Tissue samples were collected from 80 patients with liver cirrhosis who underwent surgical resection for HCC, between 1997 and 2007, in two hospitals (Istituto Clinico Humanitas and San Paolo Hospital, Milan). The study was carried out with Local Ethical committee approval. Patients included 62 men and 18 women (mean age: 67 years, range: 42–83). The number of patients positive for HCV, HBV was 54 and 9, respectively. None of the patients received chemotherapy, radiation and/or alcoholization therapy before surgery. In each case, samples of HCC and surrounding cirrhotic tissue were snap-frozen into cryovial with 1 ml of RNAlater^® and stored at -80°C. Routinely Hematoxylin-Eosin-stained sections of HCCs and surrounding tissue were reviewed by two pathologists according to WHO (2000) and TNM (2002). The HCCs were histologically graded as I-IV on the basis of cellular atypia and architectural complexity, according to Edmonson. The evaluated clinicopathological features were: sex, age at diagnosis, viral infection, tumour histological features (histotype, stage, number and diameter of nodules, presence of capsule, microvascular invasion, Edmonson's grade of differentiation). These are summarized in table 1.

Clinical outcome data were available in 69 patients (86%). The follow-up period ranged from 1 to 75 months (average 34 months). At the last follow-up, 15 patients were deceased for HCC disease, whereas 22 patients were alive with HCC recurrence.

Samples were homogenized in 1 ml of TRIzol^® reagent (Invitrogen, Milan, Italy) with a tissue lyser (Qiagen). Total RNA was purified according to the manufacturers' protocol. Spectrophotometrical RNA quantification was conducted by GeneQuant II (Pharmacia Biotech) at 260 nm. Total RNA was stored at -80°C until molecular investigation was performed. In reverse transcription reactions, 1 μg of total RNA from each sample was used for cDNA generation in a final reaction volume of 100 μl with High Capacity cDNA Archive Kit (Applied Biosystems).

qPCR reactions of target (n = 9) and housekeeping (HKG n = 8) genes were performed using Assay-on-Demand™ chemistry in an ABI PRISM 7900 HT Sequence Detection System (Applied Biosystems, Foster City, CA, USA). The assay identification numbers of target and housekeeping genes are as follow: Hs00244967_m1 (BIRC1), Hs00357350_m1 (BIRC2), Hs00154109_m1 (BIRC3), Hs00236913_m1 (XIAP), Hs00213882_m1 (XAF1), Hs00153353_m1 (BIRC5), Hs00212288_m1 (BIRC6), Hs00223384_m1 (BIRC7), Hs00219876_m1 (Smac/DIABLO), Hs00609297_m1 (HMBS), Hs99999905_m1 (GADPH), Hs 99999907_m1 (β-2M), Hs00188166_m1 (SDHA), Hs00824723_m1 (UBC), Hs00427620_m1 (TBP), Hs99999909_m1 (HPRT1), Hs99999903_m1 (ACTβ). The BIRC5 assay is specific for its two splicing variants: NM_001168.2 and NM_001012271.1 (Survivin and Survivin2B, respectively). The BIRC7 assay recognizes all splicing variants of this gene.

Instrument raw data (fluorescence) of all the samples were converted in threshold cycles (Ct) by SDS 1.2 software (Applied Biosystem, Foster City, CA, USA). Ct values were then imported in Excel worksheet for relative quantification (RQ).

For RQ calculation, the geometrical mean of the three more stable HKGs (HMBS, β-2M and GAPDH, GeNorm software) [9], was used as normalization factor.

The Fold Change (FC) was calculated, defined as the ratio between averaged normalized expression level of targets in neoplastic and corresponding non-neoplastic samples. Normalized RQ were log2-transformed for statistical analysis.

The ratios of XIAP and its antagonists were calculated as follows: first, for every patient, the ratio between normalized expression level of XIAP and its antagonists both in cancer and in cirrhotic tissue was calculated. Then the average value of these single ratios for all cancerous and cirrhotic samples, respectively, was calculated. Finally the fold change ratios of XIAP/XAF1 and XIAP/Smac were calculated.

Representative tissue blocks from surgical resection of 40 patients (all included in qPCR study) were chosen to construct 2 TMA as previously described [10]. For each patient, seven cylindrical tissue cores were included in the TMA: three from carcinoma, two from cirrhotic tissue close to the cancer (< 1 cm) and two cirrhotic tissues taken at a distance of at least 3 cm from neoplastic lesion. From each TMA block a 4 μm-thick section was cut, stained with H&E and inspected for adequacy before immunohistochemical studies.

Serial 4 μm-thick sections from each TMA block were stained with a series of primary antibody: XIAP (clone 48, BD Biosciences, San Jose, CA USA) 1:100; XAF1 (ab32023, Abcam, Cambridge, USA) 1:100; full-length Survivin (NB 500-201, Novus Biologicals, Littleton, USA) 1:2000. Immunohistochemistry was performed using a Dako immunostainer (Dako, Glostrup, Denmark) and immunostaining was revealed by Dako EnVision™ Detection Kit with Peroxidase/DAB as chromogen. All slides were counterstained with haematoxylin.

Immunohistochemical results were evaluated by two pathologists blinded to clinical data. Intensity and percentage of positive cells were calculated by averaging out replicate cores. The percentage of immunoreactive tumor cells and cirrhotic liver cells was determined and assigned to one of the following five categories: 0, (<5%); 1, (5%–24%); 2, (25%–49%); 3, (50%–74%) and 4, (≥75%). The intensity of immunostaining was scored as absent (0), mild (1), moderate (2) and marked (3). The percentage of positive cells and staining intensity were multiplied to produce a weighted score for each case. Cases with weighted score 0 were defined as negative; 1–6 as low expressors and 7–12 as high expressors.

Statistical analysis was conducted for both sets of results: those from qPCR and those relative to immunohistochemical studies. For samples clustering (qPCR results), log2-transformed data were imported in dChip software (dChip 2006, DNA-Chip Analyzer available for free at http://www.hsph.harvard.edu/~cli/complab/dchip). Relative abundance of each gene was standardized by software pre-processing steps (subtracting its mean and dividing the result by its standard deviation) and hierarchical clustering analysis was then performed. The 1-Rank correlation was employed as distance metric and the Average as a linkage method.

The genes with a global FC exceeding 2 or lower than 0.5 were statistically analyzed by univariate statistics (paired t test).

For statistical purposes tumors with Edmondson's grade I and II, III and IV were categorized in "low" and "high" grade subsets, respectively. HCCs staged pT3 and pT4 were merged in the same group. Correlations of gene and protein expression with clinicopathological features (histotype, viral infection, stage, grade, vascular invasion, perineoplastic capsule, number of nodules, and diameter of nodules) were analyzed by ANOVA and by Chi-square, respectively.

To investigate whether expression levels of each gene and protein were associated to patient survival, HCC samples were categorized in high or low expressor groups if the target level was above or below the median expression value, respectively. For XIAP/XAF1 and XIAP/Smac mRNA ratios correlation to patients' outcome, HCCs were categorized in "low" and "high" groups if the ratio value was lower or higher than 2, respectively.

Overall and disease-free survival curves of HCC patients were plotted according to the Kaplan-Meier method using the log-rank statistics to test for statistical significant difference of generated curves (GraphPad Prism version 4.00 for Windows, San Diego, California, USA, http://www.graphpad.com). Data with a p value lower than a threshold limit (p < 0.05) were considered significant.

All investigated genes were expressed at detectable levels (Ct <40) in all HCCs and corresponding non-neoplastic samples. To explore whether HCC and non-neoplastic tissue could be distinguished at molecular level by the expression profile of IAPs and inhibitors, unsupervised hierachical clustering (dChip) was performed. The majority of HCC samples (K, n = 62, 88%, p < 0.0001 by Fisher's exact test) clustered in the same branch of the dendrogram clearly separated from matched non-neoplastic counterparts (N) (see Additional file 1).

Survivin and Livin/ML-IAP mRNAs exhibited significant overexpression in cancer tissue compared to non-neoplastic tissue (FC = 6.86, p < 0.001 and FC = 2.33, p < 0.001, respectively) (table 2). Only NAIP mRNA revealed a slightly lower expression level in cancer tissue compared to non-neoplastic tissue, although the fold change was not significant (FC = 0.67). The expression of other genes was comparable between cancer and non-neoplastic tissue (FC between 0,91 and 1.15).

Expression values of XAF1 and Smac/DIABLO were comparable between HCCs and non-neoplastic tissues (FC = 0.88 and FC = 1.10 respectively).

XIAP/XAF1 ratio was significantly higher in tumor than in non-neoplastic parenchyma (FC = 3.02, p < 0.001; table 2). On the contrary Smac/DIABLO expression did not show any significant difference between HCC and non-neoplastic liver (FC = 1.09).

A significant correlation was found between high Survivin mRNA level and high tumor stage (pT3 and pT4, p = 0.03), high tumor grade (III and IV, p = 0.01) and vascular invasion (p = 0.001). Moreover, high c-IAP2 mRNA level was found to be significantly correlated with absence of perineoplastic capsule (p = 0.02) and high NAIP mRNA level with pseudoglandular histotype (p = 0.03). Finally, high mRNA levels of c-IAP1 and Smac/DIABLO genes were significantly correlated with younger patient age (p = 0.03 and p = 0.009 respectively).

Livin/ML-IAP and Survivin overexpression showed no correlation with disease outcome, although there was a trend for shorter overall survival in patients with high Survivin expression (p = 0.09). Kaplan-Meyer analysis demonstrated a significantly shorter overall survival in patients with high XIAP/XAF1 ratio (p = 0.03) (figure 1).

The gene expression profiles obtained suggested that Survivin, XIAP and XAF1 could be important in determining clinical outcomes of HCC patients. To confirm this hypothesis, we investigated Survivin, XIAP and XAF1 protein expression by immunohistochemistry on a tissue microarray platform.

Survivin immunoreactivity was intense in the cytoplasm of cirrhotic liver cells, whereas HCC cells generally showed weak cytoplasmic staining (figure 2). Nuclear Survivin immunoreactivity was found only in five HCC (12,5%). Survivin immunoreactivity was not associated with any clinical variable: although samples of HCC with high level of Survivin were correlated with a shorter overall survival, this trend did not reach statistical significance (p = 0.39).

XIAP immunoreactivity was detected exclusively in cytoplasm of both neoplastic and non-neoplastic liver cells. Most HCC (82,5%, n = 33) displayed XIAP immunoreactivity, whereas 70% (n = 28) of cirrhotic liver tissue (distant from the tumor) expressed XIAP (table 3). Survival analysis demonstrated a significantly shorter overall survival in patients with high XIAP expression (p = 0.03) (figure 3); no correlations with other clinicopathological parameters were found.

XAF1 immunoreactivity was detected in the cytoplasm and nucleus of neoplastic and cirrhotic liver cells (figure 3). More than fifty percent of the HCC and cirrhotic tissues did not show XAF1 expression (table 3) and the expression score of XAF1 was significantly lower than that of XIAP. XAF1 immunoreactivity did not reveal any significant correlation with clinicopathological parameters and survival.

Since qPCR results showed that XIAP/XAF1 ratio strongly correlated with prognosis, we divided patients in four groups for survival analysis, depending on the following immunoreactivity patterns: group 1, XIAP low and XAF1 negative; group 2, XIAP low and XAF1 positive; group 3, XIAP high and XAF1 negative and group 4, XIAP high and XAF1 positive. Interestingly, groups 3 and 4 showed a trend for worse disease free and overall survival, although it did not reach statistical significance (p = 0,10 and p = 0,08 respectively) (data not shown).