Comparison of microarray-based RNA expression with ELISA-based protein determination of HER2, uPA and PAI-1 in tumour tissue of patients with breast cancer and relation to outcome

Abstract

Purpose

Prognostic and predictive markers in breast cancer are currently determined by single analysis of protein amounts. If RNA-based multi-gene analyses enter clinical practice, simultaneous determination of currently established markers like human epidermal growth factor receptor 2 (HER2), urokinase plasminogen activator (uPA) and its inhibitor (PAI-1) would represent an elegant simplification. To investigate the correlation between RNA and protein levels, we assessed HER2, uPA and PAI-1 in patients with breast cancer. In addition, we evaluated the influence of these factors on patient outcome.

Methods

We collected tumour samples from 133 patients with primary breast cancer. Protein and mRNA levels were measured for HER2, uPA and PAI-1. Protein concentration was measured by ELISA, mRNA expression was analysed by Affymetrix A133U Gene Chip and validated by quantitative PCR.

Results

We were able to demonstrate a statistically significant correlation between mRNA and protein expression for HER2 (r = 0.67, P < 0.001) and uPA (r = 0.7, P < 0.001) but not for PAI-1 (r = 0.27). We observed a prognostic information for PAI-1 mRNA and protein values. Patients with high PAI-1 mRNA expression had a reduced 10-year disease-free survival (DFS) rate (60 vs. 70%, P = 0.071) and 10-year overall survival (OS) rate (68 vs. 79%, P = 0.034). Patients with PAI-1 protein levels above 14 ng/mg protein had a reduced disease-free (10-year DFS rate 54 vs. 71%, P = 0.006) and overall survival rate (10-year OS-rate 63 vs. 83%, P = 0.018). In the patient cohort with no chemotherapy, PAI-1 mRNA levels were the strongest prognostic factor for OS in univariate and multivariate analysis.

Conclusions

Results of RNA-based multi-gene analyses of the prognostic and predictive markers HER2 and uPA correlate with the corresponding protein levels. This is not the case for PAI-1. However, PAI-1 mRNA expression might reveal new clinically relevant information in addition to PAI protein levels.

Introduction

Important factors of breast cancer classification include prognostic and predictive biomarkers like the human epidermal growth factor receptor 2 (HER2) and the urokinase plasminogen-activator (uPA) with its inhibitor (PAI-1). These factors are currently determined by individual protein assays measuring single biomarkers with the need to perform multiple tests on the same tumour specimen.

RNA expression arrays offer the opportunity to simultaneously assess multiple biological markers or drug target genes (“Expression Profiling”). Gene chip assays for prognostic use in patients with breast cancer are prospectively evaluated in large cohorts (Bogaerts et al. 2006; van’t Veer et al. 2005), but results will not be available in the near future. If multi-gene analyses find their way into clinical practice, the determination of currently established markers like HER2, uPA and PAI-1 by the same method would represent an elegant simplification. Preliminary reports indicate that the evaluation of prognostic markers can be done as part of a clinical breast cancer specific multi-gene assay and may prove to be useful (Esteva et al. 2005; Paik et al. 2004). However, it is unclear for most of the markers whether the measured gene expression correlates with protein expression. Transcriptional cell activity does not necessarily translate into an active protein as posttranscriptional regulations and modifications take place with high complexity and dynamics. Therefore, several studies investigated the correlation between gene and protein expression.

HER2 represents the currently most promising target for therapy and is also discussed as predictive marker (Dowsett and Dunbier 2008). A correlation between gene expression by PCR and protein expression by immunohistochemistry (IHC) and gene copy number by fluorescence in situ hybridization (FISH) was described (Gjerdrum et al. 2004; Vinatzer et al. 2005).

The proteolytic factors uPA and PAI-1 are prospectively evaluated as prognostic markers in patients with axillary node-negative breast cancer (Annecke et al. 2008). The determination of these factors is performed from fresh frozen tissue by ELISA. Conflicting results for the correlation between gene and protein expression show either a good or a weak or no existing correlation (Biermann et al. 2008; Castello et al. 2007; Spyratos et al. 2002).

Taken together, the potential of RNA-based multi-gene analysis is not completely clear. Therefore, we examined the correlation between mRNA values of HER2, uPA and PAI-1 and the corresponding protein expression in fresh frozen tumour tissue of 133 patients with primary breast cancer. In addition, we investigated the prognostic impact of these factors on mRNA and protein level.

Materials and methods

Patients

Tissue samples of 133 patients with primary breast cancer were collected during surgery, snap-frozen and stored in liquid nitrogen. All patients were treated for breast cancer at the University Medical Center Hamburg Eppendorf between 1992 and 2002. Patient selection was based upon availability of tumour tissue.

Patients gave written informed consent to access their tissue and review their medical records according to our investigational review board and ethics committee guidelines.

No radiotherapy or neoadjuvant chemotherapy had been performed prior to surgery. Breast conserving surgery was performed in 55% of the patients, 45% were treated by mastectomy. Seventy-three patients (60%) received adjuvant chemotherapy, 79 patients (66%) radiotherapy and 72 (59%) adjuvant endocrine treatment. Detailed patient characteristics at the time of primary diagnosis are listed in Table 1. At the time of recurrence, 25 patients (24%) had pulmonary metastases as first site of metastasis, 22 patients (21%) had liver metastases, 43 patients (41%) had bone metastases and 10 patients (10%) had brain metastases as first site of metastasis (n = 29 were unknown).

Table 1.

Tumour and treatment characteristics of the patients with primary breast cancer

	No. of patients	Percent
Tumour size
T1	30	24.4
T2	86	69.9
T3	2	1.6
T4	5	4.1
Unknown	10	7.5
Grading
G1	12	9
G2	50	37.6
G3	60	45.1
Grading unknown	11	8.3
Oestrogen receptor status
Positive	94	75.2
Negative	26	20.8
Unknown	6	4.8
Progesterone receptor status
Positive	90	67.7
Negative	38	28.6
Unknown	5	3.8
Axillary lymph node status
Positive	93	69.9
Negative	35	26.3
Unknown	5	3.8
Adjuvant treatment
Chemotherapy (CT) alone	37	31
Endocrine therapy (ET) alone	36	30
CT plus ET	30	29
No therapy	12	10

Median age at the time of surgery: 56.3 years

Protein extraction/tissue preparation for ELISA analyses

For all tumours, samples were snap-frozen and stored in liquid nitrogen. Tumour cell content exceeded 70% in all samples, verified by H&E staining of cryo-cut sections. Pieces of approximately 100 mg were cut from the samples and pulverized using a microdismembrator (Braun-Melsungen, Melsungen, Germany) for 2 × 45 s at 200 rpm. The resulting powder was immediately suspended in 670 μl TBS with 1% Triton X-100. The suspension was incubated overnight at 4°C with gentle shaking and subjected to ultracentrifugation for 1 h at 25,000 rpm at 4°C in order to remove insoluble cell material. The supernatants were stored in liquid nitrogen until use.

HER2-determination by ELISA

Tissue HER2 was quantified by a commercially available enzyme-linked immunosorbent assay (Siemens Healthcare Diagnostics Inc., Tarrytown, NY, USA) as previously described (Mueller et al. 2004). The results were expressed in ng per mg of total protein.

Quantitative analysis of uPA and PAI-1 by ELISA

uPA and PAI-1 antigen concentrations were determined applying commercially available ELISA kits (Femtelle™ uPA/PAI-1 ELISA kit, American Diagnostica Inc., Stamford, CT, USA). Antigen concentrations in the detergent extracts were expressed in ng per mg of total protein.

RNA isolation

Approximately 50 mg of frozen breast tumour tissue was pulverized in liquid nitrogen. RLT-Buffer (QIAGEN, Hilden, Germany) was added and the homogenate was centrifuged through a QIAshredder column (QIAGEN). From the eluate, total RNA was isolated by the RNeasy Kit (QIAGEN) according to the manufacturer’s instructions. RNA yield was determined by UV absorbance and RNA quality was assessed by analysis of ribosomal RNA band integrity on an Agilent 2100 Bioanalyzer RNA 6000 LabChip kit (Agilent Technologies, Palo Alto, CA, USA).

Microarray analysis

The Affymetrix (Santa Clara, CA, USA) HG-U133A array and GeneChip System™ was used to quantify the relative transcript abundance in breast cancer tissues. Starting from 5 μg total RNA, labelled cRNA was prepared using the Roche Microarray cDNA Synthesis, Microarray RNA Target Synthesis (T7) and Microarray Target Purification Kit, according to the manufacturer’s instructions. In the analysis settings, the global scaling procedure was chosen which multiplied the output signal intensities of each array to a mean target intensity of 500. Samples with suboptimal average signal intensities (i.e., scaling factors >25) or GAPDH 3′/5′ ratios >5 were relabelled and rehybridized on new arrays.

PCR

Aliquots of total RNA used for GeneChip expression analysis were used for quantitative RT-PCR for uPA and PAI-1 with an ABI PRISM 7900 Sequence Detection System (Applied Biosystems, Foster City, CA, USA). cDNA for PCR amplification was generated by oligo dT-primed reverse transcription (Superscript First Strand System, Invitrogen Corporation, Carlsbad, CA, USA) including DNAse I treatment. Primers and probes were designed with the Primer Express software (Applied Biosystems, Foster City, CA, USA) and spanned the same gene region of the respective Affymetrix probe set. Labelled oligonucleotides were obtained from Eurogentec s.a. (Liege, Belgium). Absolute copy numbers were normalized according to GAPDH as a reference gene. The primer/probes were prepared by mixing 25 μl of 100 μM stock solution “upper primer”, 25 μl of 100 μM stock solution “lower primer” plus 12.5 μl of the 100 μM stock solution TaqMan-probe (FAM/TAMRA) and adjusted to 500 μl with H₂O (Primer/probe-mix). PCR reactions using cDNA generated from 1.5 ng total RNA were performed in duplicates in a volume of 10 μl. This included TaqMan universal Mix (Eurogentec s.a.) according to manufacturer’s protocol in a 384-well format and 1 μl of the P&P mix. Thermal cycler parameters were 2 min at 50°C, 10 min at 95°C and 40 cycles, each consisting of a 15 s denaturation step at 95°C and a 1-min annealing/extension step at 60°C. Relative abundance of a gene transcript was calculated either by the ΔΔCt method or by arbitrarily defined RNA copy number estimates at a Ct = 24 as 106 copies. Subsequent analysis included normalization steps such as median centring and per gene median division.

The primer/probe sets used for amplification of the target genes were the following:

• uPA Probe ACTACATCGTCTACCTGGGTCGCTCAAGG

• Forward Primer CGCCACACACTGCTTCATTG

• Reverse Primer CCCCTTGCGTGTTGGAGTT

• PAI-1 Probe TGGGCCAAGTGATGGAACCCTGAC

• Forward Primer GCACAACCCCACAGGAACA

• Reverse Primer GTCCCAGATGAAGGCGTCTTT

Statistical analysis

Correlations between protein or mRNA expression and clinical or histological tumour parameters were calculated by Spearman analysis using the SPSS 15.0 software (SPSS Inc, Chicago, IL, USA). A correlation coefficient below 0.3 (r < 0.3) was interpreted as weak correlation. For prognostic parameters, the following groups were compared: Tumour size <5 cm (pT1 + 2) versus more than 5 cm (pT3 + 4), G1/G2 vs. G3; node-positive versus node-negative tumours; ER/PR positive versus ER/PR negative tumours; age <56 vs. 56 years and older. For survival analyses, the mRNA values were first divided into quartiles. Survival analyses were then performed for all quartiles. The cut-offs that resulted in the most significant difference in outcome were used. For HER2, the upper 25% of values were compared with the lower 75%, for uPA and PAI-1, values below and above the median were compared. Overall survival was computed from the date of surgery to the date of death due to breast cancer. Survival curves were compared with the logrank test. Univariate as well as multivariate P values for the respective risk factors in the survival model were obtained by a Cox proportional hazards model. All tests were performed at a significance level of P = 0.05 (two-sided).

Results

Correlation of biological factors determined by ELISA and mRNA expression determined by microarray analysis

Growth factor receptor HER2

A high correlation between HER2 levels measured on protein and mRNA level was observed (r = 0.67, P < 0.001, Fig. 1). The median HER2 protein level in tissue extracts was 18 ng/mg protein (2.5–750 ng/mg protein), the median HER2 mRNA expression value on the Affymetrix chip was 286 (54–4300).

Fig. 1.

Correlations between HER2 protein expression and HER2 mRNA expression (r = 0.67)

Invasion factors uPA and PAI-1

We found highly significant correlations between uPA protein and mRNA levels (r = 0.7, P < 0.001, Fig. 2a) but a weak association between PAI-1 protein and mRNA levels (r = 0.27, P = 0.001, Fig. 2b). The median uPA protein level was 2.55 ng/mg protein (0.17–21.36 ng/mg protein), the median uPA mRNA level was 135 (40–800). The median PAI-1 protein level was 12.2 ng/mg protein (0.21–111.33 ng/mg protein), the median PAI-1 mRNA level was 77 (4.8–448). Taking previously published cut-off levels (Harbeck et al. 1999) into account, 43% of patients had elevated uPA protein levels (>3 ng/mg protein) and 41% of patients had elevated PAI-1 protein levels (>14 ng/mg protein).

Fig. 2.

Correlations between uPA protein expression und uPA mRNA expression (r = 0.70) (a) and between PAI-1 protein expression und PAI-1 mRNA expression (r = 0.27) (b)

Taken together, we observed a highly significant concordance between protein expression determined by ELISA and mRNA expression measured by microarray-based analyses for HER2 and uPA (P < 0.001), but not for PAI-1 (Table 2). The same results could be obtained in different subsets of patients according to their hormone receptor status (ER and PR positive vs. ER and PR negative).

Table 2.

Correlation between protein and mRNA expression data (ELISA vs. RNA-based analysis)

	Number (n)	Correlation (r)	95% Confidence-interval	P value
HER2	133	0.67	0.56–0.76	<0.001
uPA	133	0.70	0.53–0.81	<0.001
PAI-1	133	0.27	0.18–0.38	0.001

Validation of microarray-based RNA measurements by quantitative PCR

In order to validate results obtained from microarray analysis, we performed a quantitative PCR for uPA and PAI-1 in a subset of patients (n = 46). Strong correlations between RNA levels measured by both methods were found (r = 0.72 for uPA, P < 0.001 and r = 0.78 for PAI-1, P < 0.001).

mRNA expression and protein levels and their association with clinical and histomorphological parameters

The relation between mRNA expression and protein levels of the three markers and clinicopathological parameters including tumour size, nodal status, oestrogen and progesterone receptor status, age and grading were analysed in the cohort of 133 patients.

Regarding the RNA data, we found significant positive correlations between uPA and PAI-1 (r = 0.45, P < 0.001). In addition, we observed a negative correlation between oestrogen receptor and uPA (r = −0.37, P < 0.001) and PAI-1 (r = −0.38, P < 0.001), respectively. We also found a positive correlation between uPA mRNA and PAI protein levels (r = 0.33, P < 0.001) as well as a weak correlation between PAI-1 mRNA and uPA protein levels (r = 0.22, P = 0.01). A positive PR status (IHC) correlated inversely with PAI-1 mRNA levels (r = −0.3, P = 0.01). No additional correlations were found.

Disease free and overall survival analysis

Analyses were performed separately for mRNA and protein expression and for established clinicopathological parameters. mRNA values were divided into two groups as described earlier (see statistics).

After a median follow-up time of 88 months (range 8–169 months), 30% of the patients (34/117) had recurrent disease and 24% of the patients (30/126) had died.

Univariate analysis

A non-significant trend for shorter disease-free survival was observed for high PAI-1 mRNA expression (10-year disease-free survival (DFS) rates 60 vs. 70%, P = 0.071, Fig. 3a; Table 3).

Fig. 3.

Disease-free survival in patients with high and low PAI-1 mRNA levels (P = 0.071) (a). Overall survival in patients with high and low PAI-1 mRNA levels (P = 0.034) (b). q1–2: quartiles 1 to 2; q3–4: quartile 3 to 4

Table 3.

mRNA expression and protein levels in relation to disease-free survival in 133 patients (univariate analysis)

	Cut-off	Number of patients in each group above and below the cut-off	Number of events	10-year-DFS-survival rate (percent)	P value
Gene analysis	mRNA values
uPA	134	62	23	62	0.97
		61	16	66
PAI-1	76	64	21	60	0.071
		59	17	70
HER2	425	31	12	59	0.54
		92	25	66
Protein analysis	Protein values
uPAa	3	72	20	56	0.062
		51	19	70
PAI-1a	14	72	19	54	0.006
		51	19	71
HER2a	31	31	12	60	0.638
		92	23	65

^ang/mg protein

Bold value indicates P < 0.05

In addition, a significantly shorter overall survival could be demonstrated by univariate analysis for high PAI-1 mRNA expression (10-year overall survival (OS) rates 68 vs. 79%, P = 0.034, Fig. 3b; Table 4).

Table 4.

mRNA expression and protein levels in relation to overall survival in 133 patients (univariate analysis)

mRNA	Cut-off	Number of patients in each group above and below the cut-off	Number of events	10-year-OS-survival rate (percent)	P value
Gene analysis	mRNA values
uPA	134	64	15	73	0.71
		62	13	74
PAI-1	76	65	18	68	0.034
		61	10	79
HER2	425	32	5	84	0.59
		94	23	70
Protein analysis	Protein values
uPAa	3	54	11	65	0.177
		72	9	82
PAI-1a	14	52	12	63	0.018
		74	8	83
HER2a	31	31	4	83	0.428
		95	23	70

^ang/mg protein

Bold values indicate P < 0.05

For protein determinations, a shorter disease-free survival was observed in the univariate analysis for high PAI-1 levels (>14 ng/mg protein, 10-year DFS rates 54 vs. 71%, P = 0.006, Table 3). In addition, a shorter overall survival was observed for high PAI-1 levels (10-year OS rates 83 vs. 63%, P = 0.018, Table 4).

For uPA as well as HER2 mRNA and protein levels, we observed no prognostic impact on DFS and OS except a trend pointing to a shorter DFS in cases with high uPA protein levels (P = 0.062, Tables 3, 4).

Regarding clinicopathological parameters, tumour size (T1 + 2 vs. T3 + 4) correlated with shorter disease-free (10-year DFS rates 64 vs. 43%, P = 0.04) and overall survival (10-year OS rates 74 vs. 53%, P = 0.04, data not shown). Positive nodal status and high grading (G3) were of borderline significance for DFS (P = 0.08 for nodal status and P = 0.075 for grading) but not for OS (P > 0.10). The use of adjuvant hormone therapy correlated with longer overall survival (10-year OS-rate 78 vs. 60%, P = 0.045) but not with DFS (P > 0.05) and the use of adjuvant chemotherapy had no impact on DFS and OS (P > 0.05) in univariate analysis.

In the subgroup of patients who received no chemotherapy, the prognostic impact of PAI-1 was even stronger. Patients with high PAI-1 protein levels or high PAI-1 mRNA levels had a shorter disease-free survival (10-year DFS rate 41 vs. 79%, P = 0.01 for PAI-1 protein levels, 10-year DFS rate 44 vs. 74%, P = 0.03 for PAI-1 mRNA levels, Fig. 4a). A shorter overall survival in the group of patients with no adjuvant chemotherapy was only observed for high PAI-1 mRNA levels (10-year OS rate 41 vs. 79%, P = 0.004, Fig. 4b).

Fig. 4.

Disease-free (a) and overall survival (b) in the subgroup of patients with no chemotherapy with high and low PAI-1 mRNA levels. q1–2: quartiles 1 to 2; q3–4: quartile 3 to 4

Multivariate analysis

Only PAI-1 protein levels correlated with shorter disease-free survival (HR = 1.02, 95% CI 1.004–1.034, P = 0.014) and overall survival (HR = 1.02, 95% CI 1.0–1.04, P = 0.06) by multivariate analysis adjusted for age, nodal status, grading, ER status and PAI-1 mRNA-expression (not shown).

However, in the subgroup of patients with no adjuvant systemic treatment, only PAI-1 mRNA levels correlated with overall survival (HR 4.38, 95% CI 1.45–13.02, P = 0.009) by multivariate analysis adjusted for the same parameters (not shown).

Discussion

Our aim was to evaluate whether RNA-based analysis of established prognostic markers could provide clinically useful results comparable to protein determination. In the present study, we observed a significant correlation between protein and mRNA levels for HER2 and uPA but not for PAI-1.

Stratification of patients with breast cancer regarding adjuvant therapy is currently based on clinicopathological parameters. In the near future, multi-gene analyses may be added to the clinical routine once the results from large prospectively designed, randomized studies will be available. As soon as these new techniques enter clinical practice, it will be essential if relevant biomarkers like HER2, uPA and PAI-1 can be measured reliably in the same specimen. RNA analysis by microarray or multiplex PCR represents a promising approach to assess biological factors requiring very small amounts of tumour tissue and delivering quantitative estimates of the mRNA expression. In addition, establishment of RNA-based analysis also seems feasible from paraffin-embedded tissue sections (Ross et al. 2008).

Patients with tumours overexpressing HER2 are candidates for the treatment with anti-HER2 agents like trastuzumab and lapatinib (Di Cosimo and Baselga 2008). The selection of patients with HER2-overexpressing tumours for treatment is usually based on IHC combined with FISH for validation in cases with moderate expression (2+) by IHC (Birner et al. 2001; Wolff et al. 2007). Despite published standards, a relevant rate of discordance between IHC and FISH was found also in recent clinical trials (Perez et al. 2006). Thus, new HER2 tests like gene expression analyses might be useful. Several studies showed a good correlation between HER2 status measured by gene expression analysis and IHC or FISH. Schlemmer et al. (2004) found a strong correlation between RT-PCR, IHC and FISH for HER2 results in 112 patients without clinical follow-up. Berqvist et al. (2007) showed a good concordance between microarray-based RNA expression analysis and IHC for HER2 in 306 primary breast cancer tumours with a significant impact of HER2 levels measured by RNA on survival. Vinatzer et al. (2005) also observed a positive correlation between RT-PCR and IHC for HER2 in 136 tumours of 85 patients with breast cancer. HER2 overexpression determined by RT-PCR correlated with reduced disease-free survival (P < 0.003).

In our study, we also demonstrated a strong correlation between mRNA and protein levels for HER2, but we were not able to show a correlation of HER2 expression with survival which is in line with the current opinion that HER2 is not a strong prognostic factor. HER2 protein levels were determined by ELISA, currently not a standard diagnostic procedure in primary breast cancer tissue. The ELISA test used in our study is an FDA-approved method to monitor HER2 levels in the serum of women with metastatic breast cancer (Carney et al. 2003). Our group as well as others were able to demonstrate that this approach represents a useful method to determine quantitative HER2 levels in tumour tissue (Konecny et al. 2001; Mueller et al. 2003).

For uPA and PAI-1 protein determination, the measurement by ELISA is the standard method and has been validated in several studies demonstrating the clinical relevance of these proteins as prognostic factors (Harbeck et al. 1999, 2002; Jaenicke et al. 1993). These factors are already part of clinical practice recommendations and are further evaluated in the context of large studies particularly in patients with node-negative breast cancer (Annecke et al. 2008). The currently established uPA and PAI-1 measurement by ELISA requires a relatively large amount of fresh frozen tumour tissue. Measurement of mRNA expression offers the opportunity to determine multiple markers from limited tumour tissue specimens and should help to reduce the amount of tissue required (Castello et al. 2002, 2006). This is of clinical relevance since the average size of newly diagnosed tumours is decreasing with the improvements in early breast cancer detection.

We evaluated the protein and gene expression of uPA and PAI-1 by array analysis and ELISA and found a high concordance between mRNA levels and protein values for uPA, but not for PAI-1. Thus, our results show that PAI-1 antigen concentrations do not reflect the mRNA expression levels. This might be due to posttranscriptional regulation processes which were described for components of the plasminogen activation system (Nagamine et al. 2005). As we could show the same correlation coefficients for uPA, PAI-1 and HER2 consistently in different subsets of patients, we exclude that the weak correlation for PAI-1 is due to analytical problems directly related to the assay utilized. The lack of correlation between PAI-1 measured on mRNA and protein level was previously observed by Biermann et al. (2008) and Castelló et al. (2007), who similarly found no correlation for PAI-1 (r < 0.3) in a smaller group of 74 and 70 patients, respectively, with the same method of protein extraction for ELISA and RT-PCR for determination of mRNA values. In addition, Castelló et al. (2007) found no prognostic value of PAI-1 mRNA values in their patient cohort (50% node-positive) with a median follow-up time of 30 months.

Interestingly, in our patient cohort with a median follow-up time of 88 months PAI-1 mRNA levels as determined by microarray analysis were prognostic indicators by univariate analysis, similar to PAI-1 protein levels despite the weak correlation with PAI-1 protein levels. In the subgroup of patients with no chemotherapy, PAI-1 mRNA levels were even a better prognostic factor. In multivariate analysis, only high PAI-1 mRNA remained a significant indicator for shorter overall survival. This indicates that PAI-1 mRNA expression may give additional prognostic information and that the worse prognosis of patients with high PAI-1 mRNA levels might be overcome by chemotherapy reflecting also a predictive effect of PAI-1. Our results are in accordance with previous studies showing that PAI-1 mRNA had a stronger prognostic impact for metastasis-free-survival and breast cancer–specific survival than uPA in lymph node-positive patients receiving adjuvant treatment (n = 87) (Leissner et al. 2006). Sternlicht et al. (2006) also showed an association with shorter overall survival for high PAI-1 mRNA expression in two breast cancer data sets (van de Vijver cohort, n = 295, and University of California San Francisco (UCSF) cohort, n = 118). In the UCSF dataset, PAI-1 was prognostic in node-positive patients, in the van de Vijver dataset in node-negative patients. However, more studies with clinical follow-up data of the patients are necessary to verify the influence of PAI-1 mRNA expression on clinical outcome in patients with breast cancer.

The rate of elevated uPA and PAI-1 protein levels in our patient cohort was in the range of published studies (Harbeck et al. 1999) but almost 70% of the patients were node positive. This might be the explanation why uPA which was mainly evaluated as an important marker in node-negative breast cancer patients had no impact on patient outcome in our study.

Potential drawbacks of our study are that patients received adjuvant treatment not regarded as current standard. Trastuzumab, e.g., was not available during the collection time. Nevertheless, we were able to collect clinical follow-up data for all patients, in contrast to many published studies evaluating methods for determination of biological factors in breast cancer.

In conclusion, we found that prognostic and predictive markers like HER2 and uPA can be measured by RNA-based analyses in patients with breast cancer. We observed additional prognostic information for PAI-1 mRNA expression. Our findings indicate that gene expression analyses deliver high-quality determination of biological markers. With expression profiling now integrated into several ongoing multinational breast cancer adjuvant trials, attention to breast tumour PAI-1 mRNA levels, measured alone or as part of gene expression signature, will further enhance the prognostic and predictive value of this important biomarker.