic (gel “Hypoxia”) conditions, were separated on 2D-PAGE gels, as described in Experimental Procedures. Two representative gels are shown, in which the IPG strips pH range was 3-10. Molecular weight markers were included and their values are indicated on the left. Two selected areas (a and b) in which differences in protein expression patterns could be appreciated are shown, on a magnified scale and in triplicate, at the bottom of each respective 2D-gel (N1a, N2a and N3a, N1b, N2b and N3b: triplicates of normoxic HUVEC samples, magnification of areas a and b; H1a H2a and H3a, H1b, H2b and H3b: triplicates of hypoxic HUVEC samples, magnification of areas a and b). All the spots indicated by a number were excised from the gel, trypsin-digested and subjected to mass spectrometric analysis. A circled number identifies a protein over-expressed in hypoxia. The correspondence between spot numbering and protein identification is given in Table 4.
90
Table 4 . Protein identification in cytoplasmic fraction of HUVEC grown in normoxic and hypoxic conditions
Table 4 The table lists the proteins identified in cell fractionation “Pellet 1” of HUVEC grown in normoxic or hypoxic conditions, and shown in Fig. 14. A software-aided quantization of spot intensities in normoxic and hypoxic samples, as well as the ratio of hypoxic vs. normoxic spot intensities, the experimental isoelectric point and molecular weight are provided for each identified protein. Theoretical isoelectric point and molecular weight are derived from the amino acidic sequences published under the different SwissProt accession numbers and refer to the unprocessed precursor.
In order to get a better insight about the presence of differentially expressed proteins,
we performed a protein fractionation according to scheme of Figure 13. Figure 14
shows gels run with proteins of “Pellet 1”. Again, while most proteins exhibited
similar patterns of expression, an up regulation for procollagen-lysine 2-oxoglutarate
5-dioxygenase and α-enolase could be detected in hypoxic conditions. The most
striking observation of the study was the extremely small number of differentially
expressed proteins detected. The quantitative results of this analysis are reported in
Table 4. Proteins known to be present at different levels in normoxic and hypoxic
adrenomedullin (Cormier-Regard, Nguyen et al. 1998), procollagen proline 4-
hydroxylase (Takahashi, Takahashi et al. 2000), insulin growth factor-binding protein
3 (Feldser, Agani et al. 1999)) and other genes for which an increased expression in
hypoxia-related conditions in vitro or in vivo has been already reported (lysyl oxidase
(Brody, Kagan et al. 1979), taurine transporter (Saransaari and Oja 1999; Schaffer,
Pastukh et al. 2002), stanniocalcin (Lal, Peters et al. 2001), N-myc downstream
regulated (Park, Adams et al. 2000), inhibin beta A (Lai, Sirimanne et al. 1996)).
However, the list includes also many other genes and ESTs (16 annotated genes, 8
cDNA clones, 4 hypothetical proteins) for which an up-regulation in mRNA
expression upon exposure to hypoxia, to our knowledge, has never been reported
before. Among these genes, claudin 3, CD24 and tetranectin deserve some comment.
Claudin 3 is a member of a family of integral membrane proteins, the claudins, which,
together with occludins, represent the major constituents of tight junctions (Morita,
Furuse et al. 1999), critical structures for the maintenance of cellular polarity, as well
as for the establishment of a permeability barrier for paracellular transport in epithelial
and endothelial cells (Tsukita, Furuse et al. 2001). A strong up-regulation of claudin-3
and claudin-4 mRNA levels, as well as an over-expression at the protein level has
been reported in ovarian carcinomas, as compared to ovarian cystadenomas and
normal ovary (Rangel, Agarwal et al. 2003). It has been hypothesized that claudin-3,
when over expressed, might interfere with normal tight junction formation and
function (Rangel, Agarwal et al. 2003). An over-expression of claudin-3 might be
100
involved also in the impairment of tight junctions in endothelial cells, that leads to the
increased vascular permeability registered in hypoxia (Gonzalez and Wood 2001).
CD24, a sialoglycoprotein that is anchored to the cell surface by a GPI anchor
(Fischer, Majdic et al. 1990), is expressed on many tumor cells and is a ligand for P-
selectin (Aigner, Sthoeger et al. 1997), a Ca2+-dependent endogenous lectin that can
be expressed by activated vascular endothelium and platelets. P-selectin interacts with
P-selectin glycoprotein ligand-1 on leukocytes, taking part in the process of leukocyte
capture, rolling and extravasation (Yang, Furie et al. 1999). The differential
expression of surface antigens has been studied on activated endothelium and it has
been shown that CD24, among others, was increased in endothelial cells upon
stimulation with either TNF-α, interferon γ or thrombin (Favaloro 1993). Our results
show that CD24 mRNA expression is up regulated at late times of exposure to
hypoxia in HUVEC cells. Although the biological meaning of this observation
remains unclear, CD24 expression might be regarded as a useful marker of hypoxic
activation in vascular endothelial cells.
Tetranectin, a tetrameric protein isolated from human plasma, has a specific binding
affinity for sulfated polysaccharides and the kringle 4 domain of plasminogen (Wewer
and Albrechtsen 1992). It is induced during the mineralization phase of osteogenesis,
and a role of this protein in human disorder affecting bone and connective tissue has
been postulated (Wewer, Ibaraki et al. 1994). In situ hybridization studies for
tetranectin messengers on tissue sections of colon carcinomas and normal colon
tissues revealed a strong and distinct hybridization signal of stromal cells in colon
carcinomas but not of tumor cells. Only a few stromal cells were labeled in the normal
101
colon (Wewer and Albrechtsen 1992). Immunohistochemically, tetranectin was found
in a fibrillar-like pattern in the extracellular matrix around the tumor islands and was
not detectable in the normal colon stromal tissue. We have shown (see above) by both
transcriptomic and proteomic experiments (Bumke, Neri et al. 2003) that tetranectin is
strongly up regulated in vitro by serum starvation in normal human fibroblasts, but
that this up regulation is dampened by a simultaneous shift to an acidic milieu. The
finding that tetranectin gene expression is also strongly up-regulated in HUVEC
exposed to hypoxia reinforce the possibility that the plasmin cascade play a
remarkable role in the degradation and remodeling of the extracellular matrix, one of
the key steps in tumor angiogenesis.
When one examines the list of the genes which resulted to be strongly down regulated
by exposure to hypoxia of HUVEC cells, the most striking findings is by far the
decreased expression of messengers for bFGF, especially in view of the concomitant,
strong up-regulation of VEGF. Hypoxia has been shown to lead to an increase in
bFGF mRNA levels in human breast carcinoma cells (Le and Corry 1999), in
pulmonary vascular pericytes (Wang, Xiong et al. 2000) and, in vivo, in the adult
mouse retina (Grimm, Wenzel et al. 2002). By contrast, it has been reported that, in in
vivo xenografts of rat and human prostatic cancers, exposure to hypoxia leads to an
increase in VEGF, but not bFGF protein expression (Joseph and Isaacs 1997).
Among the genes of the extracellular matrix, a particularly interesting case is
represented by procollagen proline 4 hydroxylase A and procollagen lysine 4
hydroxylase 1 and 2.
102
As already discussed, exposure of HUVEC to hypoxia strongly up-regulates the
expression of procollagen proline 4-hydroxylase 1 gene (P4HA). In collagen
synthesis, P4HA is a key enzyme that catalyzes the formation of 4-hydroxyproline, an
essential residue for the folding of the procollagen polypeptide chains into triple
helical molecules (Kivirikko and Pihlajaniemi 1998). Many observations point to the
fact that systemic and cellular hypoxia modulates collagen synthesis in several types
of cells (Falanga, Martin et al. 1993; Durmowicz, Parks et al. 1994; Ostadal, Kolar et
al. 1995; Kim, Kang et al. 1996). Lysyl hydroxylase (PLOD) catalyzes the
hydroxylation of lysine in -X-Lys-Gly- sequences in collagens and exists in three
different isoforms (Kivirikko and Pihlajaniemi 1998). The resulting hydroxylysine
residues have two important functions: they act as attachment sites for carbohydrate
units and are essential for the stability of the intermolecular collagen cross-links
(Kivirikko and Pihlajaniemi 1998). The importance of proper lysyl hydroxylase
function is clearly seen in patients with Ehler-Danlos syndrome type VI, an autosomal
recessive disorder of connective tissue characterized by hyperextensible, friable skin
and joint hypermobility. Severe scoliosis and ocular fragility are present in some
patients. The disease is associated to a relative deficiency of specific hydroxylysine-
derived collagen crosslinks, owing to a reduced functionality of the enzyme (Ha,
Marshall et al. 1994; Pasquali, Still et al. 1997).
Our data show that both PLOD1 and PLOD2, but not PLOD3, are up regulated at late
times of exposure to hypoxia in HUVEC. We examined the promoter region of human
PLOD1 and PLOD3 and of the murine PLOD2 genes and found that both hPLOD1
and mPLOD2 contain a putative functional hypoxia responsive element (HRE). A
functional HRE consists of a pair of contiguous transcription factor binding sites, at
least one of which contains the core HIF-1 binding sequence 5’-RCGTG-3’
103
(Semenza, Jiang et al. 1996). In genes that contain two contiguous HIF-1 binding
sites, they are arranged as either directed or inverted repeats, separated by 4-12 bp.
Presence of a single binding site is necessary but not sufficient for HRE function
(Semenza, Jiang et al. 1996). Figure 15 shows a comparison of sequences of promoter
regions of hPLOD1 (GenBank AF490514, positions 453-489) and mPLOD2
(GenBank AF283255, positions 503-534) with those of other genes that have been
demonstrated to be under the transcriptional control of HIF-1. hPLOD1 promoter
hPLOD1 5’-GTGTCCGTGTCTCTGAACGCATCCGTGCCCACCCTCA-3’ hENO1 5’-GAGTGCGTGCGGGACTCGGAGTACGTGACGGAGCCCC-3’ hALDO 5’-CCCCCTCGGACGTGGACTCGGACCACAT-3’ hEPO 5’-GGGGCTGGGCCCTACGTGCTGTCTCACACAGCCTGTC-3’ mPLOD2 5’-AGCCCCGGCGTGAGGCGGTGCACGTAGGCGTA-3’ mGLUT1 5’-TCCACAGGCGTGCCGTCTGACACGA-3’ mLDHA 5’-GCCCAGCCTACACGTGGGTTCCCGCACGTCCGCTGGG-3’ Figure 15: Comparison of sequences of promoter regions of hPLOD1 and mPLOD2 with other HIF-1 regulated genes. The promoter regions of hPLOD1 (GenBank AF490514, positions 453-489) and mPLOD2 (GenBank AF283255, positions 503-534) were aligned with those of other human and murine genes whose expression has been demonstrated to be controlled by the transcription factor HIF-1. hPLOD1, human procollagen lysyl hydroxylase 1; hENO1, human enolase alpha (Semenza, Jiang et al. 1996); hALDO, human aldolase A (Semenza, Jiang et al. 1996); hEPO, human erythropoietin (Semenza, Nejfelt et al. 1991); mPLOD2, murine procollagen lysyl hydroxylase 2; mGLUT1, murine glucose transporter 1 (Ebert, Firth et al. 1995); mLDHA, murine lactate dehydrogenase A (Firth, Ebert et al. 1994).
104
sequence is very similar to that found in human enolase alpha gene (Semenza, Jiang et
al. 1996), mPLOD2 sequence strongly resembles to that of murine glucose transporter
1 (Ebert, Firth et al. 1995). Although these findings per se are not sufficient to draw
any conclusion about the functionality of the HREs identified, and await confirmation
based on transient transfection studies using constructs containing the PLOD
promoter(s) linked to a reporter gene, they at least suggest that the PLOD 1 and 2
gene regulation might be controlled by the HIF-1 transcription factor. A coordinated
up-regulation of proline and lysine hydroxylases involved in synthesis and structural
organization of collagens, the major components of the extracellular matrix, might be
crucial for the remodeling of the extracellular matrix that occurs in the early steps of
hypoxia-induced tumor angiogenesis.
In the ontological category of ligands, our analysis identified an interesting up-
regulated gene, Del-1 (developmentally regulated, endothelial locus 1). The protein
encoded in this locus contains three EGF-like repeats, homologous to those in Notch
and related proteins, including an EGF-like repeat that contains an RGD motif, and
two discoidin I-like domains (Hidai, Zupancic et al. 1998). The expression pattern of
Del-1 is unique in that it is initially expressed already in endothelial progenitor cells
of the extraembryonic mesoderm, but then declines and disappears completely by
birth (Hidai, Zupancic et al. 1998). It is completely absent in practically all adult
tissues (except in the superficial layer of articular cartilage (Pfister, Aydelotte et al.
2001)), but its expression is strongly reactivated in some tumor cell lines (Hidai,
Zupancic et al. 1998). Del-1 is deposited in the extracellular matrix, where it promotes
endothelial cell attachment and migration by binding, via its RGD motif, to the αvβ3
integrin receptor (Penta, Varner et al. 1999). It has been shown also that Del-1 is
105
massively expressed in the modified extracellular matrix of solid tumors in breast
carcinoma, melanoma and colon carcinoma specimens, being associated with both
tumor cells and angiogenic endothelial cells (Aoka, Johnson et al. 2002). Del-1
promises to become an interesting marker for tumor angiogenesis, and a new target
for therapeutic intervention.
3.2.4.2. Proteomics
A proteomic study was performed to complement the transcriptomic analysis since
proteins carry dynamic (e.g. phosphorylation) and static modifications (e.g. disulfide
linkage) that may not be apparent from genomic information or from mRNA
abundance (Corthals, Wasinger et al. 2000). Furthermore, a study comparing 2D-gel
protein-expression measurements with the corresponding message data derived from
differential gene expression showed that the correlation between mRNA and the
cognate protein is poor, suggesting that post-transcriptional regulation of gene
expression is a frequent phenomenon (Anderson and Seilhamer 1997).
We measured more than 700 trypsin-digested samples derived from 2D gels by
tandem mass spectrometry. About a third of the measured spots were clearly
identified with two or more peptides and a good correlation coefficient (> 2.5) by the
SEQUEST algorithm, one third of the samples could only be assigned with one
peptide.
The expression of procollagen lysine, 2-oxoglutarate 5-dioxygenase 2, (procollagen
lysyl hydroxylase 2, PLOD2) was found to be up regulated in HUVECs incubated for
48 hours in hypoxic conditions (Fig. 14, spots 1 to 3). This result mirrors the
106
corresponding increase in PLOD2 mRNA, observed in the transcriptomic experiment
and already discussed (see above).
Another remarkable difference is represented by the increased expression of the α-
enolase protein in hypoxic HUVEC (Fig. 14, spots 4 and 5), with respect to normoxic
controls. The glycolytic enzyme α-enolase gene is known to be controlled by the
transcriptional complex HIF-1 (Semenza, Jiang et al. 1996), yet in our cell model
system it did not show an hypoxia-dependent up-regulation at the level of mRNA
transcripts (see Table 1). The finding of an increased expression of the enolase α at
the protein level might indicate some additional post-translational regulation of the
expression of this enzyme in hypoxic HUVEC.
The poor correspondence between transcriptomic and proteomic analysis is not an
unusual finding. In the above described study, aimed at investigating the modulation
of gene expression by extracellular pH variations in human fibroblasts and making
use of the same transcriptomic and proteomic approach (Bumke, Neri et al.), only one
out of six proteins differentially expressed in the proteomic analysis was also found to
be modulated at the transcriptional level. Reasons for this are inherent to the very
different technical limitations imposed by sample preparation, dynamic range and
sensitivity of the two techniques that ought to be regarded as complementary rather
than alternative.
The most striking observation of our proteomic analysis was the extremely small
number of differentially expressed proteins detected in hypoxic vs. normoxic
HUVEC. In particular, membrane protein solubility problems limited the number of
detectable proteins to the most abundant ones, with no significant variations between
107
the two conditions under study. We have established a cell surface protein
biotinylation procedure that ought to facilitate the recovery of biotin-labeled
membrane proteins by avidin affinity chromatography, followed by their proteolytic
digestion and quantitative LC-MALDI-TOF of the obtained peptide mixture
(Scheurer, Rybak et al. 2005). Tandem mass spectrometry of the relevant peptide
fractions would ultimately lead to the identification of the interesting, modulated
peptides.
108
3.3. Identification and relative quantification of membrane proteins by surface
biotinylation and two-dimensional peptide mapping
Membrane proteins play a variety of fundamental roles in all living organisms. In
particular, transmembrane proteins and proteins anchored to the cell membrane
represent more than 30% of all proteins in the human genome (Paulsen, Sliwinski et
al. 1998; Wallin and von Heijne 1998), and more than two-thirds of the known protein
targets for drugs (Stevens and Arkin 2000). However, in spite of their relevance,
membrane proteins have often escaped a systematic analysis and quantification in
biological systems, due to their limited solubility in water and their relatively low
abundance (Scheurer, Rybak et al. 2004). Methods for the identification and relative
quantification of membrane proteins are likely to have an impact in many areas of
biological research, including immunology and pharmaceutical sciences.
When monoclonal antibodies specific to membrane proteins are available, the relative
quantification of membrane proteins in tissues and cells is facilitated by
immunohistochemical and fluorescence-activated cell sorting analysis (FACS)
(Herzenberg, Parks et al. 2002). However, antibody-based methods have some
limitations in the simultaneous detection and quantification of a large number of
different membrane proteins.
Since its introduction in 1975, 2D-PAGE has met an enormous success for the
simultaneous separation and relative quantification of several hundreds of proteins in
biological specimens. Even though successful examples (Celis, Ostergaard et al.
1996; Celis, Rasmussen et al. 1996; Celis, Celis et al. 1999; Celis, Celis et al. 2002)
109
and technical improvements have become available, 2D-PAGE methodologies are
often unsuitable for the analysis of membrane proteins, primarily because of poor
protein solubility in the buffers used for protein separation in the first dimension
(Santoni, Molloy et al. 2000).
Isotope-coded affinity tag (ICAT) methodologies (Gygi, Rist et al. 1999) are rapidly
establishing themselves as a valuable tool for the relative quantification of proteins in
biological samples, but the need to label primary amino groups in membrane proteins
(rather than thiol groups) may result in complex distributions of fractionally labeled
peptides, thus complicating peptide recovery, separation and relative quantification
(Haynes and Yates 2000; Patton, Schulenberg et al. 2002). Other mass-spectrometry
based approaches (such as the multi-dimensional protein identification technique
(Washburn, Wolters et al. 2001) or the shotgun analysis of soluble and membrane
proteins according to Wu et al. (Wu, MacCoss et al. 2003) have shown remarkable
success in the simultaneous identification of hundreds of membrane proteins, but are
inherently not quantitative.
Surface biotinylation with reactive chemical derivatives of biotin, followed by
purification on streptavidin, has been investigated as a method to restrict proteomic
analysis to membrane proteins. These approaches have clearly shown a potential for
the selective chemical modification and recovery of membrane proteins and
extracellular proteins, both in vitro (Busch, Hoder et al. 1989; Brandli, Parton et al.
1990; Sabarth, Lamer et al. 2002) and ex vivo (Rybak, Scheurer et al. 2004).
110
In this thesis, we describe a method for the simultaneous recovery, separation,
identification and relative quantification of membrane proteins. After covalent
modification with a cleavable reactive ester derivative of biotin, cells are lysed in the
presence of detergents and membrane proteins are purified on streptavidin-coated
resin. Proteins are then eluted and tryptically digested. The resulting peptides are used
to generate a two-dimensional map, based on HPLC separation in the first dimension
and mass-spectrometric analysis in the second dimension. The use of internal peptide
standards in MALDI-TOF mass spectrometry facilitates the relative quantification of
peptides (and thus of the corresponding protein) (Nelson, McLean et al. 1994). We
have exemplified this method by studying i) the detection of peptides from a
biotinylated BSA-spike, added to an aliquot of HEK biotinylated membrane proteins
before capture on streptavidin Sepharose, with respect to the non-spiked counterpart,
and ii) the relative changes of membrane protein expression of HUVEC cells, cultured
in normoxic or hypoxic conditions. The results exposed in the previous section report
about the comparative transcriptomic and proteomic analysis of the response of
HUVEC cells to hypoxia, using Affymetrix gene chip technology and 2D-PAGE
(Scheurer, Rybak et al. 2004).
3.3.1. Surface biotinylation and two-dimensional peptide mapping for the proteomic
study of membrane proteins
We have established a method (termed “2D peptide mapping”) for the selective
isolation and relative quantization of membrane proteins (Figure 16). For sample
preparation, the cell surface is biotinylated with the biotin derivative sulfo-NHS-SS-
biotin, a cleavable and water-soluble biotinylation reagent specific for primary amino
111
groups. The biotinylated cells are lysed in the presence of 0.2% SDS and 2% NP-40
and biotin-modified proteins are then purified on a streptavidin-coated resin, eluted by
reduction of the biotin linker and digested with trypsin. The resulting tryptic peptides
are fractionated by microcapillary reversed-phase HPLC, followed by MALDI-TOF
MS analysis of the HPLC fractions. This procedure yields a collection of spectra,
which can be displayed as a two-dimensional peptide map, with HPLC fractions in the
first dimension, mass/charge ratios (m/z) in the second dimension, and MS peak
intensity represented with a grayscale. For this purpose, we developed a flexible
igure 16
F . Schematic representation of the experimental methodology for the identification and relative uantification of membrane proteins by surface biotinylation and two-dimensional peptide mapping. q
Cell surface proteins are covalently modified with sulfo-NHS-SS-biotin, a biotin derivative carrying a cleavable linker and reactive for primary amino groups. The cells are lysed in the presence of detergents and biotin-labeled proteins are purified on a resin coated with streptavidin. Isolated proteins are enzymatically digested following their elution from the resin. The resulting peptides are fractionated by reversed-phase microcapillary high-performance liquid chromatography (RP-HPLC). Eluting fractions are splitted in two: one set of fractions is analyzed by matrix-assisted laser ionization time of flight mass spectrometry (MALDI-TOF MS) and the other set by microcapillary-liquid chromatography tandem mass spectrometry (µLC-MS/MS). A two-dimensional peptide map (2D-peptide map) is established with the data obtained by MALDI-TOF MS, reporting the HPLC fractions on the y-axis and the mass to charge ratio of the measured peptides on the x-axis.
112
software (termed “Spectational”; the software is available f on the following website:
ich is not used for
hen the 2D peptide mapping procedure is performed in parallel on two closely
In order to assess ion suppression effects (Kratzer, Eckerskorn et al. 1998; Wall,
Berger et al. 2002) in peptide mixtures, we studied the influence of an increasing
amount (0.1, 1, 2, 4, 6, 8, 10, 20 or 40 pmol) of competitor peptide
KVPQVSTPTLVEVSR on the ion signal intensity of 5 pmol peptide
LVNEVTEFAK. The competitor was mixed with 5 pmol of peptide LVNEVTEFAK,
and five replicates of each sample were measured by MALDI-TOF MS. The addition
Figure 17. MALDI-TOF MS with test peptides. Five replicates of a dilution series (ranging from 0.1 pmol to 40 pmol) of a test peptide with the sequence LVNEVTEFAK were measured by MALDI-TOF mass spectrometry. (A) The averaged peak signal intensities are plotted relative to the corresponding peptide amounts. (B) The same dilution series (five replicates) were mixed with 5 pmol of a standard peptide with the sequence KVPQVSTPTLVEVSR. The averaged peak signal intensities of the test peptide were normalized to the peak signal intensity of the standard peptide and plotted versus the corresponding amounts of test peptide. (C) Five pmol of the peptide LVNEVTEFAK were analyzed with increasing amounts of the peptide KVPQVSTPTLVEVSR (0.1 pmol to 40 pmol). Panel C displays the averaged peak signal intensities of 5 pmol of the peptide LVNEVTEFAK relative to different amounts of the peptide KVPQVSTPTLVEVSR added. (D) Panel D shows a plot of the ratio between the ion signal intensities of peptide LVNEVTEFAK and peptide KVPQVSTPTLVEVSR plotted versus the ratio of their relative concentration (logarithmic scale).
114
of increasing amounts of competitor peptide led to a decrease in ion signal intensity
for peptide LVNEVTEFAK (Figure 17 C). This observation suggests that relative ion
signal intensity can be used for the relative quantification of peptides in samples
containing comparable total amounts of peptides, while suppression effects may be
experienced in peptide mixtures that greatly differ in total peptide quantity.
However, the ratio of ion signal intensities of the peptide LVNEVTEFAK and the
competitor peptide was found to be proportional to the ratio of their relative amount
(Figure 17 D), confirming that the use of an internal standard peptide helps
minimizing the influence of ion suppression on the relative quantification analysis.
3.3.2. Detection of tryptic peptides derived from a BSA-spike added to an HEK
membrane protein extract
We performed a spiking experiment with bovine serum albumin, in order to determine
whether our method is suitable to the detection of differences in protein composition
of biotinylated samples.
We carried out a biotinylation experiment on two independent samples of HEK cells.
Biotinylated proteins were extracted from each sample and, to one of them, 100
pmoles of biotinylated BSA were added. The two samples were processed in parallel
and the tryptic peptides obtained were subjected to two-dimensional peptide mapping,
as described in the Materials and Methods section.
At least three different signals, corresponding to BSA peptides, could be detected only
in the BSA-spiked sample. As an example, Figure 18 A shows a portion of the 2D
peptide map obtained for fractions 18 to 28 of the BSA-spiked sample versus the non-
spiked control, after linear software normalization of peptide signal intensities relative
to the MS intensity of the internal standard peptide. In the boxed region of the map,
115
Figure 18. 2D peptide map of HEK cell surface proteins, in the presence or absence of a biotinylated BSA spike. Replicate HEK cultures were biotinylated and processed in parallel. Biotinylated BSA was added as a spike to one-half of the replicates before capture the samples on streptavidin Sepharose, processing them for HPLC separation and MALDI-TOF analysis and establishing a 2D peptide map. Panel A shows a portion of the 2D peptide map corresponding to HPLC fractions 18 to 28 (y-axis) and to m/z interval 700-2500 (x-axis). Peak signal intensities were normalized to the intensity of an internal standard peptide added at known amount to each HPLC fraction prior to MALDI-TOF MS measurements. Asterisks indicate BSA-spiked fractions. Panels B and C display the MALDI-TOF spectra of the non-spiked and spiked fraction 26, respectively corresponding to the section marked with a square in Panel A. The white arrows in Panel A and C point to a tryptic peptide of mass 1439.9, specifically present only in the BSA-spiked sample, and corresponding to the BSA peptide RHPEYAVSVLLR, whose calculated monoisotopic mass is 1439.8117.
the presence of a signal in the fraction 26 of the BSA-spiked sample, which is absent
in the control, is clearly detectable. Panels B and C show the MALDI-TOF spectra
corresponding to the boxed region of the non-spiked (panel B) and of the BSA-spiked
(panel C) samples. A peak at m/z ratio of 1439.9 is clearly detected and corresponds
116
to the BSA peptide RHPEYAVSVLLR, whose calculated monoisotopic mass is
1439.8117.
3.3.3. Two-dimensional peptide mapping of membrane proteins of HUVEC cell
cultures
We explored the applicability of the 2D peptide mapping technique for the study of
differentially expressed membrane proteins in HUVEC exposed to hypoxic (2% O2)
or normoxic (21% O2) growth conditions.
Figure 19 shows a section of the 2D peptide map from replicate HUVEC cultures,
Figure 19. 2D peptide map of cell surface proteins of HUVEC grown in hypoxic conditions. Replicate HUVEC cultures were grown in hypoxic conditions and processed in parallel as described in Materials and Methods. A 2D peptide map was established using the Spectational software. The figure displays a region of the 2D peptide map, corresponding to HPLC fractions 20 to 40 (y-axis) and m/z interval 500-3500 (x-axis). Peak signal intensities were normalized to the intensity of an internal standard peptide added at known amount to each HPLC fraction prior to MALDI-TOF MS measurements. Each fraction was also analyzed by µLC-MS/MS to identify the proteins contained in the sample. Peptides representing identified proteins are indicated with a number, which corresponds to the protein identification in Supplementary Table 1 in Appendix A (see also the text for reference to our website).
117
grown for 48 hours in hypoxic conditions. The replicate samples were processed in
parallel and independently from each other according to the scheme depicted in
Figure 16. MALDI-TOF MS spectra of HPLC fractions 20-40 are displayed in
sequential order, after linear software normalization of peptide signal intensities
relative to the MS intensity of the internal standard peptide.
The commercial preparation of the internal standard peptide we used appears as a
mixture of three different peptides (observed, in the average, at m/z 1639.4, 1538.4
and 1511.3). Signals of the internal standard peptides and of most tryptic peptides can
be observed for all fractions at comparable positions and relative intensity, thus
confirming the reproducibility of the 2D-peptide mapping procedure in independent
experiments performed in duplicate.
In total, 71 proteins were identified, 41 % of which could be assigned as type I
membrane proteins (PECAM-1, integrins, vascular endothelial cadherin and others),
integral membrane proteins (monocarboxylate transporter and sodium channel protein
type I alpha subunit) and membrane associated proteins (Caveolin-1, alpha-1 catenin),
by tandem mass spectrometry. Supplementary Table 1 (Appendix A; also available at
the website: http://www.pharma.ethz.ch/bmm/div/2D_pepmap_HUVEC) shows a list
of the proteins identified in the hypoxic HUVEC samples and shown in Figure 19,
with their SwissProt accession number, as well as the corresponding peptides and
fraction number. In the course of this experiment, we also identified extracellular
matrix proteins (13%) and cytoplasmic proteins (32%). Another part of the proteins
identified consists of contaminants (keratins), serum components (such as serum
3.3.4. Comparison of membrane protein expression in HUVEC cells exposed to
hypoxia and normoxia
While the majority of peptides from biotinylated surface proteins did not show
substantial changes of expression in the normoxic or hypoxic conditions, some
peptides were reproducibly found to be more abundant in one of the two experimental
conditions. Figure 20 A, C and E show portions of the 2D peptide map obtained,
normalized to the signal intensity of the internal standard peptide peak. Arrows point
to peptide signals which were either found not to be modulated between the two
conditions under study (Fig. 20 C) or which were specifically present only in one
condition (Fig. 20 A and 20 E). Figure 20 B, D and F represents the MALDI-TOF MS
spectra corresponding to the 2D peptide map sections shown in Fig. 20 A, C and E,
respectively (only one of the two replicates shown) and confirm the specificity of
peptide peak detection in normoxic or hypoxic HUVEC samples.
The peptide indicated by “a” in the hypoxic replicate samples (Fig. 20 A, B) was
identified by LC-MS/MS as NEGVATYAAAVLFR, a peptide of beta-catenin, a
protein involved in the Wnt growth factor signaling cascade (Lampugnani and Dejana
1997) and component of adherens junctions (Biswas, Canosa et al. 2003), suggesting
that this protein could be over-expressed in hypoxic conditions. The peptide “b” (Fig.
20 C, D) was identified as the peptide KPLIGTVLAMDPDAAR of VE-cadherin. The
amount of this protein and of alpha-catenin (not shown), two other major components
of adherens junctions, appear not to be modulated by hypoxia in 2D peptide mapping
analysis. These findings were confirmed by western blotting analysis, performed
using normoxic and hypoxic HUVEC whole cell lysates (Fig. 20 H), showing that in
hypoxic HUVEC samples beta-catenin is significantly over-expressed with respect to
the normoxic counterpart, while VE-cadherin and alpha-catenin are not modulated.
119
Figure 20. Comparative analysis of HUVEC grown in hypoxic and normoxic conditions. Panels A, C and E show sections from 2D-peptide map of two replicates (“1” and “2”) of normoxic (“N”) and hypoxic (“H”) HUVEC samples, respectively. Arrows indicate reproducible differences. The MALDI-TOF MS spectra corresponding to the 2D peptide map sections, presented in panels A, C and E are shown in panels B, D and F, respectively. The white arrows (“a” to “d”) point to relevant differences detected in the two samples. The peptide indicated by the arrow “a” (panels A and B) and over-expressed in hypoxic HUVEC was identified by LC-MS/MS as NEGVATYAAAVLFR, a peptide of beta-catenin. The peptide “b” (panels C and D) was identified as the peptide KPLIGTVLAMDPDAAR of VE-cadherin. Peptides “c” and “d” (panels E and F) were identified as GGVNDNFQGVLQNVR and VDVIPVNLPGEHGQR, belonging respectively to thrombospondin 1 and fibronectin. The over-expression of beta-catenin, thrombospondin 1 and fibronectin in hypoxic HUVEC, as well as the lack of modulation of VE-cadherin and alpha-catenin in the two conditions were confirmed by western blot analysis. Panel H shows immunoblot analysis of HUVEC whole cell lysates. Two replicate cultures (“1” and “2”) of hypoxic (“H”) and normoxic (“N”) HUVEC were lysed and 10 µg of protein were separated by SDS-PAGE. The proteins were transferred to a nitrocellulose membrane and probed for the different proteins. Panel G shows a SDS-PAGE gel replicate, stained with Sypro Ruby, as a loading control.
120
Some more peptides were also found to be up regulated in hypoxic HUVEC. For
example, peptide “c” and peptide “d” (Fig. 20 E, F) were identified as GGVN
DNFQGVLQNVR and VDVIPVNLPGEHGQR, belonging respectively to
thrombospondin 1 and fibronectin. Also in this case, Western blot analysis of
normoxic and hypoxic HUVEC whole cell lysates (Fig. 20 H) could confirm the over
expression of these two proteins in hypoxia.
A Sypro Ruby stained replicate of the SDS-PAGE gel is shown in Fig. 20 G as a
loading control, indicating comparable amounts of proteins in the four different lanes.
3.3.5. Discussion
The 2D peptide mapping procedure described in this thesis allows a relative
quantification of surface proteins in parallel cell cultures exposed to different
experimental conditions.
In a simple experimental setup, represented by biotinylated membrane proteins from
HEK cells, to which a BSA-spike was added, we could identify different BSA
peptides specifically present only in the spiked sample. In HUVEC cells, grown in
normoxic or hypoxic conditions, we could detect peptides that are differentially
expressed in one of the two conditions. Until now, abundant surface proteins of
endothelial cells (including integrins, collagens, endoglin and a number of CD
antigens) were preferentially identified. However, improved protein solubilization
after elution from streptavidin resin or direct tryptic digestion on the resin may further
increase the number of proteins identified.
121
The major part of the proteins identified in our analysis was membrane proteins or
extracellular matrix components. However, we also observed peptides corresponding
to cytoplasmic proteins. It has been shown that biotinylation of intracellular proteins
is greatly reduced when using sulfo-NHS-SS-Biotin as compared to sulfo-NHS-LC-
Biotin (Peirce, Wait et al. 2004), suggesting that the reducing milieu of the cytoplasm
could be responsible for the cleavage of the disulfide bond between protein and biotin
and as a consequence prevent the isolation of cytoplasmic proteins. It is possible,
nonetheless, that cytoplasmic proteins are co-purified by virtue of a strong interaction
with membrane proteins and despite of the strong detergents used, or are not
completely eliminated during the washing of the streptavidin-coated resin.
MALDI-TOF MS ion signal intensity allows the relative quantification of proteins in
2D peptide mapping. Ion suppression effects are minimized both by the use of internal
standards and by the comparison of closely related biological specimens (e.g., the
same cell cultures grown in different experimental conditions). The analysis of
replicate HUVEC cultures depicted in Figures 19 and 20 illustrates the reproducibility
of the procedure, while outlining surface proteins that are differentially expressed in
normoxic and hypoxic conditions. Like other methodologies [e.g., ICAT (Gygi, Rist
et al. 1999)], 2D peptide mapping only provides a relative quantization of proteins in
closely related samples. Absolute protein quantification is hindered not only by
limitations of mass spectrometry, but also by differences in tryptic digestion
efficiency among individual proteins.
122
The identification of beta-catenin as a protein over expressed in hypoxic conditions
deserves some considerations. Beta-catenin is a component of adherens junctions,
which mediates cell-cell adhesion by providing a physical link between the
transmembrane protein vascular endothelial-cadherin and the actin cytoskeleton via
interaction with alpha-catenin (Lampugnani and Dejana 1997). Those four proteins
have been identified in our analysis (Figure 4 and 5), revealing a tight interaction of
VE-cadherin with its intracellular binding partners (Huber and Weis 2001; Stow
2004), which resists washing steps in 1% NP-40 and 0.1% SDS. However, the levels
of expression of VE-cadherin and of alpha-catenin appeared not to be influenced by
HUVEC exposure to hypoxic growth conditions. In hypoxic conditions, beta-catenin
expression does not appear to be differentially regulated at the mRNA level
(Scheurer, Rybak et al. 2004), thus suggesting a post-transcriptional nature of beta-
catenin over-expression in hypoxia. A functional link between beta-catenin
accumulation and endothelial cell proliferation has recently been suggested (Biswas,
Canosa et al. 2003).
Thrombospondin 1 (TSP-1) is a glycoprotein with major roles in cellular adhesion and
vascular smooth muscle cell proliferation and migration and is also a well known
inhibitor of angiogenesis (Tucker 2004). It has been already shown in the past that
exposure of endothelial cells to hypoxic environment up regulates the TSP-1 gene and
protein expression (Phelan, Forman et al. 1998). Although less clear and concordant,
other data of the literature have shown both in in vitro (Chen, Yang et al. 2004) and in
in vivo (Berg, Breen et al. 1998) studies that levels of fibronectin protein and/or
transcript can also be up regulated by exposure to hypoxia.
123
Recent methodologies based on the multidimensional chromatographic separation of
membrane fractions at extreme pH values (Wu, MacCoss et al. 2003) or protein
biotinylation (Peirce, Wait et al. 2004; Zhao, Zhang et al. 2004) have allowed the
identification of a large number of membrane proteins. In contrast to 2D peptide
mapping, these methods, however, are not inherently quantitative. While our
technology is applicable to several biological systems, the most immediate challenges
are likely to include the comparison of closely related cell lines (Speers and Cravatt
2004) (e.g., metastastic vs. non-metastatic cancer cell lines (Clark, Golub et al.
2000)), or cells of the immune system at different stages of activation (e.g., dendritic
cells or natural killer cells).
124
3.4. In vivo protein biotinylation for the identification of organ-specific antigens
accessible from the vasculature.
The endothelium is a highly dynamic structure, morphologically and functionally
adapted to meet the unique needs of the underlying tissue (Madri and Williams 1983;
Auerbach, Alby et al. 1985; Aird, Edelberg et al. 1997). Recently, a comprehensive
description of endothelial cell (EC) diversity as a result of differences in vascular
beds, differentiation programs, and distinct adaptation to physiological and
pathological changes has been proposed (Chi, Chang et al. 2003). Indeed, different
patterns of global gene expression are emerging for ECs of arterial, venous and
lymphatic origin (Lawson and Weinstein 2002; Chi, Chang et al. 2003; Hirakawa,
Hong et al. 2003; Veikkola, Lohela et al. 2003; Yamashita 2004) . In addition, vessel
size (Muller, Hermanns et al. 2002), flow properties (Braddock, Schwachtgen et al.
1998), anatomical location (Pasqualini and Ruoslahti 1996; Chi, Chang et al. 2003),
physiological and developmental processes (Cleaver and Melton 2003) further
modulate EC gene expression. Other lines of evidence (Wyder, Vitaliti et al. 2000;
Oh, Li et al. 2004) indicate that different tumors can have a differential influence on
the expression of EC surface proteins. Furthermore, the ubiquitous distribution of
capillaries in all tissues and the partial accessibility of parenchymal cells via stromal
capillaries define body compartments, which are accessible to agents coming from the
bloodstream.
The high specialization of endothelia from different anatomical locations and
pathological conditions makes it possible to carry out a selective molecular targeting
of vascular structures, with immediate implications for imaging and therapy
125
(Pasqualini and Ruoslahti 1996; Neri, Carnemolla et al. 1997; Birchler, Viti et al.
1999; Nilsson, Kosmehl et al. 2001; Posey, Khazaeli et al. 2001; Borsi, Balza et al.
2002; Carnemolla, Borsi et al. 2002; Castellani, Borsi et al. 2002; Halin, Rondini et al.
2002; Borsi, Balza et al. 2003; Halin, Gafner et al. 2003; Nanus, Milowsky et al.
2003; Santimaria, Moscatelli et al. 2003). Advances in this field will require an in-
depth proteomic analysis of ECs from different anatomical locations or pathological
conditions, followed by an extensive validation of the putative markers identified and
the development of specific binding molecules (e.g., human antibodies (Winter,
Griffiths et al. 1994)).
Jan Schnitzer and coworkers have pioneered the use of colloidal silica for the in vivo
coating of vascular structures in tumors and in normal organs (Jacobson, Schnitzer et
al. 1992; Durr, Yu et al. 2004). This physical modification allowed the isolation (by
centrifugation and fractionation) of silica-coated structures (luminal cell plasma
membranes and caveolae of the endothelium), providing enough material for
proteomic investigations, for example by immunization or by 2D-PAGE. Two
candidate targets, aminopeptidase P and annexin A1 were identified in lung and solid
tumors, respectively, and were validated by immunohistochemistry and scintigraphic
imaging (Oh, Li et al. 2004).
In principle, it would be desirable to characterize antigens accessible via the
vasculature using chemical methods, which allow the covalent modification, recovery
and identification of accessible proteins. Such methods would offer the possibility to
perform unbiased chemical reactions with all proteins in a certain compartment. The
chemical properties of the reagent (charge, lipophilicity, size, etc.) could be shaped to
126
control its distribution (in vivo or ex vivo), thus influencing the class of proteins to be
preferentially labeled for proteomic investigations.
In this paper, we describe the terminal perfusion of tumor-bearing mice with
sulfosuccinimidyl-6-(biotinamide)-hexanoate (sulfo-NHS-LC-biotin) for the in vivo
covalent modification of accessible proteins in normal organs and in solid tumors.
The resulting biotinylated proteins can then be recovered from the excised organ and
tumor tissue by purification on streptavidin resins, followed by on-resin proteolytic
digestion and mass spectrometric analysis. Our methodology led to reliable,
reproducible and efficient in vivo labeling of structures accessible from the
bloodstream. A number of biotinylated, organ-specific proteins could be identified in
the mouse kidney, liver, muscle and heart, as well as in two experimental tumor
models.
3.4.1. Terminal perfusion and in vivo biotinylation
Figure 21 illustrates the relevant steps of the in vivo biotinylation method, for the
discovery of markers preferentially expressed in different organs or in sites of disease
(such as solid tumors). The approach is based on the terminal perfusion of rodents
with a broadly reactive derivative of biotin. We used a charged active ester derivative
of biotin (sulfo-NHS-LC-biotin; Pierce), with an impaired diffusion through
biological membranes (Dentler 1995), but other reactive biotin derivatives could be
considered (see below). As a result, accessible proteins and certain amine-containing
glycolipids and phospholipids can be covalently modified with biotin. Biotinylated
proteins can be efficiently purified on streptavidin resins and submitted to a
comparative proteomic analysis, thus revealing markers which are differentially
127
Target tissues (healthy organs, tumor)
Blood
Terminal perfusion of tumor-bearing mice
Extraction (SDS) and purification of biotinylated proteins on SA-Sepharose
Comparative proteomic analysis Target
identification
Biotinylation of accessible proteins
O S NH HN
O S NH HN
O
S NH
HN
O S
NH
HN
O
S
NH
HN
O
S
NH
HN
O
S
NH
HN
O
S
NHHN
O
S
NHHN
O
S
NHHN
Figure 21. Schematic representation of the relevant steps of the in vivo biotinylation method. Rodents are anaesthetized and subjected to the terminal perfusion with a broadly reactive derivative of biotin. As a result, accessible proteins and certain amine-containing glycolipids and phospholipids can be covalently modified with biotin. The excess of unreacted biotin derivative is quenched with Tris. Organs of interest and tumors are excised, homogenized and a total protein extract is prepared. Biotinylated proteins can be efficiently purified on streptavidin resins and submitted to a comparative proteomic analysis, thus revealing markers that are differentially expressed in organs and diseased tissues.
expressed in organs and diseased tissues. Healthy organs and the tumor tissue from
two different models [subcutaneous F9 teratocarcinoma in SvEv129 mice (Neri,
Carnemolla et al. 1997) and the RENCA carcinoma grafted orthotopically in a kidney
of BALB/c mice (Ahn, Jung et al. 2001)] were investigated.
128
The large circulation of mice was perfused at a constant pressure of 100 mm Hg
through the left heart ventricle by means of a butterfly cannula fitted with a barb.
Mice were perfused first with PBS to wash away blood components, and then with a 1
mg/ml sulfo-NHS-LC-biotin solution, followed by a quenching step with PBS
containing a 50 mM solution of the primary amine Tris (hydroxymethyl)
aminomethane (Tris). All perfusion solutions contained Dextran 40 as a plasma
expander, and were pre-warmed at 38°C. Omitting the PBS wash step prior to in vivo
biotinylation improved the perfusion of the microvasculature of subcutaneously
induced tumors. Organs of interest and tumors were excised and subjected to further
investigation.
3.4.2. Histochemical analysis of biotinylated structures in organ and tumor sections
Histochemical analysis of organ and tumor sections with streptavidin-alkaline
phosphatase confirmed that discrete structures were labeled as a result of the in vivo
biotinylation reaction in normal organs and in tumors (Fig. 22). Normal organs
resulted to be easily accessible to the biotinylation reagent. Skeletal muscle, tongue,
kidney, liver (Fig. 22 e-h) and heart (not shown) of every biotinylated mouse showed
positive staining, while in negative control mice perfused with saline (Fig. 22 a-d) the
staining reaction was essentially negative. As expected, kidneys showed the strongest
staining, with a preferential accumulation in the tubular structures of the cortex and in
the glomeruli (Fig. 22 g). The liver was mainly stained around vascular structures
(Fig. 22 h). The biotinylation reagent reached also the liver parenchyma, although to
a variable extent in different portions of the liver as well as in different mice. In
muscle, strong staining was found around vessel structures as well as in the
129
Figure 22. Histochemical and immunofluorescence analysis of organ and tumor sections. Organs from saline-perfused (a-d; i) or sulfo-NHS-LC-biotin perfused (e-h; j-l) F9 tumor-bearing SvEv129 mice were snap-frozen in cryoembedding medium. Sections from skeletal muscle (a,e), tongue (b, f) kidney (c, g), liver (d, h) or F9 tumor (i-j) were cut and subjected to histochemical staining with streptavidin:biotinylated alkaline phosphatase complex, followed by incubation with a freshly prepared solution of Fast-Red TR (a-j). Other section from F9 tumor were incubated with rat anti-mouse CD31 (1:100), followed by simultaneous incubation with goat anti-rat IgG-Cy3 conjugate (l) and streptavidin-Alexa Fluor 488 (k) to investigate the distribution of biotinylated structures relative to the localization of tumor endothelial cells. Bars = 100 µm intercellular spaces between the muscle fibers, where the endomysial capillaries are
located (Fig. 22 e). By contrast, the inner parts of the muscle fibers were not stained.
The same was true for the skeletal muscle fibers in the tongue where, in addition, the
lamina propria (sub-epithelial connective tissue containing numerous glands) was
intensely stained (Fig. 22 f).
Tumor tissues could be successfully biotinylated, with a preferential staining of
vascular structures; a section of a subcutaneous F9 tumor in a biotinylation reagent-
perfused SvEv mouse is shown in Fig. 22 j. However, the perfusion of tumors was
much less efficient and more heterogeneous compared to normal organs. While some
tumors showed a homogenous staining of virtually all vessels, more often the staining
was heterogeneous throughout the solid mass. An immunofluorescence co-staining of
tumor sections for biotin (Fig. 22 k) and the endothelial marker CD31 (Fig. 22 l)
130
revealed the co-existence of neighboring vascular structures, which were labeled to a
different extent. With the original 3-step perfusion protocol, only <10% of tumors
could be biotinylated successfully (42 perfused mice), whereas about 30% of tumors
showed successful staining when the PBS wash step was omitted (28 perfused mice;
data not shown). This modification had no effect on the perfusion efficiency of
normal organs.
3.4.3. Purification and identification of biotinylated proteins by proteomic techniques
Total proteins were extracted from the different excised organs or tumors and
quantified as described in Materials and Methods. Biotinylated proteins could be
Figure 23. Chromatographic elution profiles of tryptic peptides from healthy and tumor kidney in three independent RENCA mice. Tryptic peptides obtained by digestion of biotinylated proteins in samples of healthy kidney (left) or tumor kidney (right) from three independently perfused RENCA mice were subjected to microcapillary liquid chromatography-tandem mass spectrometry for protein identification. Chromatographic elution profiles show a good reproducibility of the in vivo biotinylation reaction in different experiments.
131
efficiently captured on streptavidin-Sepharose (SA) resin slurries, even in the
presence of strong anionic detergents, and the bulk of non-biotinylated proteins
washed away by increasingly stringent washing steps (Rybak, Scheurer et al. 2004).
We then performed a digestion step with trypsin while the proteins were still bound to
the SA-Sepharose resin. The resulting tryptic peptides were subjected to
microcapillary liquid chromatography-tandem mass spectrometry for protein
identification. Although a number of peptides from streptavidin could also be detected
(particularly in samples from saline-perfused control mice), this did not compromise
the analysis of peptides from biotinylated organs. Overall, the procedure turned out to
be extremely reproducible. Figure 23 shows the chromatographic profile of tryptic
peptides obtained from biotinylated proteins of healthy kidneys and of tumor bearing
kidneys from three independent RENCA mice.
Strikingly, the profiles exhibit consistent differences among normal and tumor
kidneys. A good reproducibility could also be seen at the level of protein
identification, finding in most cases the same peptides in specimens from different
mice (Supplementary Table 2 in Appendix B). The percentage of proteins
reproducibly identified in three independent mice ranged from 60% in the tumor, to
almost 70% in muscle and kidney, to more than 85% in the liver.
Table 5 shows a list of some of the proteins that could be identified in all tissues, or
solely in one of the organs or tumors. Figure 24 shows an analysis of the subcellular
distribution of the identified proteins, based on their SwissProt database accession
number, in the kidney, liver, skeletal muscle, heart, as well as F9 and RENCA tumors.
Cell surface proteins and extracellular matrix proteins were present in all organs
examined, in a proportion that ranged cumulatively between 20 and about 50%.
132
Table 5. Proteins identified in all tissues, or selectively in only one organ or only in tumors
Figure 24. Subcellular localization of proteins identified Subcellular localization of the proteins identified in kidney (n=155), liver (n=152), skeletal muscle (n=56), heart (n=62), F9 tumor (n=44) and RENCA tumor (n=85). A legend reporting the correspondence between subcellular compartments and color-coding of the pie chart is shown in the lower part of the panel.
However, cytoplasmic and cytoskeletal proteins were also significantly represented,
as well as blood components, which were particularly abundant in the tumor and
muscle samples.
A few proteins could be identified in the organs and tumors from the control mice
groups, which had been perfused with a saline solution. These proteins (mainly
pyruvate carboxylase, propionyl CoA carboxylase and methylcrotonyl CoA
carboxylase) are enzymes that have multiple covalent binding sites for biotin and use
this vitamin as a cofactor. In muscle of control mice, a few other proteins (isoforms of
myosin and actin) could also be identified.
134
3.4.4. Validation of some candidate marker proteins identified
In order to validate the results of the mass spectrometry analysis and confirm the
patterns of expression of the proteins identified, we performed a Western Blot and
immunofluorescence analysis. Proteins extracted from heart, kidney, liver, muscle and
tumors were normalized using the bicinchoninic acid assay and loaded on replicate
gels, separated by SDS-PAGE (Fig 25, panel A), then blotted on nitrocellulose.
Membranes were probed with commercial antibodies against Kidney-specific
A B
Figure 25. Western Blot analysis of organ–specific markers
(A) Equal amounts of heart, kidney, liver, skeletal muscle, F9 and RENCA tumor total protein extracts from a saline–perfused SvEv mouse were separated by SDS–PAGE and stained by Coomassie Brilliant Blue staining. Molecular weight markers are indicated on the left. (B) Other replicate gels were directly transferred to nitrocellulose and a staining of bands immunoreactive towards anti–kidney specific cadherin 16 (anti–Ksp cadherin), anti–calcium channel L–type DHPR alpha 1 subunit, (anti–Cacna1s), anti–calcium channel, voltage gated α2/δ–1 subunit (anti–Cacna2d1) anti–protein tyrosine kinase 7 (anti–PTK7) and anti–T lymphoma invasion and metastasis inducing protein 1 (anti–TIAM1) was performed, followed by incubation with the appropriate secondary antibody conjugated to horseradish peroxidase and ECL detection.
al. 1995) is a member of the cadherin superfamily of cell adhesion molecules, whose
expression in the kidneys is confined to the basolateral membranes, facing the
intercellular spaces, of tubular epithelial cells (Thomson and Aronson 1999). During
the in vivo perfusion procedure, these structures are accessible to the biotinylation
reagent through the small fenestrated peritubular capillaries (Figure 26, upper left
panel).
In cross section of skeletal muscle (Figure 26, lower left panel) the anti calcium
channel L type DHPR alpha 1 subunit antibody related-immunostaining is seen in the
136
+ -
Figure 26. Immunofluorescence detection of organ–specific markers Cryostat sections from mice kidney and skeletal muscle, fixed with 3% PFA + 0.01% GA by vascular perfusion; the binding sites of the primary mouse antibody are revealed by FITC–conjugated anti mouse IgG. Nuclear DNA is stained by DAPI, added to the secondary antibody. In the kidney (upper panel, +) ksp–cadherin 16 is seen in lateral cell membranes (green staining) bordering the intercellular spaces between the tubular epithelial cells. The extent of lateral cell membranes varies among nephron segments. Cell nuclei are shown in red. In cross section of skeletal muscle (lower panel, +) the anti calcium channel L type DHPR alpha 1 subunit antibody related–immunostaining is seen in the transverse tubule system, surrounding the muscle fibers and in a fine punctuate pattern within the fibers. In negative controls (–) the primary antibody was omitted. The weak background staining of the basement membranes is due to the cross reaction of the secondary anti mouse antibody with endogenous mouse IgG. Bars = 50 µm.
transverse tubule system, surrounding the muscle fibers and in a fine punctuate
pattern within the fibers.
137
3.4.5. Discussion
A number of laboratories, including ours, have experimentally demonstrated (in
animal models and in patients with cancer) that monoclonal antibodies specific to
tumor-associated vascular antigens can be useful for the diagnosis and therapy of
cancer and possibly of other angiogenesis-associated diseases (Bicknell 2002; Carver
and Schnitzer 2003; Brack, Dinkelborg et al. 2004; Thorpe 2004). Future advances in
this field will crucially rely on the identification of vascular antigens, which are
preferentially expressed in certain pathological conditions. Furthermore, a proteome-
wide analysis of accessible proteins may increase our understanding of the dynamic
nature of the endothelium in physiological processes.
Until now, the search for tumor-associated vascular antigens has mainly relied on
serendipitous discovery (Zardi, Carnemolla et al. 1987; Liu, Moy et al. 1997;
Carnemolla, Castellani et al. 1999), on immunization strategies (Buhring, Muller et al.
1991), on transcriptomic analysis of tumor-derived endothelial cells (St. Croix, Rago
et al. 2000; Wyder, Vitaliti et al. 2000) and on bioinformatics approaches (Gerritsen,
Soriano et al. 2002; Huminiecki, Gorn et al. 2002). Furthermore, the groups of
Ruoslahti and Pasqualini have pursued the in vivo panning of peptide phage libraries
for the discovery of “signature peptides” capable of identifying vascular structures in
normal organs and in tumors (Pasqualini and Ruoslahti 1996). A comparatively
smaller number of proteomic investigations have also been performed, hampered by
the difficulty to recover, by cell purification (Alessandri, Chirivi et al. 1999) or by
laser capture microdissection (Craven, Totty et al. 2002), a sufficient amount of pure
endothelial cells either from tumor or from normal vasculature.
138
In principle, the most direct way to assess differences in protein abundance between
the tumor neo-vasculature and normal vasculature would consist in the in vivo
chemical labeling of accessible structures (as described in this thesis) or in the
physical separation of surface proteins [as pioneered by Jan Schnitzer (Jacobson,
Schnitzer et al. 1992; Durr, Yu et al. 2004)], followed by a rapid recovery and
comparative proteomic analysis of the two samples. De la Fuente et al. have described
the artificial ex vivo perfusion of lungs isolated from normal and hyperoxic rats with
sulfo-NHS-LC-biotin (De La Fuente, Dawson et al. 1997). After SDS-PAGE, the
biotinylated proteins were visualized using a chemiluminescence substrate for the
streptavidin-horseradish peroxidase conjugate, outlining differences in rats exposed to
hyperoxia for 48-60 hours (De La Fuente, Dawson et al. 1997). However, the
procedure did not allow protein identification.
Our method offers a number of potential advantages compared to other methods
aimed at characterizing accessible proteins in normal tissues and in sites of disease.
First, unlike 2D-gel-based techniques, our approach is compatible with the use of
strong anionic detergents (e.g., SDS), which are indispensable for the solubilization of
most membrane proteins and extracellular matrix components during the critical
phase of protein extraction from the tissues. Second, the chemical properties of the
reactive organic molecule used in the perfusion reaction can be changed, in order to
influence the structures, which are accessible, in vivo and which can be labeled and
recovered. Indeed, it should be possible to choose labeling reagents, which mimic the
pharmacokinetic properties of the targeting agents that one intends to use for imaging
or therapeutic applications. For example, we can alter the charge, size (e.g., by
conjugation to dextran) and lipophilicity of the reactive derivatives of biotin. Indeed,
139
the approach is not limited to biotin, and any other reactive molecule for which a
high-affinity reagent is available could be considered (e.g., fluorophores and the
corresponding antibodies). Third, the proteomic analysis described in this thesis is
easy to implement, sensitive, and allows the identification of hundreds of different
accessible proteins in the tissues of interest. The method is biased towards the most
abundant antigens, which are also the preferred ones for ligand-based tumor targeting
applications (Brack, Dinkelborg et al. 2004). However, advances in mass
spectrometry (Geoghegan and Kelly 2004) or the use of enzyme-specific labeling
methods (Adam, Sorensen et al. 2002) may facilitate the identification of less
abundant tissue specific antigens.
In our analysis, a number of proteins were found to be biotinylated in several tissues
(e.g., glyceraldehyde 3-phosphate dehydrogenase, collagen type VI (alpha 3), actin;
Supplementary Table 1). In addition, a number of previously known organ-specific
antigens were found to be expressed only in the expected organ (e.g., kidney
transporters, muscle and cardiac channels; Table 5 and Appendix B). Interestingly, in
a given organ we were able to identify both organ-specific proteins [e.g., kidney
specific transmembrane protein 27 (Zhang, Wada et al. 2001), kidney specific
transport protein (Lopez-Nieto, You et al. 1997) and Na+K+ transporting ATPase
gamma chain (Forbush, Kaplan et al. 1978)] and proteins which are known to be
expressed ubiquitously (e.g., Na+K+ transporting ATPase alpha-1 chain in kidney).
The fact that these last proteins could be identified only in one organ might reflect
either a difference in their relative abundance among different organs, or a facilitated
accessibility in that particular organ, or both. Moreover, some hypothetical proteins
140
and proteins of unknown function could also be specifically identified in different
organs.
The results obtained by in vivo biotinylation can differ in some cases from previous
transcriptomic studies of genes expressed in human tumor endothelium. For example,
St. Croix et al. (St. Croix, Rago et al. 2000) described collagen type IV (alpha 2) as a
pan-endothelial marker, while presented collagen type VI (alpha 3), nidogen and
collagen type I as genes up-regulated in angiogenic vessels. In our analysis, collagen
type VI and nidogen were found in various normal organs as well as in tumors, while
collagen type IV was preferentially found in both tumors studied. However, St. Croix
and colleagues analyzed endothelial cells from colorectal tumors and normal mucosa
specimens, while we studied mouse organs and transplanted tumors, in which
basement membrane is more abundant. Indeed, a number of studies have shown that
collagen type IV is abundantly expressed by basal lamina in solid tumors. In clear cell
type renal carcinomas, Droz and colleagues found collagen type IV (alpha 1) and
collagen type IV (alpha 2) in basal laminae surrounding tumor islets (Droz, Patey et
al. 1994). Histochemical studies performed on tissues from 11 patients showed that
collagen type IV was always present in the basement membrane surrounding tumor
nests of basal cell carcinoma, although tumors with different degrees of invasivity
showed a different distribution of the collagen type IV chains (Tanaka, Iyama et al.
1997). Confocal microscopic studies (Baluk, Morikawa et al. 2003) have shown that
basement membrane (of which collagen type IV is the major constituent) covered
>99.9% of the surface of blood vessels in the three different tumor types examined
(Baluk, Morikawa et al. 2003). Our results could reflect a preferential accessibility of
collagen type IV in tumor basement membrane, due to the increased fenestration of
tumor blood vessels (Maeda, Fang et al. 2003).
141
Some of the antigens identified exclusively in tumor specimens have already been
shown to be activated or over expressed in solid tumors or in in vitro tumor cell lines.
For example, tumor tissues from patients with a histopathological diagnosis of
glioblastoma multiforme or anaplastic astrocytoma were analyzed for activity of the
lysosomal enzyme glucosylceramidase (and other lysosomal enzymes) which was
found to have an increased activity, as compared with that in normal brain tissue
(Nygren, von Holst et al. 1997).
A vast literature has demonstrated an association between integrin alpha V over
expression and tumor cell proliferation (Levinson, Hopper et al. 2002; Cruet-
Hennequart, Maubant et al. 2003) or tumor propensity to metastasize to distal organs
(Pecheur, Peyruchaud et al. 2002). Human ovarian OV-MZ-6 cancer cells express the
integrin alpha(v)beta3, which associates with vitronectin and correlates with ovarian
cancer progression (Hapke, Kessler et al. 2003). Our experiments show a selective in
vivo biotinylation of alphaVβ3 integrin in tumors, mirroring magnetic resonance
imaging results obtained targeting a paramagnetic contrast agent to endothelial
alphaVβ3 in rabbit carcinomas through a specific monoclonal antibody (Sipkins,
Cheresh et al. 1998). Indeed, alphaVβ3 integrin has been proposed as a marker of
tumor angiogenesis (Auerbach and Auerbach 1994; Kumar 2003) and a humanized
monoclonal antibody to alphaVβ3 (Vitaxin) (Gutheil, Campbell et al. 2000) is in phase
II clinical trials as a therapeutic agent for blocking tumor-induced angiogenesis in
melanoma and prostate cancer (Tucker 2003).
Transketolase, a thiamine diphosphate-dependent enzyme linking the nonoxidative
branch of the pentose phosphate pathway to the glycolytic pathway, has been shown
to control in vivo tumor growth in mice with Ehrlich’s ascites tumor (Comin-Anduix,
142
Boren et al. 2001), and to be over expressed in highly metastatic adenoid cystic
carcinoma cell lines compared with their poorly metastatic counterparts (An, Sun et
al. 2004). Studies reported in the Anatomical Viewer of the SAGE Genie website
(Boon, Osorio et al. 2002) (http://cgap.nci.nih.gov/SAGE) show that transcripts
coding for transketolase are abundant only in white blood cells of the healthy adult.
By contrast, high levels of mRNAs for this enzyme can be found in lung, kidney and
colon tumor tissues.
Finally, acetyl CoA carboxylase isozymes have been reported to show distinctive
tissue distribution and regulation. The polypeptide of Mr 265,000 (Acetyl CoA
carboxylase 265) is the sole form expressed in white adipose tissue, while the isozyme
Acetyl CoA carboxylase 280 predominates in cardiac and skeletal muscle (Thampy
1989; Bianchi, Evans et al. 1990; Iverson, Bianchi et al. 1990; Louis and Witters
1992). Both forms are present in liver, brown adipose tissue, lactating mammary
gland and pancreatic islets (Thampy 1989; Bianchi, Evans et al. 1990; Iverson,
Bianchi et al. 1990; Louis and Witters 1992). Recently, acetyl CoA carboxylase 265
has been identified as a partner of the protein encoded by the breast cancer
susceptibility gene BRCA1 (Magnard, Bachelier et al. 2002). Our studies show that
acetyl CoA carboxylase 265 is either more abundant and/or more easily accessible in
the tumors tested, than in other normal tissues, thus suggesting its use as antigen for
ligand-based tumor targeting applications.
A couple of other proteins, specifically identified in tumors, have not yet been
extensively characterized as tumor markers and represent interesting candidates for
development of targeting antibodies. We identified the murine homolog of the human
PTK7 protein. PTK7 is a receptor protein tyrosine kinase-like molecule that lacks
Tubulin beta-1 chain or 30 P07437 (R) FPGQLNADLR 26 Tubulin beta-1 chain or Q9H4B7 (K) LAVNMVPFPR 31 Tubulin beta-2 chain or P05217 (R) AILVDLEPGTMDSVR 31 Tubulin beta-4 chain or Q13509 Similar to tubulin, beta 4 or Q9BUF5 Tubulin beta-5 chain or P05218 Tubuin beta-5 chain or P04350 Tubulin beta-4q chain or Q99867 Beta-tubulin 4q or Q8WZ78
Hypothetical protein or Q969E5 Tubulin, beta polypeptide or Q9BVA1 Probable beta tubulin (Fragment) or O43209 Beta tubulin or Q13885 Tubulin, beta 4 Q8WUL7
of 2-oxoglutarate dehydrogenase complex, mitochondrial [Precursor] or Alpha-ketoglutarate dehydrogenase complex Q16187 dihydrolipoyl succinyltransferase or
Human full-length cDNA 5-PRIME end of clone Q86TQ8
O09164 F Fatty acid transport protein 2 O88560 Fibrinogen A alpha polypeptide Q99K47 Fibrinogen, B beta polypeptide Q8K0E8 Fibronectin precursor P11276 Fructose-bisphosphate aldolase B Q91Y97 Fumarylacetoacetase P35505 G
Ubiquinol-cytochrome C reductase complex core protein I
Vacuolar ATP synthase catalytic subunit A, ubiquitous isoform Very-long-chain acyl-CoA synthetase
Proteins marked in red were identified only in kidney extracts
237
1300018L09Rik protein Q9DBB818 days embryo cDNA, RIKEN full-length enriched library Q9D160 2-hydroxyphytanoyl-CoA lyase Q9QXE03,2-trans-enoyl-CoA isomerase, mitochondrial precursor P42125 3-hydroxy-3-methylglutaryl-coenzyme A lyase Q8QZS6 3-ketoacyl-CoA thiolase Q8BWT14-hydroxyphenylpyruvate dioxygenase P49429 4-trimethylaminobutyraldehyde dehydrogenase Q9JLJ2 60 kDa heat shock protein, mitochondrial precursor P19226 A Actin, aortic smooth muscle P03996 Actin, cytoplasmic 1 P02570 Acyl-CoA-binding protein P31786 Acyl-coenzyme A oxidase 1, peroxisomal Q9R0H0 Adenosylhomocysteinase P50247 Alcohol dehydrogenase [NADP+] Q9JII6 Alcohol dehydrogenase A chain P00329 Alcohol dehydrogenase class III P28474 Aldehyde dehydrogenase 1A1 P24549 Aldehyde dehydrogenase, mitochondrial precursor P47738 Aldo-keto reductase family 1 member C13 Q8VC28 Alpha enolase P17182 Amine oxidase Q8C0B2 Amylo-1 Q8CE68 Apolipoprotein C-III precursor P33622 Apolipoprotein E precursor P08226 Arginase 1 Q61176 Argininosuccinate synthase P16460 Asialoglycoprotein receptor 1 (Hepatic lectin 1) P34927 Aspartate aminotransferase, cytoplasmic P05201 Aspartate aminotransferase, mitochondrial precursor P05202 ATP synthase alpha chain, mitochondrial precursor Q03265 ATP-binding cassette, sub-family D, member 3 P55096 ATP-citrate synthase Q91V92 B Basigin precursor P18572 Beta-alanine oxoglutarate aminotransferase Q8BZA3 Betaine-homocysteine S-methyltransferase O35490 C Calcium-binding mitochondrial carrier protein Aralar2 Q9QXX4Carbonic anhydrase III P16015 Carboxylesterase 3 Q8VCT4 Catalase P24270 Cathepsin B precursor P10605 Cis-retinol/androgen dehydrogenase type 3 Q8K5C8 Cystathionine gamma-lyase Q8VCN5D Delta-aminolevulinic acid dehydratase P10518 Dipeptidyl peptidase IV P28843 E
238
Ectonucleotide pyrophosphatase/phosphodiesterase 1 P06802 Electron transfer flavoprotein alpha-subunit Q99LC5 Elongation factor 1-alpha 1 P10126 Elongation factor 2 P58252 Endoglin precursor Q63961 Endoplasmin precursor P08113 Esterase 10 Q9CWI4 Estradiol 17 beta-dehydrogenase 5 P70694 F Fatty acid synthase Q9EQR0 Fatty acid transport protein 2 O88560 Fatty acid-binding protein, liver P12710 Fibrinogen, B beta polypeptide Q8K0E8 Fibronectin precursor P11276 Fructose-1,6-bisphosphatase Q9QXD6Fructose-bisphosphate aldolase B Q91Y97 Fumarylacetoacetase P35505 G Gamma-glutamyltransferase 5 precursor Q9Z2A9 Gamma-glutamyltransferase-like activity 1 Q8C7B4 Glutamate dehydrogenase, mitochondrial precursor P26443 Glutaryl-CoA dehydrogenase, mitochondrial precursor Q60759 Glutathione peroxidase P11352 Glutathione S-transferase Mu 1 P10649 Glutathione S-transferase theta 2 Q61133 Glutathione S-transferase Yc P30115 Glutathione transferase omega 1 O09131 Glyceraldehyde 3-phosphate dehydrogenase P16858 Glycogen phosphorylase, liver form Q9ET01 H H-2 class I histocompatibility antigen, D-B alpha chain precursor P01899 H-2 class I histocompatibility antigen, K-B alpha chain precursor P01901 Hemoglobin alpha, adult chain 1 Q9CY10 Hemoglobin beta-1 chain P02088 Histidine ammonia-lyase P35492 Histone H2A.1 P22752 Histone H2B F P10853 Homogentisate 1,2-dioxygenase O09173 Hydroxymethylglutaryl-CoA lyase, mitochondrial precursor P38060 Hydroxymethylglutaryl-CoA synthase P54869 Hypothetical protein Q91W19 Hypothetical protein Q8CC86 Hypothetical protein (Activated leukocyte cell adhesion molecule) Q8R2T0 Hypothetical protein (Fragment) Q99KD0 Hypothetical protein (Staphylococcal nuclease domain containing 1) Q922L5 I Integrin beta-1 precursor P09055 Isocitrate dehydrogenase [NADP] cytoplasmic O88844 K Keratin, type II cytoskeletal 1 P04104
239
Ketohexokinase P97328 L L-lactate dehydrogenase A chain P06151 M Macrophage mannose receptor precursor Q61830 Maleylacetoacetate isomerase Q9WVL0Methylcrotonyl-CoA carboxylase alpha chain Q99MR8N N system amino acids transporter NAT-1 Q9JLL8 NADH-ubiquinone oxidoreductase chain 4 P03911 NADH-ubiquinone oxidoreductase chain 5 P03921 NADP-dependent malic enzyme P06801 NDRG2 protein Q9QYG0Neural-cadherin precursor (N-cadherin) P15116 O Ornithine carbamoyltransferase, mitochondrial precursor P11725 P Peptidyl-prolyl cis-trans isomerase A P17742 Peroxiredoxin 6 O08709 Peroxisomal multifunctional enzyme type 2 P51660 Phosphatidylethanolamine-binding protein P70296 Phosphoglucomutase Q9D0F9 Phosphoglycerate kinase 1 P09411 Platelet glycoprotein IV Q08857 Probable urocanate hydratase Q8VC12 Propionyl-coenzyme A carboxylase, alpha polypeptide Q80VU5 Pyruvate carboxylase Q05920 R RIKEN cDNA 1500002K10 gene Q91VS7 S Scavenger receptor class B member 1 Q61009 SEC14-like protein 2 Q99J08 Selenium-binding protein 1 P17563 Senescence marker protein-30 Q64374 Serum albumin precursor P07724 Short chain 3-hydroxyacyl-CoA dehydrogenase Q61425 Similar to acetyl-CoA acyltransferase, 3-oxo acyl-CoA thiolase A, peroxisomal
Q8VCH0
Similar to acetyl-coenzyme A acyltransferase 2 Q8JZR8 Similar to aldehyde dehydrogenase 4 family, member A1 (Fragment) Q8CHT0 Similar to aldehyde dehyrdogenase 8 family, member A1 Q8BH00 Similar to CG6385 gene product Q99LB7 Similar to DJ-1 protein Q99LX0 Similar to DKFZP586B1621 protein Q8VC30 Similar to fibrinogen, gamma polypeptide Q8VCM7Similar to glutamic-pyruvate transaminase Q8QZR5 Similar to glycine N-methyltransferase Q91WN7Similar to methionine adenosyltransferase I, alpha Q91X83 Similar to mitochondrial aconitase Q99KI0 Sodium/potassium-transporting ATPase beta-1 chain P14094
240
Sodium/potassium-transporting ATPase beta-3 chain P97370 Soluble epoxide hydrolase P34914 Solute carrier family 21 member 10 Q9JJL3 Solute carrier family 25, member 13 Q9QXX4Sorbitol dehydrogenase Q64442 Sulfotransferase-related protein O35403 Superoxide dismutase [Cu-Zn] P08228 T T-cell surface glycoprotein CD1d1 precursor P11609 T-complex protein 1 Q8CAY6Thioether S-methyltransferase P40936 Trifunctional enzyme alpha subunit Q8BMS1Triosephosphate isomerase P17751 U UDP-glucuronosyltransferase 1-1 precursor, microsomal Q63886 V Vitamin D-binding protein precursor P21614 W Weakly similar to carbamoyl-phosphate synthase (Fragment) Q8C196 Proteins marked in blue were identified only in liver extracts
Q9JK53 Propionyl-coenzyme A carboxylase, alpha polypeptide Q80VU5 Pyruvate carboxylase Q05920 S Serine proteinase inhibitor A3K precursor P07759 Serotransferrin precursor Q921I1 Serum albumin precursor P07724 Similar to myosin, heavy polypeptide 2, skeletal muscle, adult Q922D2 T
Heart 1700007H16Rik protein Q9DAM43,2-trans-enoyl-CoA isomerase, mitochondrial precursor P42125 A Actin, alpha cardiac P04270 Actinin alpha 2 Q8K3Q4 Acyl coenzyme A thioester hydrolase, mitochondrial precursor Q9QYR9ADP,ATP carrier protein, heart/skeletal muscle isoform T1 P48962 Alpha-1-antitrypsin 1-1 precursor P07758 Alpha-1-antitrypsin 1-3 precursor Q00896 Alpha-2-macroglobulin precursor Q61838 Antithrombin-III precursor P32261 Aspartate aminotransferase, mitochondrial precursor P05202 ATP synthase alpha chain, mitochondrial precursor Q03265 ATP synthase B chain, mitochondrial precursor Q9CQQ7ATP synthase D chain, mitochondrial Q9DCX2B Basement membrane-specific heparan sulfate proteoglycan core protein precursor
Q05793
C Cardiac Ca2+ release channel Q9ERN6 Carnitine palmitoyltransferase I O35287 Collagen alpha 2(VI) chain precursor Q02788 Creatine kinase, M chain P07310 F Fructose-bisphosphate aldolase A P05064 G Glyceraldehyde 3-phosphate dehydrogenase P16858 H Hemoglobin alpha, adult chain 1 Q9CY10 Hemoglobin beta-1 chain P02088 Hypothetical protein Q91WP2 I Integrin alpha-7 precursor Q61738 Integrin beta-1 precursor P09055 Isocitrate dehydrogenase 2 Q8C2R9 L Laminin alpha-2 chain precursor Q60675 Laminin beta-1 chain precursor P02469 Liver carboxylesterase precursor P23953 L-lactate dehydrogenase A chain P06151 L-lactate dehydrogenase B chain P16125 Lumican precursor P51885 M Malate dehydrogenase, mitochondrial precursor P08249
Proteins marked in green were identified only in muscle extracts
243
Methylcrotonyl-CoA carboxylase alpha chain Q99MR8M-protein O55124 Myoglobin P04247 Myosin heavy chain, cardiac muscle alpha isoform Q02566 Myosin-binding protein C, cardiac-type O70468 N NADH-ubiquinone oxidoreductase chain 4 P03911 NADH-ubiquinone oxidoreductase chain 5 P03921 Neural-cadherin precursor P15116 P Peroxiredoxin 6 O08709 Platelet glycoprotein IV Q08857 Propionyl-coenzyme A carboxylase, alpha polypeptide Q80VU5 Pyruvate carboxylase Q05920 S Sarcoplasmic/endoplasmic reticulum calcium ATPase 2 O55143 Serotransferrin precursor Q921I1 Serum albumin precursor P07724 Short chain 3-hydroxyacyl-CoA dehydrogenase, mitochondrial precursor Q61425 Similar to acetyl-coenzyme A acyltransferase 2 Q8JZR8 Similar to mitochondrial aconitase Q99KI0 Solute carrier family 25, member 13 Q9QXX4T Titin homolog (Fragment) Q8BUJ0 Transthyretin precursor P07309 Trifunctional enzyme alpha subunit Q8BMS1Tubulin alpha-1 chain P05209 Type VI collagen alpha 3 subunit O88493 U Ubiquinol-cytochrome C reductase complex core protein I, mitochondrial precursor
Q9CZ13
Ubiquitin-like 1 Q9CQZ2 V Voltage-dependent anion-selective channel protein 2 Q60930 Voltage-dependent anion-selective channel protein 3 Q60931 Proteins marked in violet were identified only in heart extracts
F9 Tumor 0 day neonate head cDNA, RIKEN full-length enriched library Q9D2K8 4F2 cell-surface antigen heavy chain P10852 A Acetyl-CoA carboxylase 265 (Fragment) Q925C4 Actin, cytoplasmic 1 P02570 Alpha-1-antitrypsin 1-3 precursor Q00896 Alpha-2-macroglobulin precursor Q61838 Apolipoprotein A-I precursor Q00623
244
Apolipoprotein A-IV precursor P06728 B Basement membrane-specific heparan sulfate proteoglycan core protein precursor
Q05793
C P08122
Collagen alpha 2(VI) chain precursor Q02788 Complement C3 precursor P01027 E Elongation factor 1-alpha 1 P10126 F Fibrinogen A alpha polypeptide Q99K47 Fibrinogen, B beta polypeptide Q8K0E8 Fibronectin precursor P11276 G
P17439 Glyceraldehyde 3-phosphate dehydrogenase P16858 H Heat shock protein HSP 90-alpha P07901 Hemoglobin alpha, adult chain 1 Q9CY10 Hemoglobin beta-1 chain P02088 Histone H2B F P10853 I Inter-alpha-trypsin inhibitor heavy chain H3 precursor Q61704 L Liver carboxylesterase precursor P23953 M Methylcrotonyl-CoA carboxylase alpha chain Q99MR8
Q9QXZ0Murinoglobulin 1 precursor P28665 N Nidogen-2 precursor O88322 P Propionyl-coenzyme A carboxylase, alpha polypeptide Q80VU5 Pyruvate carboxylase Q05920 R
Q9DBK8S Serine proteinase inhibitor A3K precursor P07759 Serotransferrin precursor Q921I1 Serum albumin precursor P07724 Similar to fibrinogen, gamma polypeptide Q8VCM7
Q921X9 Similar to myosin, heavy polypeptide 2, skeletal muscle, adult Q922D2
Biglycan precursor P28653 C Cadherin-16 precursor O88338 Carbonic anhydrase XII precursor Q8CI85 Collagen alpha 1(I) chain precursor P11087 Collagen alpha 1(IV) chain precursor P02463 Collagen alpha 1(VI) chain precursor Q04857 Collagen alpha 2(VI) chain precursor Q02788 Collagen type XIV precursor Q80X19 Complement C3 precursor P01027 E Elongation factor 1-alpha 1 P10126 Elongation factor 2 P58252 Esterase 10 Q9CWI4 F Fatty acid transport protein 2 O88560 Fibrinogen A alpha polypeptide Q99K47 Fibrinogen, B beta polypeptide Q8K0E8
Zinc finger protein (Fragment)
Proteins marked in orange were identified only in F9 tumor extracts
246
Fibronectin precursor P11276 G Gamma-glutamyltranspeptidase 1 precursor Q60928 Glyceraldehyde 3-phosphate dehydrogenase P16858 H H-2 class I histocompatibility antigen, K-D alpha chain precursor P01902 Hemoglobin alpha chain P01942 Hemoglobin beta-1 chain P02088 Hemoglobin beta-2 chain P02089 I Integrin alpha-3 precursor Q62470 Integrin alpha-V precursor P43406 Integrin beta-1 precursor P09055 Inter-alpha-trypsin inhibitor heavy chain H3 precursor Q61704 L Laminin alpha-5 chain precursor Q61001 Laminin beta-1 chain precursor P02469 Laminin gamma-1 chain precursor P02468 Lipoprotein receptor-related protein Q91ZX7 L-lactate dehydrogenase A chain P06151 Lutheran blood group Q99K86 M Malate dehydrogenase, mitochondrial precursor P08249 Meprin A alpha-subunit precursor P28825 Methylcrotonyl-CoA carboxylase alpha chain Q99MR8Microsomal dipeptidase precursor P31428 Monocyte differentiation antigen CD14 precursor P10810 Murinoglobulin 1 precursor P28665 N NADH-ubiquinone oxidoreductase chain 4 P03911 Nidogen precursor P10493 Nidogen-2 precursor O88322 Nonmuscle heavy chain myosin II-A Q8VDD5P Pancreas sodium bicarbonate cotransporter O88343 Plasma glutathione peroxidase precursor P46412 Propionyl-coenzyme A carboxylase, alpha polypeptide Q80VU5 Pyruvate carboxylase Q05920 S Serine proteinase inhibitor A3C precursor P29621 Serine proteinase inhibitor A3K precursor P07759 Serotransferrin precursor Q921I1 Serum albumin precursor P07724 Similar to fibrinogen, gamma polypeptide Q8VCM7Similar to mitochondrial aconitase Q99KI0 Similar to tubulointerstitial nephritis antigen Q91XG7 Sodium/potassium-transporting ATPase beta-1 chain P14094 Solute carrier family 25, member 13 Q9QXX4Solute carrier family 4 Q8BUG0T
247
T-lymphoma invasion and metastasis inducing protein 1 Q60610 Transketolase P40142 Transthyretin precursor P07309 Tubulin alpha-1 chain P02551 Tubulin beta-2 chain P05217 Type VI collagen alpha 3 subunit O88493 Type XV collagen O35206 U UDP-glucuronosyltransferase 1-1 precursor, microsomal Q63886 V Vitronectin precursor P29788 Voltage-dependent anion-selective channel protein 1 Q60932 Proteins marked in pink were identified only in RENCA tumor extracts