A Three-Gene Signature for Outcome in Soft Tissue Sarcoma

Abstract

Finding markers or gene sets that would further classify patients into different risk categories and thus allow more individually adapted multimodality treatment regimens in soft tissue sarcomas is necessary. In this study, we investigated the prognostic values of hypoxia-inducible factor 1a (HIF1a), heparin-binding epidermal growth factor-like growth factor (HB-EGF), vascular endothelial growth factor (VEGF), and other angiogenesis-related gene expressions, as well as their interrelationships.
Formalin-fixed paraffin-embedded tissue samples were obtained from 45 patients with soft tissue sarcoma (median age 57 years, range 16-85 years). After laser capture microdissection direct quantitative real-time reverse transcription-PCR (TaqMan) assays were done in triplicates to determine HIF1a, HB-EGF, VEGF, and other gene expression levels.
Multivariate Cox [corrected] regression analysis revealed significant independent associations of HB-EGF, HIF1a, and VEGF-C gene expression to the overall survival (P < 0.0001). A combined factor of these three genes showed a relative risk for shorter survival of 5.5, more than twice higher as in an increasing International Union against Cancer Stage. Receiver operating characteristic curve analysis showed a significant sensitivity of 73% and specificity of 82% of this factor for the diagnosis of short (<3 years) versus long (3-9 years) survival (P = 0.0002). VEGF-A showed significant gender differences in the association to survival.
Measuring HIF1a, HB-EGF, and VEGF-C expression may contribute to a better understanding of the prognosis of patients with soft tissue sarcoma and may even play a crucial role for the distribution of patients to multimodal therapeutic regimens. Prospective studies investigating the response to different adjuvant or palliative therapies seem to be warranted.

Full text

Imaging, Diagnosis, Prognosis
A Three-Gene Signature for Outcome in Soft Tissue Sarcoma
Andreas-Claudius Hoffmann,
1,2,3
Kathleen D. Danenberg,
3
Helge Taubert,
4
Peter V. Danenberg,
2
and Peter Wuerl
5
Abstract Purpose: Finding markers or gene sets that would further classify patients into different
risk categories and thus allow more individually adapted multimodality treatment regi-
mens in soft tissue sarcomas is necessary. In this study, we investigated the prognostic
values of hypoxia-inducible factor 1a (HIF1a), heparin-binding epidermal growth factor–
like growth factor (HB-EGF), vascular endothelial growth factor (VEGF), and other
angiogenesis-related gene expressions, as well as their interrelationships.
Experimental Design: Formalin-fixed paraffin-embedded tissue samples were obtained
from 45 patients with soft tissue sarcoma (median age 57 years, range 16–85 years).
After laser capture microdissection direct quantitative real-time reverse transcription-
PCR (TaqMan) assays were done in triplicates to determine HIF1a, HB-EGF, VEGF,
and other gene expression levels.
Results: Multivariate Cox regression analysis revealed significant independent asso-
ciations of HB-EGF,HIF1a,andVEGF-C gene expression to the overall survival (P<
0.0001). A combined factor of these three genes showed a relative risk for shorter sur-
vival of 5.5, more than twice higher as in an increasing International Union against
Cancer Stage. Receiver operating characteristic curve analysis showed a significant
sensitivity of 73% and specificity of 82% of this factor for the diagnosis of short
(<3 years) versus long (3-9 years) survival (P= 0.0002). VEGF-A showed significant
gender differences in the association to survival.
Conclusions: Measuring HIF1a,HB-EGF,andVEGF-C expression may contribute to a
better understanding of the prognosis of patients with soft tissue sarcoma and may
even play a crucial role for the distribution of patients to multimodal therapeutic regi-
mens. Prospective studies investigating the response to different adjuvant or palliative
therapies seem to be warranted. (Clin Cancer Res 2009;15(16):5191–8)
Although soft tissue sarcomas have an incidence rate of <1%
per year in the United States, they range among the five leading
causes of death in adolescent and young adult (1). The group of
soft tissue sarcomas consists of several histologic types, with
malignant fibrous histiocytomas being very common in pa-
tients older than 40 years and neurogenic sarcomas being more
common in younger ones (2).
Soft tissue sarcomas are considered to be relatively aggressive.
Every second diagnosed patient will ultimately die from the dis-
ease. The tumors are relatively resistant to radio and chemother-
apy (3, 4). Whenever possible patients receive surgery in
combination with radiotherapy, second-line treatment, or treat-
ment of metastatic disease, mainly consists of ifosfamide or da-
carbazine and doxorubicin, but with limited success and no
relevant survival benefit for the combination regimes (5, 6).
The locally invasive growth and the prognosis determining high
rate of metastases often lead to difficulties accomplishing a
complete resection. Therefore, the necessity to find new targets
for tumor therapy is obvious (7).
Imatinib mesylate has been shown to be a successful treat-
ment option in Gastrointestinal Stromal Tumors (GIST-Tumors)
and could also be potentially used in soft tissue sarcomas
(8, 9). Johnson and others recently described the strong induc-
tion of epidermal growth factor–like growth factor (HB-EGF)
release through imatinib mesylate (10). HB-EGF is able to bind
to EGF receptors, the main target of the tyrosine kinase inhibi-
tors (11), with a higher affinity than EGF itself and has recently
been discussed as a potential molecular target (12, 13). Hepar-
anase (HPSE) functions as an endoglycosidase that cleaves he-
paran sulfate chains of proteoglycans (14, 15). Enclosed in the
heparan sulfate glycosaminoglycans are growth factors and an-
giogenic factors, like basic fibroblast growth factor (bFGF) and
Authors' Affiliations:
1
Department of Medicine (Cancer Research), West
German Cancer Center, Molecular Oncology Risk-Profile Evaluation,
University Hospital Essen, Essen, Germany;
2
Department of Biochemistry
and Molecular Biology and Norris Comprehensive Cancer Center,
University of Southern California;
3
Response Genetics, Inc., Los Angeles,
California;
4
Department of Pathology, University of Halle-Wittenberg,
Halle, Germany; and
5
Department of Surgery, Malteser St. Franziskus-
Hospital, Flensburg, Germany
Received 10/2/08; revised 3/13/09; accepted 4/24/09; published OnlineFirst
8/11/09.
The costs of publication of this article were defrayed in part by the payment
of page charges. This article must therefore be hereby marked advertisement
in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
Requests for reprints: Andreas C. Hoffmann, Department of Medicine (Cancer
Research), West GermanCancer Center, Molecular OncologyRisk-ProfileEval-
uation,University HospitalEssen, Hufelandstrasse 55,Essen, 45122, Germany.
Phone: 49- 201-723-85036; Fax: 49-201-723-5733; E-mai l: ach
@
o117.com.
F2009 American Association for Cancer Research.
doi:10.1158/1078-0432.CCR-08-2534
5191 Clin Cancer Res 2009;15(16) August 15, 2009www.aacrjournals.org

HB-EGF that are released upon degradation with HPSE (16).
Thus far, these genes and the interrelationship have not been
further evaluated in soft tissue sarcoma.
Hypoxia-inducible factor 1a (HIF1a) has been described as
one of the main drivers of angiogenesis and has been shown
to correlate with prognosis in many cancers (17, 18). Lately,
Shintani and others used immunohistochemical staining to
show, for the first time, that HIF1a protein overexpression in
malignant fibrous histiocytoma correlates with poor overall
survival (19). Thus far, there has been no approach to measure
mRNA expression levels of HIF1a and the dependent angiogen-
ic markers in soft tissue sarcoma. It is well known that HIF1a
regulates vascular endothelial growth factor (VEGF) expression
under hypoxic conditions (20). The correlation of bFGF and
platelet-derived growth factor (PDGF) to HIF1a and VEGF
was previously described in various cancers but never scruti-
nized in soft tissue sarcoma (21, 22).
In this study, we used a multigene panel to evaluate the
expression of HIF1a,HB-EGF,VEGF-C, and other angiogenic
markers; the prognostic values of these gene expressions; and
their interrelationships in soft tissue sarcoma. We measured
the mRNA expression levels of these genes with quantitative
real-time reverse transcription-PCR (RT-PCR) in formalin-
fixed paraffin-embedded tissue and then further analyzed
the abovementioned genes and their correlation with clinical
and histopathologic variables.
Materials and Methods
Study population, demographic data, and staging procedures. Formalin-
fixed paraffin-embedded samples were gathered from 45 patients with
either malignant fibrous histiocytoma (MFH; 19 of 45, 42.2%) or neu-
rogenic sarcoma (NS; 26 of 45, 57.8%) with a median age of 57 (MFH
70, 39-85; NS 53, 16-72) years at time of operation who were scheduled
for primary surgical resection and were treated with postoperative radio-
therapy (patient characteristics; Table 1). All patients were treated at the
University Hospital of Leipzig, Germany. Tumor-node-metastasis stag-
ing was done according to the criteria of the International Union against
Cancer (23).
Microdissection. After a review of representative H&E-stained slides
of the formalin-fixed paraffin-embedded blocks by a pathologist to es-
timate the tumor load per sample, section slides of 10-μmthickness
were obtained for laser captured microdissection (P.A.L.M. Microlaser
Technologies AG). All tumor slides were prepared as described exten-
sively by Vallbohmer and others in 2005 (24).
Isolation of RNA and cDNA synthesis. The isolation of RNA from
tumor tissue isolated by the microdissection was done in accordance
with a patented procedure at Response Genetics, Inc. (U.S. Patent
6248,535). The cDNA preparation steps were accomplished as de-
scribed previously (25).
Quantitative real-time PCR. To quantify HIF1a,HPSE,bFGF,
PDGFA, PDGFRA, HB-EGF,VEGF,VEGF-C, vascular endothelial growth
factor receptor 1 (VE GFR1), VEGFR2,andVEGFR3 mRNA expression
levels, we used an endogenous reference gene (β-actin) and our gene
set on a method based on real-time fluorescence detection of amplified
cDNA [ABI PRISM 7900 Sequence Detection System (TaqMan) Perkin-
Elmer Applied Biosystem]. The RT-PCR was implemented as previously
described by Kuramochi and others in 2006 (26). All genes were run on
all samples in triplicates. The detection of amplified cDNA results in a
cycle threshold (C
t
) value, which is inversely proportional to the
amount of cDNA. The higher the ensuing cycle threshold (C
t
)value,
the more PCR cycles were necessary to attain detection limit, which
means less cDNA. Colon, liver, and St. Universal Mix RNAs (Stratagene)
were used as control calibrators on each plate. All primers were selected
using the Gene Express software (Applied Biosystems) but were adapted
to the needs of RNA/cDNA as extracted out of paraffin-embedded tis-
sue. All primers were validated before use, analogical to the described
method of Schneider and others in 2005 (27). All results are expressed
as ratios between two absolute measurements (gene of interest/endog-
enous reference gene) to account for loading differences. We used a log
transformation before statistical analysis, including a multiplier which
accounts the average CT values maintained for each gene during the
validation process on the calibrators and therefore allows comparing
samples which were run on different RT-PCR well plates.
The primary sequences of the herein used genes were designed as
previously published by Azuma and colleagues (28).
Statistical analysis. The interrelationship among gene expression
levels were tested with Spearman's test for bivariate correlations. The
Mann-Whitney Utest for not normally distributed samples was used
to test whether the gene expression levels were influenced by the differ-
ent clinicopathologic parameters. Every gene was tested with the
Kaplan-Meier method to estimate the different associations of gene
expression levels with overall survival. Differences in survival between
the high- and low-expression group were analyzed with the log-rank test.
To evaluate independent prognostic factors associated with survival
multivariate cyclooxygenase (Cox) proportional hazards, regression
analysis with stepwise selection was used with the gene set and the tu-
mor stage [International Union against Cancer (UICC)] as covariates
Translational Relevance
This manuscript describes important relationships
between hypoxia-inducible factor 1a and down-
stream angiogenesis genes with potential use for
targeted therapy in soft tissue sarcoma.
Table 1. Patient characteristics
Parameter No. patients (%)
Median age all 57; MFH: median 70, range 39-85; NS: median 53,
range 16-72
Gender
Male 18 (40.0%)
Female 27(60.0%)
Histology
Malignant fibrous histiocytoma 19 (42.2%)
Neurogenic sarcoma 26 (57.8%)
pT category
pT1 10 (22.2%)
pT2 35 (77.8%)
pN category
pN1 38 (84.4%)
pN2 7 (15.6%)
Grading
G
2
20 (44.4%)
G
3
25 (55.6%)
Residual tumor category
R
0
45 (100%)
UICC stage
IIA/B 15 (33.3%)
IIIA/B 20 (44.4%)
IVA/B 10 (22.3%)
Abbreviations: UICC, International Union Against Cancer; UICC
1997; pTNM, TNM pathologic classification; pN, regional lymph
node metastasis; G, grade of differentiation.
5192Clin Cancer Res 2009;15(16) August 15, 2009 www.aacrjournals.org
Imaging, Diagnosis, Prognosis

after adjustment for potential confounders (tumor staging, type of
tumor resection, and age of patients).
A data mining technique provided by the SAS Institute was used to
split gene expression in high- and low-level groups based on a platform
that recursively partitions data according to the relationship between
the Xand Yvalues, creating a tree of partitions (recursive descent par-
tition analysis). By searching all possible cuts, it finds a set of cut points
of Xvalues (gene expression) that best predict the Yvalue (survival
time). These data splits are done recursively forming a tree of decision
rules until the desired fit is reached; the most significant split is deter-
mined by the largest likelihood ratio χ
2
statistic. In either case, the
split is chosen to maximize the difference in the responses between
the two branches of the split. This method was previously used by Lu
and others (29).
We used receiver operating characteristic (ROC) curve analysis to test
the ability of the chosen cut points to discriminate short survivors
(<3 y) from long survivors (3-9 y; refs. 30, 31). The level of significance
was set to P< 0.05. All Pvalues reported were based on two-sided tests.
All statistical tests were done using the Software Packages SPSS for
Windows, Version 16.0 and JMP 7.0 Software (SAS).
Results
Spearman's test for bivariate correlations. Spearman's test on
the log-transformed δCT values showed significant correlations
between some of the gene expressions. HIF1a was correlated to
nearly every gene we tested (Fig. 1).
Comparison of gene expression levels throughout subgroups. To
test whether gene expression levels were significantly different
in-between clinicopathologic subgroups like primary tumor
expansion (pT), Grading, and UICC stage, we used the
Mann-Whitney Utest for not normally distributed samples.
There was no significant difference in gene expression levels
of HIF1a (P= 0.36), HB-EGF (P= 0.8), VEGF-C (P= 0.19),
bFGF (P= 0.89), PDGFA (P=0.68),PDGFRA(P= 0.23),
VEGF (P= 0.58), VEGFR1 (P=0.45),VEGFR2 (P= 0.82),
VEGFR3 (P=0.99),orHPSE (P= 0.31) between pT1 and
pT2 tumors.
The gene expression levels of HIF1a,VEGF-C,HPSE,bFGF,
VEGF,VEGFR1,VEGFR3, and PDGFRA did not differ signifi-
cantly between low- and high-grade tumors (P=0.37;P=
0.79; P= 0.55; P= 0.28; P= 0.45; P= 0.1; P= 0.85); they were,
however, close to significance regarding HB-EGF and PDGFA
(P=0.06;P= 0.06) and even significant concerning VEGFR2
and grading (P= 0.04).
With respect to the distinct UICC stages, HIF1a,HB-EGF,
VEGF-C,bFGF,VEGF,VEGFR2, PDGFRA, and HPSE (P= 0.34;
P= 0.26; P=0.74;P=0.55;P=0.78;P= 0.11; P=0.23;P=
0.32) did not show a significantly different expression; VEGFR1,
VEGFR3, and PDGFA gene expression were however significant
differently (P= 0.04; P= 0.01; P= 0.01) expressed in the discrete
UICC stages.
Partition tree analysis of genes based on survival time. HB-
EGF,HIF1a, and VEGF-C were the most significant divisors in
the recursive partitioning tree for all patients for survival. The
cut point for HB-EGF was the 80th percentile, HIF1a and
VEGF-C expression had their cut-point at the 40th percentile.
Survival analysis using the Kaplan-Meier method. Kaplan-
Meier analysis showed that high expression of HIF1a was corre-
lated to a significantly more favorable prognosis (P= 0.0036;
Fig. 2). The range of expression was 0.408 to 1.529 in the low-
expression group with a median of 1.061 and 1.535 to 8.689 in
the high-expression group with a median of 3.144. The survival
difference between the low-expression group (1 year, 16 of 45
patients, 36.6%) and the high-expression group (3 years, 29 of
45 patients, 64.4%) was 24 months.
In contrast, high expression of HB-EGF was significantly cor-
related to a shorter overall survival and therefore to an unfavor-
able prognosis (P= 0.005, Fig. 3). The cut point was at the 80th
percentile and divided the patients into a group with 9 months
median survival (high expression, 8 of 41 patients, 19.5%,
range 0.396-2.175, median 0.795) and a group of patients
with 27 months of median survival (low expression, 33 of 41
patients, 80.5%, range 0.000434-0.355, median 0.0137).
VEGF-C expression higher than the 40th percentile cut-point
was significantly associated with longer survival (P=0.0023;
Fig. 4). The low-expression group (range 0.000888-0.0985, me-
dian 0.00516) survived a median time of 15 months (17 of 42
patients, 40.5%) and the high-expression group survived 34
months in median (25 of 42, 59.5%), a survival difference of
19 months.
Multivariate Cox proportional hazard regression analysis. Cox
proportional hazard regression analysis with stepwise selection
used to test for the significance of independent association be-
tween the three genes, the UICC stage, and survival revealed
that HB-EGF was the strongest independent factor (P= 0.003)
with a relative risk of 4.4 [exp(b)] of higher HB-EGF for shorter
survival. The relative risk of a higher UICC stage for shorter sur-
vival was 2.5 [exp(b)]. HIF1a and VEGF-C were also significant-
ly included in the model, but their negative coefficient showed
a smaller exponent of 2.6 and 3.9, respectively. The overall
model fit was P< 0.0001.
We designed a combined group with patients showing high-
HIF1a expression (>40th percentile), high–VEGF-C expression
(>40th percentile), and low–HB-EGF expression (<80th per-
centile), assuming that this group should have the best chance
for a long survival according to the Kaplan-Meier and Cox re-
gression analysis; the group consisted of 14 of 45 patients
(31.1%). Stepwise multivariate Cox regression analysis re-
vealed that the combined variable was the strongest indepen-
dent significantly associated factor for survival and superior to
Fig. 1. The used genes and their interrelationship with Spearman's test for
bivariate correlation (P< 0.05). The grayscale represents the number of
connections to other genes. The lighter the grayscale of the gene the fewer
correlations were found on this particular sample set.
5193 Clin Cancer Res 2009;15(16) August 15, 2009www.aacrjournals.org
Three-gene signature in sarcoma

every other factor (gene expression, as well as UICC stage) by
itself [P= 0.0002; exp(b) = 5.48]. Figure 5 illustrates the sur-
vival curves for the combined factor out of the three genes
HB-EGF,HIF1a, and VEGF-C against the combined UICC clas-
sifications (IIA/B, IIIA/B, IVA/B) in the Cox proportional ha-
zards regression model.
All three genes, HIF1a,HB-EGF, and VEGF-C did show signif-
icant gender differences in the association to survival. In the
subgroup of female patients, the association of HB-EGF mRNA
expression with outcome was significantly stronger than in
male patients with a relative risk of 10.19 (P= 0.0007). Using
the combined factor of HIF1a,VEGF-C, and HB-EGF, there were
no significant gender differences.
ROC curve analysis. ROC curve analysis was used to assess
the sensitivity and the specificity of the combined gene expres-
sion of HIF1a,HB-EGF,andVEGF-C to distinguish short (3
years and less) from long survivors (3-9 years). The sensitivity
(true positive rate) was 72.73% and the specificity (true nega-
tive rate) was 82.35% for the diagnosis of short versus long sur-
vival in all patients. The area under the curve was 0.775
(confidence interval 0.626-0.886) with a significance level of
P= 0.0002. The positive likelihood ratio (true positive rate/false
positive rate) was 4.12, and the negative likelihood ratio (false
negative rate/true negative rate) was 0.33 (Fig. 6). Using each of
the single genes at the chosen cut points, HB-EGF,HIF1a, and
VEGF-C failed to properly classify patients according to the sur-
vival criteria used for the combination. The time point of 3
years had the highest sensitivity and specificity, but the combi-
nation of the three genes was significant also at 1, 2, and 5 years
to distinguish short from long survivors.
VEGF-A, VEGF-C, VEGFR1, VEGFR2, and VEGFR3. In-
terestingly, analysis of the used VEGF genes revealed that, espe-
cially VEGF-A, showed significant gender differences in survival
of the examined patients. The 75th percentile cut point of
VEGF-A splits male patients in a high-expression group with a
median survival of 83.5 months and a low-expression group
with a median survival of nearly 5 years less (26 months,
P= 0.03). The female patients showing higher VEGF-A expres-
sion had a significantly worse outcome with a median survival
time of 10 months than the female patients with a low expres-
sion (median survival 27 months, P= 0.02). As mentioned ear-
lier VEGF-C showed significance for survival using Kaplan-
Meier log-rank test at the 40th percentile (P= 0.002). VEGF-
R2 and VEGF-R3 showed significance for survival in Kaplan-
Meier log-rank tests at the 60th and 70th percentile, respectively
(P= 0.03; P= 0.02). Nonetheless, in a multivariate Cox regres-
sion model containing the five VEGF genes used in this study
with the cut points defined by the recursive partitioning tree,
only VEGF-C evinced as independently and significantly associ-
ated with outcome, i.e., low expression as an unfavorable prog-
nostic factor (P= 0.0049).
Discussion
In this study, we determined the gene expressions of HIF1a,
HB-EGF, and other members of the angiogenic pathway in
formalin-fixed paraffin-embedded samples of soft tissue sarco-
ma patients, malignant fibrous histiocytoma, and neurogenic
sarcoma who did not receive any chemotherapy before surgery.
By using laser capture microdissection to isolate tumor tissue
from the clinical specimens along with quantitative RT-PCR,
we hoped to achieve a more precise characterization of the as-
sociations of these gene expressions with each other and with
the outcome of the patients than was previously available.
By multivariate Cox regression analysis, we revealed signifi-
cant independent associations of HB-EGF,HIF1a, and VEGF-C
gene expression to the overall survival. By using the three most
significant genes, we were able to build a combined factor with
a relative risk for shorter survival of 5.5 —that is twice higher
than a one-step increasing UICC stage. With ROC curve analy-
sis, we tested each of the three genes for the ability of the cho-
sen cut points to distinguish short (<3 years) and long survivors
(3-9 years). None of the genes had significant sensitivity or
Fig. 2. Kaplan-Meier plot, estimating overall survival and relapse-free
survival. Differences in survival between the high- and low-HIF1a
expression group were analyzed with the log-rank test. The black line
represents the high-expression group (cut point 40th percentile) with a
median survival of 36 mo (29 of 45 patients, 64.4%), whereas the light gray
curve represents the low-expression group with a median survival of
12 mo (16 of 45 patients, 36.6%).
Fig. 3. Kaplan-Meier plot, estimating overall survival and relapse-free
survival. Differences in survival between the high–and the low–HB-EGF
expression group were analyzed with the log-rank test. The black line
represents the high-expression group (cut point 80th percentile) with a
median survival of 9 mo (8 of 41 patients, 19.5%), whereas the light gray
curve represents the low-expression group with a median survival of 27 mo
(33 of 41 patients, 80.5%).
5194Clin Cancer Res 2009;15(16) August 15, 2009 www.aacrjournals.org
Imaging, Diagnosis, Prognosis

specificity but when using the combined factor the true positive
rate was 73% and the true negative rate was 82% with a signif-
icance of P= 0.0002.
There only have been very few studies thus far to further eval-
uate the associations of the angiogenic genes to outcome of pa-
tients with soft tissue sarcoma, although the association of
HIF1a with outcome has recently been discussed in a variety
of other entities (19, 21, 22, 32–41). All studies showed a sig-
nificant independent association of HIF1a expression to surviv-
al, although the results were different as to the fact whether
high or low levels are associated with an unfavorable prognosis.
Shintani and others reported that using semiquantitative
immunohistochemistry to detect HIF1a in 49 samples of soft
tissue sarcomas they were able to reveal a significantly indepen-
dent association of HIF1a overexpression with an unfavorable
prognosis. As previously published, we found a significantly
independent association between HIF1a overexpression and
decreased survival time in pancreatic ductal adenocarcinoma
(41). Although we were not able to confirm the linkage of
high-HIF1a expression to an aggravated prognosis in soft tissue
sarcoma, we were able to substantiate the finding that HIF1a
seems to be significantly associated with outcome in patients
with soft tissue sarcoma. Furthermore, we were able to show
this correlation with mRNA-based quantitative RT-PCR in soft
tissue sarcoma and therefore hoped to get a more precise and
objective result than is possible with immunohistochemistry.
Volm and Koomagi examined HIF1a expression in samples
of patients with squamous lung cancer for their correlation to
outcome (42). They analyzed HIF1a in formalin-fixed paraffin-
embedded non–small cell lung carcinoma samples (n= 96) by
means of immunohistochemistry to clarify the relationship of
HIF1a overexpression to survival. Despite the findings of Shin-
tani, they found HIF1a overexpression to be significantly asso-
ciated with a favorable prognosis, a result that was confirmed in
oral floor squamous cell and renal cell carcinoma (38, 39). It
has to be mentioned that HIF1a is known to undergo a rapid
posttranscriptional degradation under normoxic conditions by
the ubiquitin-proteasome system, which is restricted by an
oxygen-dependent degradation domain within HIF-1, but un-
der hypoxic conditions, the accumulation of HIF-1a involves
stabilization of the protein (43). For several genes, such as thy-
midylate synthase and dihydropyrimidine dehydrogenase, there
have been studies that provided information on a generally
close linear correlation between the expression of mRNA and
the protein expression (44, 45). Although the regulation of
HIF1a RNA expression seems to respond rapidly to the oxygen
concentrations in the cell (46), meaning that the mechanisms of
HIF1a regulation are transcriptional, as well as a posttranscrip-
tional, it is controversially discussed whether this mechanisms
correlate well with each other (47). Therefore, there might be
quite different findings in correlative studies dependent on
whether they are based on protein or mRNA expression. Further-
more, alternative splicing of HIF1a and upstream genes might
influence the different expression patterns (48–51).
The association of HB-EGF with outcome and tumor aggres-
siveness has been recently discussed in bladder, breast, pancre-
atic, and hepatocellular cancer (52–58) and is even discussed as
a molecular drug target (12, 13). However, its relationship with
the outcome of patients with soft tissue sarcoma has not been
elucidated. In our study group, we found HB-EGF to be the
strongest independent factor significantly associated with sur-
vival. Using stepwise multivariate Cox proportional hazards re-
gression models, we were able to show that HB-EGF mRNA
expression was stronger associated with outcome than the UICC
stage. Moreover, we were able to use the single factors HIF1a,
VEGF-C, and HB-EGF that showed independent association with
survival as a combined factor that had a relative risk for a exac-
erbated prognosis of 5.48 with a significance of P= 0.0002. In-
terestingly only the combined factor was feasible to distinguish
short (<3 years) from long survivors (3-9 years) with a sensitivity
(true positive rate) of 73% and a specificity (true negative rate)
of 82%. The difference in significance cannot so much be
Fig. 4. Kaplan-Meier plot, estimating overall survival and relapse-free
survival. Differences in survival between the high–and low–VEGF-C
expression group were analyzed with the log-rank test. The black line
represents the high-expression group (cut point 40th percentile) with a
median survival of 15 mo (17 of 42 patients, 40.5%), whereas the light gray
curve represents the low-expression group with a median survival of 34 mo
(25 of 42, 59.5%).
Fig. 5. Survival plot from multivariate Cox regression analysis estimating
overall survival and relapse-free survival. The upper light gray solid line
represents the patients with high-HIF1a expression (>40th percentile),
high–VEGF-C expression (>40th percentile), and low–HB-EGF expression
(<80th percentile). The lower black solid line represents patients with
low-HIF1a expression (<40th percentile), low–VEGF-C expression (<40th
percentile), and high–HB-EGF expression (>80th percentile). The dashed
lines represent the different combined UICC stages (from top to bottom)
IIA/B, IIIA/B, and IVA/B.
5195 Clin Cancer Res 2009;15(16) August 15, 2009www.aacrjournals.org
Three-gene signature in sarcoma

explained by differently sized groups (11 of 30 versus 11 of 34)
but rather from the possibly higher association of three genes
out of the angiogenesis pathway to tumor size and vitality of
the tumor cells and therefore survival of the patient. This strong
association of HB-EGF with outcome in soft tissue sarcoma is
especially noteworthy because imatinib mesylate has a strong
physiologic effect on HB-EGF release (10). Imatinib itself is
frequently discussed as a new treatment option for soft tissue
sarcomas (59, 60).
The mRNA expression levels of the genes used in the signa-
ture, HIF1a,HB-EGF, and VEGFC, did not seem to have a sig-
nificant association with tumor size, grading, or the different
UICC stages. These results potentially underline, together with
Cox regression analysis, that gene expression of HIF1a,HB-EGF,
and VEGFC in this patient group was an independent factor. We
did, however, find significant differences for some of the other
used genes like VEGFR1,VEGFR2,VEGFR3,andPDGFA
throughout the clinicopathologic subgroups. It has to be further
examined whether this could lead to markers that help in the
preoperative staging process, meaning that a preoperative biop-
sy and analysis could provide essential information about po-
tential treatment relevant facts earlier than it is currently
available. This could eventually lead to an altered staging and
treatment process with the possibility of neoadjuvant regimens
for certain patient groups.
Due to the emerging discussion about the use of Bevacizumab
in soft tissue sarcoma, we wanted to evaluate the expression pro-
file of VEGF-A, VEGF-C, and their receptors. Interestingly
VEGF-A showed gender-related differences in Kaplan-Meier–
based log-rank tests, a fact especially interesting because VEGF
is known to have strong estrogen dependence (61–63). Using
multivariate analysis, we can not substantiate this finding, only
VEGF-C seemed to be independently associated with survival.
Also HB-EGF seemed to be much stronger associated with sur-
vival in female patients than in male patients. It might however
be interesting to pursue this hypothesis in a larger patient set to
elucidate whether this possible gender differences can be sub-
stantiated and might therefore be a possible explanation for
the previously described difficulties of implementing new anti-
angiogenic drugs in soft tissue sarcoma (6).
Although the genes on themselves showed significantly differ-
ent strong associations with outcome in the gender subgroups of
the combined factor of HIF1a,VEGF-C and HB-EGF was inde-
pendent from gender significantly associated with survival. We
can only speculate that high HIF1a and VEGF-C indicate that the
vascularization of the tumors was increased, a fact that would
surely provide a better response to therapy.
It has to be mentioned that the herein used subentities, ma-
lignant fibrous histiocytoma, and neurogenic sarcoma may con-
sist of different histologic subtypes, like malignant peripheral
nerve sheath tumors, schwannomas, dedifferentiated liposarco-
mas, or other sarcomas. Therefore these results should not be
generalized at the moment. However, also due to our results
from other entities like pancreatic cancer, we think that further
studies examining these angiogenic genes on larger patient col-
lectives seem to be warranted.
Conclusions
The significant independent association of HIF1a,VEGF-C,
and HB-EGF expression with survival probability (P< 0.0001)
and the significantly higher relative risk of patients when using
a combined factor of these three genes [exp(b) = 5.5] compared
with the UICC stages suggest that these three genes may con-
tribute to a better understanding of the prognosis of patients
with soft tissue sarcoma and may even play a crucial role
for the distribution of patients to multimodal therapeutic regi-
mens. The gender differences in drug-related genes like HB-
EGF and VEGF-A lend support to the idea that hormone levels
might influence angiogenesis and therefore potentially also
affect drug efficiency in patients treated with tyrosine kinase
inhibitors or antiangiogenic drugs. Larger studies including
patients treated with actual chemotherapeutics seem to be
warranted.
Disclosure of Potential Conflicts of Interest
A. Hoffman is a Consultant for RGI, Inc.; K. Danenberg is CEO of
Response Genetics, Inc.; P. Danenberg is Scientific Advisor for RGI, Inc.
References
1. Jemal A, Siegel R, Ward E, et al. Cancer
statistics, 2008. CA Cancer J Clin 2008;58:
71–96.
2. Herzog CE. Overview of sarcomas in the adoles-
cent and young adult population. J Pediatr
Hematol Oncol 2005;27:215–8.
3. Seidel C, Bartel F, Rastetter M, et al. Alterations
of cancer-related genes in soft tissue sarcomas:
hypermethylation of RASSF1A is frequently de-
tected in leiomyosarcoma and associated with
poor prognosis in sarcoma. Int J Cancer 2005;
114:442–7.
4. van de Rijn M, Fletcher JA. Genetics of soft
tissue tumors. Annu Rev Pathol 2006;1:435–66.
5. Maki RG. Gemcitabine and docetaxel in meta-
static sarcoma: past, present, and future. Oncol-
ogist 2007;12:999–1006.
6. D'Adamo DR, Anderson SE, Albritton K, et al.
Fig. 6. ROC curve analysis of the combined genes versus short (<3 y) and
long (3-9 y) survival.
5196Clin Cancer Res 2009;15(16) August 15, 2009 www.aacrjournals.org
Imaging, Diagnosis, Prognosis

Phase II study of doxorubicin and bevacizumab
for patients with metastatic soft tissue sarcomas.
J Clin Oncol 2005;23:7135–42.
7. TschoepK, KohlmannA, SchlemmerM,Haferlach
T, Issels RD. Gene expression profiling in sarco-
mas. Crit Rev Oncol Hematol 2007;63:111–24.
8. Blanke CD, Rankin C, Demetri GD, et al. Phase
III randomized, intergroup trial assessing imati-
nib mesylate at two dose levels in patients
with unresectable or metastatic gastrointestinal
stromal tumors expressing the kit receptor ty-
rosine kinase: S0033. J Clin Oncol 2008;26:
626–32.
9. Yang JL, Crowe PJ. Targeted therapies in adult
soft tissue sarcomas. J Surg Oncol 2007;95:
183–4.
10. Johnson FM, Saigal B, Donato NJ. Induction of
heparin-binding EGF-like growth factor and
activation of EGF receptor in imatinib mesylate-
treated squamous carcinoma cells. J Cell Physiol
2005;205:218–27.
11. Higashiyama S, Iwabuki H, Morimoto C, Hieda
M, Inoue H, Matsushita N. Membrane-anchored
growth factors, the epidermal growth factor fam-
ily: beyond receptor ligands. Cancer Sci 2008;99:
214–20.
12. Miyamoto S, Hirata M, Yamazaki A, et al.
Heparin-binding EGF-like growth factor is a
promising target for ovarian cancer therapy.
Cancer Res 2004;64:5720–7.
13. Miyamoto S, Yagi H, Yotsumoto F, et al. New
approach to cancer therapy: heparin binding-
epidermal growth factor-like growth factor as a
novel targeting molecule. Anticancer Res 2007;
27:3713–21.
14. Quiros RM, Rao G, Plate J, et al. Elevated
serum heparanase-1 levels in patients with pan-
creatic carcinoma are associated with poor
survival. Cancer 2006;106:532–40.
15. Kim AW, Xu X, Hollinger EF, Gattuso P,
Godellas CV, Prinz RA. Human heparanase-1
gene expression in pancreatic adenocarcino-
ma. J Gastrointest Surg 2002;6:167–72.
16. Xu X, Rao G, Quiros RM, et al. In vivo and
in vitro degradation of heparan sulfate (HS) pro-
teoglycans by HPR1 in pancreatic adenocarcino-
mas. Loss of cell surface HS suppresses
fibroblast growth fac tor 2-mediated cell signal-
ing and proliferation. J Biol Chem 2007;282:
2363–73.
17. Couvelard A, O'Toole D, Turley H, et al. Micro-
vascular density and hypoxia-inducible factor
pathway in pancreatic endocrine tumours: nega-
tive correlation of microvascular density and
VEGF expression with tumour progression. Br
J Cancer 2005;92:94–101.
18. Banerjee S, Dowsett M, Ashworth A, Martin
LA. Mechanisms of disease: angiogenesis and
the management of breast cancer. Nat Clin Pract
Oncol 2007;4:536–50.
19. Shintani K, Matsumine A, Kusuzaki K, et al. Ex-
pression of hypoxia-inducible factor (HIF)-1αas a
biomarker of outcome in soft tissue sarcomas.
Virchows Arch 2006;449:673–81.
20. Gray MJ, Zhang J, Ellis LM, et al. HIF-1α,
STAT3, CBP/p300 and Ref-1/APE are compo-
nents of a transcriptional complex that regulates
Src-dependent hypoxia-induced expression of
VEGF in pancreatic and prostate carcinomas.
Oncogene 2005;24:3110–20.
21. Bos R, van Diest PJ, de Jong JS, van der Groep
P, van der Valk P, van der Wall E. Hypoxia-
inducible factor-1αis associated with angiogene-
sis, and expression of bFGF, PDGF-BB, and EGFR
in invasive breast cancer. Histopathology 2005;
46:31–6.
22. Giatromanolaki A, Koukourakis MI, Sivridis E,
et al. Relation of hypoxia inducible factor 1 α
and 2 αin operable non-small cell lung cancer
to angiogenic/molecular profile of tumours and
survival. Br J Cancer 2001;85:881–90.
23. Sobin LH, Fleming ID. TNM classification of
malignant tumors, fifth edition (1997). Union In-
ternationale Contre le Cancer and the American
Joint Committee on Cancer. Cancer 1997;80:
1803–4.
24. Vallbohmer D, Zhang W, Gordon M, et al.
Molecular determinants of cetuximab efficacy.
J Clin Oncol 2005;23:3536–44.
25. Lord RV, Salonga D, Danenberg KD, et al.
Telomerase reverse transcriptase exp ression is
increased early in the Barrett's metaplasia, dys-
plasia, adenocarcinoma sequence. J Gastroint-
est Surg 2000;4:135–42.
26. Kuramochi H, Hayashi K, Uchida K, et al. Vas-
cular endothelial growth factor messenger RNA
expression level is preserved in liver metastases
compared with corresponding primary colorec-
tal cancer. Clin Cancer Res 2006;12:29–33.
27. Schneider S, Uchida K, Brabender J, et al.
Downregulation of TS, DPD, ERCC1, GST-Pi,
EGFR, and HER2 gene expression after neoadju-
vant three-modality treatment in patients with
esophageal cancer. J Am Coll Surg 2005;200:
336–44.
28. Azuma M, Shi M, Danenberg KD, et al. Serum
lactate dehydrogenase levels and glycolysis sig-
nificantly correlate with tumor VEGFA and
VEGFR expression in metastatic CRC patients.
Pharmacogenomics 2007;8:1705–13.
29. Lu KH, Patterson AP, Wang L, et al. Selection
of potential markers for epithelial ovarian
cancer with gene ex pression arrays and recur-
sive descent partition analysis. Clin Cancer Res
2004;10:3291–300.
30. Metz CE. Basic principles of ROC analysis.
Semin Nucl Med 1978;8:283–98.
31. Zweig MH, Campbell G. Receiver-operating
characteristic (ROC) plots: a fundamental evalu-
ation tool in clinical medicine. Clin Chem 1993;
39:561–77.
32. Aebersold DM, Burri P, Beer KT, et al. Expres-
sion of hypoxia-inducible factor-1α: a novel pre-
dictive and prognostic parameter in the
radiotherapy of oropharyngeal cancer. Cancer
Res 2001;61:2911–6.
33. Bachtiary B, Schindl M, Potter R, et al. Overex-
pression of hypoxia-inducible factor 1αindicates
diminished response to radiotherapy and unfa-
vorable prognosis in patients receiving radical
radiotherapy for cervical cancer. Clin Cancer
Res 2003;9:2234–40.
34. Birner P, Gatterbauer B, Oberhuber G, et al. Ex-
pression of hypoxia-inducible factor-1 αin oligo-
dendrogliomas: its impact on prognosis and on
neoangiogenesis. Cancer 2001;92:165–71.
35. Birner P, Schindl M, Obermair A, Breitenecker
G, Oberhuber G. Expression of hypoxia-inducible
factor 1αin epithelial ovarian tumors: its impact
on prognosis and on response to chemotherapy.
Clin Cancer Res 2001;7:1661–8.
36. Burri P, Djonov V, Aebersold DM, et al. Signif-
icant correlation of hypoxia-inducible factor-1α
with treatment outcome in cervical cancer trea-
ted with radical radiotherapy. Int J Radiat Oncol
Biol Phys 2003;56:494–501.
37. Enatsu S, Iwasaki A, Shirakusa T, et al. Ex-
pression of hypoxia-inducible factor-1αand its
prognostic significance in small-sized adenocar-
cinomas of the lung. Eur J Cardiothorac Surg
2006;29:891–5.
38. Fillies T, Werkmeister R, van Diest PJ, Brandt
B, Joos U, Buerger H. HIF1-αoverexpression in-
dicates a good prognosis in early stage squa-
mous cell carcinomas of the oral floor. BMC
Cancer 2005;5:84.
39. Lidgren A, Hedberg Y, Grankvist K, Rasmuson
T, Vasko J, Ljungberg B. The expression of hyp-
oxia-inducible factor 1αis a favorable indepen-
dent prognostic factor in renal cell carcinoma.
Clin Cancer Res 2005;11:1129–35.
40. Schindl M, Schoppmann SF, Samonigg H,
et al. Overexpression of hypoxia-inducible factor
1αis associated with an unfavorable prognosis
in lymph node-positive breast cancer. Clin Can-
cer Res 2002;8:1831–7.
41. Hoffmann AC, Mori R, Vallbohmer D, et al.
High expression of HIF1a is a predictor
of clinical outcome in patients with pancreatic
ductal adenocarcinomas and correlated to
PDGFA, VEGF, and bFGF. Neoplasia 2008;10:
674–9.
42. Volm M, Koomagi R. Hypoxia-inducible factor
(HIF-1) and its relationship to apoptosis and pro-
liferation in lung cancer. Anticancer Res 2000;20:
1527–33.
43. Huang LE, Arany Z, Livingston DM, Bunn HF.
Activation of hypoxia-inducible transcription fac-
tor depends primarily upon redox-sensitive sta-
bilization of its αsubunit. J Biol Chem 1996;271:
32253–9.
44. Johnson MR, Wang K, Smith JB, Heslin MJ,
Diasio RB. Quantitation of dihydropyrimidine
dehydrogenase expression by real-time
reverse transcription polymerase chain reaction.
Anal Biochem 2000;278:175–84.
45. Johnston PG, Lenz HJ, Leichman CG, et al.
Thymidylate synthase gene and protein expres-
sion correlate and are associated with response
to 5-fluorouracil in human colorectal and gastric
tumors. Cancer Res 1995;55:1407–12.
46. Wang GL, Jiang BH, Rue EA, Semenza GL.
Hypoxia-inducible factor 1 is a basic-helix-loop-
helix-PAS heterodimer regulated by cellular O2
tension. Proc Natl Acad Sci U S A 1995;92:
5510–4.
47. Krieg M, Haas R, Brauch H, Acker T, Flamme
I, Plate KH. Up-regulation of hypoxia-inducible
factors HIF-1αand HIF-2αunder normoxic con-
ditions in renal carcinoma cells by von Hippel-
Lindau tumor suppressor gene loss of function.
Oncogene 2000;19:5435–43.
48. Bartel F, Meye A, Wurl P, et al. Amplification of
the MDM2 gene, but not expression of splice
variants of MDM2 MRNA, is associated with
prognosis in soft tissue sarcoma. Int J Cancer
2001;95:168–75.
49. Bartel F, Schulz J, Bohnke A, et al. Signifi-
cance of HDMX-S (or MDM4) mRNA splice
variant overexpression and HDMX gene ampli-
fication on primary soft tissue sarcoma prog-
nosis. Int J Cancer 2005;117:469–75.
50. Drutel G, Kathmann M, Heron A, et al. Two
splice variants of the hypoxia-inducible factor
HIF-1αas potential dimerization partners of
ARNT2 in neurons. Eur J Neurosci 2000;12:
3701–8.
51. Maynard MA, Evans AJ, Hosomi T, Hara S,
Jewett MA, Ohh M. Human HIF-3α4isadomi-
nant-negative regulator of HIF-1 and is down-
regulated in renal cell carcinoma. FASEB J
2005;19:1396–406.
52. Kramer C, Klasmeyer K, Bojar H, Schulz WA,
Ackermann R, Grimm MO. Heparin-binding
epidermal growth factor-like growth factor iso-
forms and epidermal growth factor receptor/
ErbB1 expression in bladder cancer and their
relation to clini cal outcome. Cancer 2007;109:
2016–24.
53. Adam RM, Danciu T, McLellan DL, et al. A nu-
clear form of the heparin-binding epidermal
growth factor-like growth factor precursor is a
feature of aggressive transitional cell carcinoma.
Cancer Res 2003;63:484–90.
54. Thogersen VB, Sorensen BS, Poulsen SS,
Orntoft TF, Wolf H, Nexo E. A subclass of
HER1 ligands are prognostic markers for surviv-
al in bladder cancer patients. Cancer Res 2001;
61:6227–33.
55. Ito Y, Higashiyama S, Takeda T, Yamamoto Y,
Wakasa KI, Matsuura N. Expression of heparin-
binding epidermal growth factor-like growth
factor in pancreatic adenocarcinoma. Int J Pan-
creatol 2001;29:47–52.
56. Ito Y, Takeda T, Higashiyama S, Noguchi S,
Matsuura N. Expression of heparin-binding
5197 Clin Cancer Res 2009;15(16) August 15, 2009www.aacrjournals.org
Three-gene signature in sarcoma

epidermal growth factor-like growth factor in
breast carcinoma. Breast Cancer Res Treat 2001;
67:81–5.
57. Ito Y, Takeda T, Higashiyama S, et al. Expression
of heparin binding epidermal growth factor-like
growth factor in hepatocellular carcinoma: an im-
munohistochemical study. Oncol Rep 2001;8:903–7.
58. Hoffmann AC, Mori R, Vallbohmer D, et al.
High expression of heparanase is significantly
associated with dedifferentiation and lymph
node metastasis in patients with pancreatic duc-
tal adenocarcinomas and correlated to PDGFA
and via HIF1a to HB-EGF and bFGF. J Gastroint-
est Surg 2008;12:1674–81, discussion 81-2.
59. Shor AC, Agresta SV, D'Amato GZ, Sondak VK.
Therapeutic potential of directed tyrosine kinase
inhibitor therapy in sarcomas. Cancer Control
2008;15:47–54.
60. Wagner A. Treatment of advanced soft tissue
sarcoma: conventional agents and promising
new drugs. J Natl Compr Canc Netw 2007;5:
401–10.
61. Herve MA, Meduri G, Petit FG, et al. Regulation
of the vascular endothelial growth factor
(VEGF) receptor Flk-1/KDR by estradiol through
VEGF in uterus. J Endocrinol 2006;188:91–9.
62. Meduri G, Bausero P, Perrot-Applanat M. Ex-
pression of vascular endothelial growth factor
receptors in the human endometrium: modula-
tion during the menstrual cycle. Biol Reprod
2000;62:439–47.
63. Ochoa AL, Mitchner NA, Paynter CD, Morris
RE, Ben-Jonathan N. Vascular endothelial
growth factor in the rat pituitary: differential dis-
tribution and regulation by estrogen. J Endocri-
nol 2000;165:483–92.
5198Clin Cancer Res 2009;15(16) August 15, 2009 www.aacrjournals.org
Imaging, Diagnosis, Prognosis