Cellular senescence predicts treatment outcome in metastasised colorectal cancer

Background: Cellular senescence is a terminal cell-cycle arrest that occurs in response to activated oncogenes and DNA-damaging chemotherapy. Whether cancer cell senescence at diagnosis might be predictive for treatment outcome is unknown. Methods: A senescence index (SI) was developed and used to retrospectively correlate the treatment outcome of 30 UICC stage IV colorectal cancer (CRC) patients with their SI at diagnosis. Results: 5-Fluorouracil/leucovorin-treated CRC patients achieved a significantly longer progression-free survival when presenting with SI-positive tumours before therapy (median 12.0 vs 6.0 months; P=0.044). Conclusion: Cancer cell senescence predicts treatment outcome in metastasised CRC. Prospective analyses of larger patient cohorts are needed.

In the past, many cancer studies sought to link quantitative assessments of tumour growth properties, namely proliferation and apoptosis, to treatment outcome, but the majority of these analyses produced negative results for various reasons. Although relevant as a prognosticator, the value of using the mitotic index to predict treatment outcome is very limited (see Schmitt, 2003Schmitt, , 2007Brown and Attardi, 2005;Yerushalmi et al, 2010 and references therein for review). A third cancer-related growth condition is oncogene-induced senescence (OIS), a terminal cell-cycle arrest initiated by activated Ras-type oncogenes that functions as a tumour-suppressive barrier in pre-malignant lesions in vivo (Braig et al, 2005;Michaloglou et al, 2005;Collado and Serrano, 2010;Reimann et al, 2010). Oncogene-induced senescence is mediated through an oncogene-evoked DNA damage response (DDR) (Bartkova et al, 2005(Bartkova et al, , 2006Gorgoulis et al, 2005), thereby explaining why DNA-damaging chemotherapy produces senescence as well (Chang et al, 1999;Schmitt et al, 2002). Therapyinduced senescence (TIS) contributes to treatment outcome in pre-clinical models (Schmitt et al, 2002), and is detectable in patient cancer biopsies after neoadjuvant chemotherapy (te Poele et al, 2002;Roberson et al, 2005), but its long-term effect on clinical courses remains undetermined. We hypothesised that senescent cells present sporadically in untreated cancers at diagnosis may indicate an increased susceptibility to TIS, and, thus, might be associated with a superior clinical outcome. Because the gold-standard senescence assay, detection of the senescenceassociated b-galactosidase (SA-b-gal) activity (Dimri et al, 1995), cannot be applied to formalin-fixed and paraffin-embedded (FFPE) clinical routine samples, we developed a senescence index (SI) based on FFPE-suitable immunohistochemical surrogate markers. We report here the first-time use of this SI in untreated human tumour specimens at diagnosis as a predictor for outcome to cancer therapy.

Patients and tumour material
Snap-frozen or FFPE samples of normal colon mucosa, colon adenoma and colon carcinoma were collected, and informed patient consent was obtained for the anonymous use of the material. FFPE tumour samples from previously untreated 30 patients with UICC stage IV colorectal cancer (CRC) all of whom received a 5-fluorouracil/leucovorin (5-FU/LV)-based first-line regimen were subjected to senescence evaluation, reflecting the available archive material in our pathology department between the years 1996 and 2002 that met these criteria. Tissue material from patients suspected or diagnosed with a hereditary cancer syndrome or inflammatory bowel disease, or whose chemotherapy was terminated for reasons other than disease progression was excluded from this study. Clinical data sets were compiled in a retrospective and anonymous manner, and included gender, age, tumour location, carcinoembryonic antigen (CEA) serum levels, TNM classification, tumour grading and progression-free survival (PFS), defined as the time from first diagnosis until diagnosis of progressive disease (i.e. local recurrence of the primary lesion, or growth of metastases or occurrence of new metastatic lesions).

Retroviral infection
IMR90 human fibroblasts (CCl-186 from American Type Culture Collection, Manassas, VA, USA) were stably transduced with the pBabe-Puro-H-RasV12 retrovirus (provided by S Lowe) or an empty vector as a control, and selected in puromycin as described (Serrano et al, 1997;Braig et al, 2005). Cell pellets were snap frozen in liquid nitrogen or processed as FFPE pellets comparable to patient tissue samples.
The formula and the cut-off values of the SI were generated based on the expression (i.e. the percentage of positive cells by immunostaining) of the three markers p-ERK, HP1g and PAI-1 in low-proliferating areas (arbitrarily defined as o12 Ki67-positive in an area of 100 cells) in a learning set of five adenoma samples. For each of the three markers, a factor was generated that reflects the reciprocal percentage of cells that stained positive in such areas averaged over these five adenomas. This factor was used as a coefficient to equalise the marker's relative individual contribution to the SI. Moreover, a linear correction value was added to set the discrimination threshold to 0. Values between À1 and þ 1 were considered 'non-conclusive'. Senescence index values of individual cancer samples were obtained in low-proliferating areas as well (see Figure 2 for further information on SI-related technical procedures). Proportions of senescent cases in cryo-preserved colorectal tissue sections of the indicated groups (normal crypt mucosa (n ¼ 5, 0 senescent); adenoma (n ¼ 12, 8 senescent), invasive carcinoma (n ¼ 6, 1 senescent)). Note that 9 out of 12 adenoma and 2 out of 6 carcinoma cases would score 'senescent' if judged in low-proliferating areas only (see D). (B) Average frequency of senescent cells per group (as in A), measured as the percentage of SA-b-gal-positive (blue) cells, and compared to the rate of Ki67-positive (red nuclear staining) cells (top: error bars represent the standard deviation; bottom: matched areas of representative photomicrographs -a non-senescent normal crypt mucosa, a senescent adenoma and a formally non-senescent carcinoma (but showing numerous senescent cells)). Notably, in normal colorectal mucosa, only cells of the most differentiated luminal mucosa stain SA-b-gal-positive. (C) FFPE sections of pelleted Ras-infected senescent (sen.) human fibroblasts (SA-b-gal-positive and Ki67-negative frozen material as a reference) show much stronger immunoreactivity for p-ERK, HP1g and cytoplasmic PAI-1 when compared with non-senescent (non-sen.) mock-infected fibroblasts. (D) SI values, based on the expression of these three markers, were obtained in 30 cases of stage IV CRC specimens at diagnosis, and used to stratify PFS following 5-FU/LV first-line chemotherapy (senescent (n ¼ 12; blue line) vs non-senescent (n ¼ 17; red line); one case scored 'not conclusive' (see Figure 2B)). Genomic sequencing in a subset of these 29 specimens identified K-Ras codon 12 or 13 mutations in 5 out of 6 senescent, but only 2 out of 11 non-senescent cases (inset). Note that a primary stratification by the K-Ras mutation status unveiled no significant differences in PFS (P ¼ 0.128).

Senescence-associated b-galactosidase activity
Senescence-associated b-galactosidase activity was detected at pH 6.0 in cryo-preserved cells or tissue sections as described (Dimri et al, 1995;Braig et al, 2005). Cases were considered senescent if their mean percentage of SA-b-gal-positive cells was higher than the mean percentage of Ki67-positive cells.

Ras mutation analysis
Genomic DNA was extracted from macro-dissected tumour areas of FFPE tissue sections using the QIAamp DNA Mini kit (Qiagen, Hilden, Germany). PCR amplification of the first exon of the k-ras gene was performed according to van den Brandt and colleagues (Brink et al, 2003). The resulting 179 bp PCR product was sequenced using the BigDye Terminator v1.1 Cycle Sequencing kit on a 3130 Genetic Analyzer (Applied Biosystems, Foster City, CA, USA). Hotspot codon 12 or 13 mutations were detected by comparison with the germ-line sequence.

Statistical analysis
Statistical analyses (SPSS software package, release 17.0; SPSS, Munich, Germany) of Kaplan -Meier survival plots were based on the log-rank (Mantel -Cox) test; additional statistical comparisons used the Wilcoxon -Mann -Whitney test (to probe equal distribution of age, frequency of Ki67-reactive and cleaved caspase-3positive cells, and serum CEA levels in patient subgroups), the Fisher's exact test (with respect to gender, tumour localisation, sites of distant metastasis (one vs more than one) and Ras mutations) and the w 2 -test (regarding the pT and the pN status). P-values o0.05 were considered statistically significant.

RESULTS
We analysed 23 snap-frozen colorectal tissue samples of normal mucosa, adenomas and untreated invasive carcinomas by the SA-bgal assay, complemented by immunostaining for the proliferation marker Ki67 and confirmed the reportedly high frequency of senescent cells in adenomas, leading to the classification of 8 out of 12 adenoma cases tested as senescent, which is well in line with their macroscopic presentation as polyps of often stable size for years ( Figure 1A and B; see the flow diagram in Figure 2A for technical details) (Bartkova et al, 2006). Importantly, as previously reported for a subset of lymphoma, lung and breast cancer specimens at manifestation (te Poele et al, 2002;Roberson et al, 2005;Reimann et al, 2010), we detected a significant fraction of senescent cells within neoplastic epithelial areas of manifest colorectal carcinomas, possibly indicating a still available senescence programme at this full-blown cancer stage ( Figure 1A and B).
Because the enzymatic SA-b-gal assay cannot be applied to FFPE routine samples, or substituted by a single marker, we tested protein expression levels of a panel of DDR mediators, cell-cycle regulators, chromatin-related proteins and others (Bartkova et al, 2006;Collado and Serrano, 2006) in a well-established OIS reference model system -sections of either snap-frozen or FFPE pellets of Ras-senescent vs proliferating fibroblasts -that allowed us to compare SA-b-gal staining and potential immunohistochemical surrogate markers in the same material side by side (Serrano et al, 1997). Using this fibroblast model system followed by first a learning and then a validation set of adenoma samples, we generated a senescence surrogate score (SI) based on the expression of p-ERK, HP1g and PAI-1 -all of them previously linked to OIS (Serrano et al, 1997;Lin et al, 1998;Narita et al, 2003) -that recapitulated SA-b-gal reactivity and adenoma senescence ( Figure 1C; Figure 2). Apply SI to CRC stage IV samples at diagnosis (n = 30), test independence from clinical and molecular parameters, and correlate with outcome to therapy (see Figure 1D) as frozen samples Exploit/confirm senescence by enzymatic SA--gal activity and Ki67 immunostaining in situ (see Figure 1A-C) Exploit/confirm senescence surrogate markers by immunostaining in situ (see Figure 1C) Adenoma learning set (n = 5) to generate a senescence surrogate signature (SI; see Figure 2B) Independent adenoma validation set (n = 17) to confirm the SI (see Figure 2C) We obtained the SI in low-proliferating areas of 30 independent UICC stage IV CRC samples before any drug treatment, and assigned these cases, based on a cut-off value, to a (partly) senescent vs a non-senescent group. All patients in this retrospective analysis received a 5-FU/LV-based first-line regimen. Patients with senescent areas in their tumour biopsies experienced a significantly longer PFS with a median PFS of 12.0 vs 6.0 months when compared to the non-senescent group ( Figure 1D). Notably, both groups did not show statistically significant differences regarding gender, age, pT and pN status as well as localisation of the primary tumour (colon vs rectum), extent of distant metastasis (one vs more than one site) or CEA serum levels at diagnosis, and showed indistinguishable rates of proliferation (i.e. averaged sample-wide Ki67 reactivity) and apoptosis (i.e. mean frequency of cleaved caspase-3-positive cells) (data not shown). Interestingly, K-Ras codon 12/13 mutations were found at a higher frequency (P ¼ 0.035) in the senescent group ( Figure 1D, inset), suggesting that Ras mutations, if associated with a senescent phenotype, may not necessarily serve as a predictor of poor outcome, as reported by several studies in the past (Benhattar et al, 1993;Andreyev et al, 1998Andreyev et al, , 2001Roth et al, 2010).

DISCUSSION
This is the first report linking cellular senescence at diagnosis to treatment outcome in cancer. Whether sporadic senescent cells in manifest tumours represent a few remainders of the premalignant condition, or indicate retained senescence susceptibility throughout the tumour is currently not clear. Our study data favour the latter explanation, postulating that TIS produced in response to DNA-damaging chemotherapy in senescence-capable tumour cells contributes to the overall outcome to therapy. Notably, an inverse correlation between the expression level of another senescence marker, macroH2A, and the risk of recurrence has recently been reported for lung cancer patients (Zhang et al, 2005;Sporn et al, 2009). Additional studies are certainly needed to clarify whether a greater extent of sporadic cancer cell senescence indeed translates into a higher frequency of TIS-positive cells in situ when analysed in re-biopsies a few days after chemotherapy. Moreover, a preserved pro-senescent cancer capability might be therapeutically exploitable by novel senescence-inducing strategies that do no longer damage DNA. The role of senescent cells at diagnosis as a novel predictor of treatment outcome should be further evaluated in larger prospective trials.