There have been frequent descriptions of apoptosis in gastrointestinal stromal tumor (GIST) cell lines treated with imatinib (Glivec, Novartis, Basle, Switzerland) or receptor tyrosine kinase (RTK) inhibitors [1,2], but apoptosis-associated morphological changes are only occasionally encountered in surgical samples of GISTs.

For the last year, we have histologically examined a number of surgical specimens taken after imatinib treatment in which, in addition to previously described highly responding areas of complete cell depletion and poorly responding areas with unchanged tumoral cellularity [3,4], we found areas in which the tumoral cells showed prominent intracytoplasmic vacuoles and relatively intact nuclei in the absence of nuclear and cytoplasmic condensation and nuclear fragmentation. These findings, which have previously been described at microscopic level in untreated GISTs [5], and at microscopic and ultrastructural levels in imatinib-treated GISTs [4], are consistent with autophagic rather than apoptotic changes.

Autophagy is a dynamic process in which portions of the cytoplasm are sequestrated within a double-membrane vesicle (or autophagosome) that fuses to the lysosome to generate what are known as autolysosomes or vacuoles. The content of the autophagosome is released into the autolysosome as a result of the action of lysosomal enzymes and then degraded by the same enzymes [6,7]. This degradation can promote cell survival by recycling the degraded nucleotides, amino acids, and fatty acids that maintain energy production or can promote cell death as a result of self-cannibalization [8].

The main difference between autophagy and apoptosis is that the latter invariably leads to cell death, whereas the former may contribute to cell survival or death depending on the threshold level: high levels of autophagy promote cell death [9].

Autophagy is known as programmed cell death type II (apoptosis is programmed cell death type I) and is highly regulated by means of a tumor-suppressor mechanism [10]. It is molecularly controlled mainly by constitutively expressed autophagy-related genes: beclin1 (the major player) and PI3K class I (PI3KI) and class III (PI3KIII). Beclin1 is a haploinsufficient tumor-suppressor that is not required for apoptosis but is necessary for autophagy [11,12]. PI3KI and PI3KIII have opposite effects on autophagy, which is inhibited by the activation of PI3KI but initiated by PI3KIII forming a complex with beclin1 [13,14]. Beclin1 can also bind bcl2, which, in addition to being an important regulator of apoptosis, may inhibit autophagy by directly interacting with beclin1. Growing evidence suggests that the beclin1/bcl2 complex may act as a rheostat ensuring the threshold for cell homeostasis or cell death depending on the presence of a beclin1 function that is respectively checked or unchecked by bcl2 [15,16]. Another widely accepted molecular autophagic marker is the posttranslational modified microtubule-associated protein light chain 3 (LC3-II), which is derived from cytosolic LC3-I because of lipidization during autophagy. LC3-II closely binds to autophagosomes, and its amount closely correlates with the number of autophagosomes [17]. Interestingly, the regulatory pathway of autophagy shares a number of molecules with the oncogenic pathways activated by RTKs, such as the PI3KI/AKT/mTOR and RAS/RAF/MEK1/2/MAPKs pathways. The first inhibits autophagy and activates cancer growth, whereas the second can promote autophagy and cancer growth [18]. The activated RTKs in GISTs areKITand PDGFRA, together with their downstream effectors (PI3KI/AKT and MAPKs), and imatinib is their main starvation inducer.

Within the framework of this quite new scenario, we evaluated the morphological, biochemical, and immunophenotypical profiles expected to be related to autophagy and apoptosis in a series of molecularly characterized GISTs taken from surgically resected imatinib-treated patients. The results showed no signs of apoptosis but a number of indirectmarkers of autophagy, which is in keeping with clinical observations suggesting prompt tumoral regrowth when imatinib is discontinued.

Table 1 Clinical Findings in Imatinib-Treated and Untreated Patients, Molecular Characterization, and Histological Response Assessment on Nodules of Treated Cases.

Clinical findings Tumor location and TNMstatus (primary [P], recurrence [R], and metastasis [M]) and the dose and duration of imatinib treatment applied to advanced (number in bold) and neoadjuvant (number in italic) cases are shown in Table 1. The neoadjuvant treatment protocol was generally shorter (approximately 12 months) than that used for surgically treated advanced cases. Imatinib treatment was discontinued 12 to 36 hours before the surgical procedures in all cases. Nine patients were judged to be clinical responders (r) and two clinical progressors (p). The two clinical progressors belonged to advanced cases and showedmicroscopically both high mitotic index andMib1 labeling, and the GIST corresponding to case no. 11, harbored a secondarymutation. Clinically, the responsewas assessed by means of both the Response Evaluation Criteria in Solid Tumors and the Choi criteria [21,22].

Gene mutation analysis The samples of the patients that had not been previously characterized were analyzed for c-Kit and PDGFRA gene mutations as described by Miselli et al. [3]. Among c-kit mutations, one secondary mutation at exon 17, known to be related to imatinib resistance, was observed. No PDGFRA gene mutations were detected. The molecular findings in all of the patients are shown in Table 1.

Histological response assessed on paraffin-embedded surgical specimens One hundred five samples/nodules from the 11 surgical specimens were assessed for tumor response based on residual cellularity, the mitotic index, and Ki-67 labeling, and the results were averaged for each case. Residual cellularity was scored as follows: a, <10%; b, 10% to 50%; c, 50% to 90%; and d, >90%. Averaging yields the following scores: high responders, 0 to <50% residual viable tumoral cells with no mitosis and no obvious Ki-67 immunostaining; moderate responders, >50% to 90% tumoral cells, no mitosis, and Ki-67 immunostaining in 0 to <10% of cells; low responders, >50% to 90% tumoral cells, mitotic index >10/50 high-power fields, Ki-67 immunostaining in 20% to 30% or >30% of cells.

Histological response assessed on frozen material The 11 frozen specimens obtained from the 11 treated patients (one sample for each case) were analyzed to assess the response to imatinib based on the morphological criterion of residual cellularity (scored as described above). For details, see Table 1.

Positive controls The NIH3T3 (American Type Culture Collection, Manassas, VA), HeLa (kindly provided by Dr. A. Greco), and HEK293 cell lines (kindly provided by Dr. A. Greco) were used as positive controls for the beclin1, PI3KIII, bcl2, and LC3-II blots; the staurosporine-treated HeLa cell line was used as a positive control for the caspase 3, caspase 7, lamin A/C, and LC3-II blots.

Protein extraction, Western blot analysis, and coimmunoprecipitation The cytoplasmic and nuclear proteins were extracted as previously described [24,25]. The protein extracts underwent electrophoresis and immunoblot analysis using standard protocols and 30 µg of cytoplasmic or nuclear protein lysates per sample.

The antibodies tested by means of immunoblot analysis included rabbit polyclonal anti-beclin1 (H-300: sc-11427; Santa Cruz Biotechnology, Santa Cruz, CA) diluted 1:200; rabbit polyclonal anti-PI3KIII (anti-Vps34, 38–2100; Zymed Laboratories Invitrogen immunodetection, South San Francisco, CA) diluted 1 µg/ml; mouse monoclonal anti-bcl2 (clone 124 code: M0887; DakoCytomation, Glostrup, Denmark) diluted 1:200; rabbit-polyclonal anti-LC3-II (LC3B, #2775; Cell Signaling Technology, Danvers, MA) diluted 1:1000; rabbit polyclonal anti-caspase 3 (#9662; Cell Signaling Technology) diluted 1:500; rabbit polyclonal anti-cleaved-caspase 7 (Asp 198, #9491; Cell Signaling Technology) diluted 1:1000; rabbit polyclonal anti-lamin A/C (#2032; Cell Signaling Technology) diluted 1:1000; and antiactin (A2066; Sigma, St. Louis, MO) diluted 1:5000.

For the coimmunoprecipitation experiments, 1 mg of protein lysate per sample was immunoprecipitated by means of incubation with 5 µl of rabbit polyclonal anti-beclin1 antibody (sc-11427; Santa Cruz Biotechnology) and protein A Sepharose (Sigma). Immunoblot analysis was performed using the same anti-beclin1 antibody, and the filters were then stripped and incubated with anti-PI3KIII and antibcl2 antibodies. The secondary antibodies included anti-rabbit (Sigma), anti-mouse (Sigma), and protein A-HRP (Amersham Biosciences, Piscataway, NJ) used at the recommended dilutions. After hybridization with the secondary antibody, the blots were incubated with ECL Western Blotting Detection Reagents (Amersham Biosciences) and exposed onto ECL Hyperfilm (Amersham Biosciences).

Regression changes in treated patients We analyzed b samples (with residual cell component of 10–50%) and c samples (with a residual cell component of 50–90%) in which acellular areas characterized by the presence of an eosinophilic proteinaceous matrix variously defined as myxoid, collagenous, or hyaline [5,26], and often associated with ectatic vessels surrounded by hyaline sclerosis (called a: residual cell component, <10%), were intermixed with clusters of viable tumoral cells with prominent perinuclear vacuolization and relatively intact nuclei in the absence of nuclear or cytoplasmic condensation and fragmentation: i.e., devoid of apoptosis hallmarks. These multivacuolated tumoral cells showed nuclear immunostaining with full-length lamin A/C antibody and a null immunophenotype with cleaved lamin A/C antibody. Scattered lymphocytes negative for lamin A/C and positive for CD3 (data not shown) were also present. Areas of more regressive changes were found in which the tumoral cells had lost their morphology and acquired histiocytic features (corresponding to a samples Figure 1). We also analyzed d samples (residual cell component, >90%) and two samples from untreated patients.

Figure 1 Post-imatinib morphological regression changes ranging from an acellular area characterized by the presence of eosinophilic proteinaceous matrix (A) sometime intermingled with ectatic vessels surrounded by hyaline sclerosis (B) and clusters of viable (more ...)

Table 2 Biochemical and Immunophenotypical Characterization of Samples From Imatinib-Treated and Untreated Patients.

Figure 2 Autophagy-related protein expression in imatinib-treated and untreated GISTs (A and B) and their interactions in the autophagic process (C). (A) Western blot analyses using anti-PI3KIII, antibeclin1 and anti-bcl2 antibodies: the arrows indicate the bands (more ...)

Beclin1 and PI3KIII expression Similar 2+ expression of beclin1 and PI3KIII was observed in the untreated cases and the treated low responders. In the moderate and high responders, beclin1 expression corresponded to respectively 2+ and 1+, whereas, although retaining the same trend throughout the groups, PI3KIII was less expressed, ranging from weakly positive (+w) to 1+. In no. 7, a sample in which residual cellularity was very low, beclin1 was positive but PI3KIII-negative.

Bcl2 expression The expression of bcl2 in the untreated cases and low responders was more heterogeneous than the two autophagyrelated markers described above, and this tendency was confirmed by IHC analysis (data shown later in the Table 2). This finding is in keeping with the observed variability in bcl2 expression in primary tumors [27] and that found in the two unselected control GISTs, in which bcl2 expression was 1+ and 2+ (comparing the expressions of tumoral cells and in-built lymphocytes). On the contrary, in the moderate and high responders, the expression profile of bcl2 was the same as that of PI3KIII.

LC3-II expression Excluding the two negative cases among the high responders, the results were similar to those relating to PI3KIII in 9 of the 11 treated and untreated cases, in keeping with the role of PI3KIII in initiating autophagy and that of LC3-II in a later phase. Furthermore, as the amount of LC3-II correlated well with the number of autophagosomes, the absence of LC3-II in the presence of PI3KIII is consistent with their degradation in the two most regressing samples (b).

Coimmunoprecipitation experiments to confirm the presence of autophagy The interactions between beclin1/PI3KIII and beclin1/bcl2 were assessed by means of coimmunoprecipitation experiments. The results are shown in Table 2 and some cases in Figure 2C.

Beclin1 All of the protein extracts from each sample immunoprecipitated using an antibody specific for beclin1 protein showed a corresponding 60-kDa band after membrane incubation. The intensity of the band (1+, 2+) varied from sample to sample in accordance with the expression revealed by the WB experiments and was in line with residual cellularity.

Beclin1/PI3KIII interaction With the exception of one case (no. 7), all of the evaluable samples coexpressed beclin1 and PI3KIII. Given the differences in residual cellularity between the high/moderate and low responders/untreated cases, coexpression was definitely greater in the former.

Beclin1/bcl2 interaction The range of beclin1 and bcl2 coexpression (from +w to 1+) was smaller than that of beclin1 and PI3KIII. Given also that the expression levels of PI3KIII and bcl2 were similar in the high and moderate responders, these findings support the view that a small amount of bcl2 protein directly interacts with beclin1.

Taken together, the high levels of proautophagy beclin1/PI3KIII complexes and low levels of antiautophagy beclin1/bcl2 complexes are consistent with the presence of autophagy.

Expression of apoptosis-related proteins All of the samples were analyzed by means of WB experiments to assess the expression of full-length and cleaved caspase 3, cleaved caspase 7, and full-length and cleaved lamin A/C proteins. The results are shown in Table 2.

Caspase 3 and lamin A/C were only present as full-length proteins in all of the treated patients (regardless of their residual cellularity or clinical/histological group) and the untreated patients. The caspase 7 antibody, which only recognizes the large fragment of the enzyme resulting from apoptosis-induced cleavage, was not found in any of the treated or untreated cases (Figure 3, which shows some representative cases).

Figure 3 Apoptosis-related protein expression in imatinib-treated and untreated GISTs. The WB analyses were performed using anti-caspase 3, anti-cleaved caspase 7, and anti-lamin A/C antibodies: the arrows indicate the bands corresponding to the expression of (more ...)

Taken together, the results show the absence of apoptotic hallmarks in all of the patients.

Expression of autophagy-related proteins To confirm the expression of autophagy-related proteins, sections obtained from paraffinembedded samples whose residual cellularity was comparable with that of the corresponding frozen samples were screened for beclin1, PI3KIII, and bcl2.

The imatinib-treated and untreated cases all showed cytoplasmic reactivity for beclin1, PI3KIII, and bcl2. The results of the IHC analysis are shown in Table 2 and Figure 4A.

Figure 4 Immunohistochemical analysis of two imatinib-treated GISTs, showing residual cellularity b (case 2) and c (case 5). (A) Expression of autophagy-related proteins. Both cases show cytoplasm decoration for PI3KIII, beclin1, and bcl2. (B) Expression of apoptosisrelated (more ...)

Expression of apoptosis-related proteins To investigate the presence of apoptosis, further sections from the same samples were incubated with cleaved caspase 3 and full-length and cleaved lamin A/C antibodies. All of the imatinib-treated and untreated cases were positive for full-length lamin A/C with nuclear reactivity, whereas no immunostaining was observed for cleaved/activated caspase 3 or cleaved lamin A/C (Figure 4B). The results of the IHC analysis are shown in Table 2.

The finding of autophagy-related protein expression in all of the treated and untreated samples in the absence of apoptosis-related protein expression supports the idea that autophagy is part of the molecular profile of GISTs.

Our morphological, biochemical, and immunophenotypical results provide a strong indirect evidence that imatinib-induced tumoral regression in patients with posttreatment surgically resected GISTs is sustained by autophagy rather than apoptosis.

It has recently been reported that, in addition to inhibiting RTK phosphorylation and inducing cell cycle arrest, imatinib activates autophagy [28]. Autophagy is a dynamic rearrangement of the subcellular membranes that leads to autophagic bodies (corresponding to vacuoles at histological level), which may contribute to tumoral death by means of self-cannibalization, although, unlike apoptosis (which invariably leads to cell death), it may also contribute to cell survival [9]. Mainly investigated on cell lines and by means of electron microscopy [29], the only assay able to demonstrate autophagosomes, autophagy is regulated by a tumor-suppressor mechanism in which the major players are the activating beclin1/PI3KIII complex, the suppressing beclin1/bcl2 complex [8], and the presence of LC3-II strictly bound to autophagosomes. In line with this, we found that most of the samples in our high, moderate, and low histological response groups expressed beclin1/PI3KIII, beclin1/bcl2, and LC3-II in the absence of any morphological, IHC, or biochemical marker of apoptosis.

There was a close correlation between the expression of autophagic gene products and the residual tumoral cellularity, a finding that is underlined by the immunophenotypical profile (Figure 1) and in keeping with the assumption that autophagy represents a single cell's adaptation to starvation [30]. There was in fact a trend toward a threshold decrease in the autophagy pathway in the a and b samples, although b samples showed a definite relative excess of autophagic markers, such as excess beclin1 in beclin1/bcl2 complexes in the presence of beclin1/PI3KIII complexes. This is in line with the idea that our morphological and biochemical findings may represent the end of a phenomenon in which the rapid nonphysiological up-regulation of autophagy triggered by imatinib-induced receptor starvation is in part extinguished.

As PI3KIII is responsible for initiating autophagy by forming complexes with beclin1 and LC3-II expression corresponds to autophagosome formation, the absence of LC3-II protein in two b samples is not unexpected: it is conceivable that most of autophagosomes in high responders (b) have already been digested by lysosomal enzymes and that, in moderate responders (c), autophagy is still at such an early stage that tumoral cells can retain their vitality.

The low bcl2 expression in beclin1/bcl2 complexes in the posttreatment regressing tumors is in keeping with the idea that the down-regulation of bcl2 induces autophagic cell death [31] and does not conflict with the view that GISTs with high bcl2 levels aremore susceptible to a KIT inhibitor such as imatinib [27]. However, as approximately 70% of GISTs express bcl2, and our series did not include any matched pretreatment and posttreatment samples, we can only say that bcl2 contributes to regulating the process of autophagy and that the bcl2 threshold in b and c samples is consistent with cell death.

The fact that the autophagy gene profile in the low responding group did not substantially differ from that in the untreated cases strongly suggests that autophagy is not acquired, but part of the molecular profile of GISTs, and that imatinib-treated GIST cells trigger the same machinery as that used to control cell growth. This assumption, which is also suggested by the marked vacuolization reported to be a notable histological feature in GIST [32,33], is in line with the recently underlined close relationship between the type of cell survival/death used by each tumor type, and its response to drug treatment [34].

Remarkably, we did not observe any morphological changes corresponding to apoptosis, a finding that was corroborated by the negative results of the biochemical and IHC experiments using caspase 3, caspase 7, and lamin A/C antibodies. This absence of apoptosis only apparently conflicts with in vitro experiments because the cell model does not work under nutrient starvation conditions [1,2] and cell lines are treated with imatinib for only a short time. Furthermore, fingerprints of autophagy (but not apoptosis) were also found in the GIST specimens taken from two untreated patients. In this context, it is worth noting that although it inhibits KIT activation, imatinib may fail to induce apoptosis in GIST882 cell lines [35,36], thus suggesting that its antiproliferative effect may be mediated by cell cycle arrest rather than apoptosis. This is in keeping with clinical observations in which the rapid switching off of tumoral TK activity is not paralleled by a decrease in tumor size: also known as “tumor dormancy,” this fits well with the metabolic characteristics of autophagy.

Given the pleiotropic role of autophagy, and the evidence that it and GISTs share a number of RTK pathways, our findings suggest various considerations. First, as autophagy seems to exist in both untreated and treated GISTs, it is conceivable that it may contribute to controlling tumoral cell growth in untreated cases, while being the result of a specific stress response to the imatinib-induced starvation of RTK activation in treated GISTs.

Second, the pleiotropism of autophagy may explain the wide spectrum of posttreatment changes which, depending on the threshold of nutrient starvation and the recognized clonal composition of GISTs, may range from areas showing a complete loss of cellularity or very few scattered and regressing residual cells, to clusters of still viable multi-vacuolated tumoral cells and clusters of tumoral cells lacking posttreatment features in nonresponsive areas.

Third, the pleiotropic role of autophagy may explain the prompt tumoral regrowth that occurs when the treatment is discontinued: removing the starvation block (imatinib) may allow the still viable multivacuolated tumoral cells to resume functioning.

Finally, in relation to the activation of RTK pathways, the evidence that rapamycin may overcome resistance to imatinib once again strongly supports the involvement of autophagy in GISTs: mTOR is a downstream intermediate of the PI3KI/AKT pathway, and rapamycin can eliminate its inhibitory effects and thus favor autophagy [37]. In this light, it is interesting to note that we have recently investigated RTK downstream signaling in 15 treated and 5 untreated GISTs (partially overlapping the present series; manuscript submitted) and found MAPK cascade activation, which is known to promote autophagy [38,39].

In conclusion, although the unavailability of paraformaldehydeglutaraldehyde- fixed material made it impossible to use electron microscopic assay and a direct demonstration of autophagy could not be provided, through the analysis of surgical specimens, we provided a substantial number of descriptive observational data which, together with clinical and imaging evidence, strongly suggest that autophagy rather than apoptosis seems to be activated in GISTs in response to imatinib treatment. Further investigations, including the analysis of imatinib-treated normal tissue, and focusing on mechanistic insights into the role of autophagy in imatinib response through in vitro assays are needed to confirm these preliminary descriptive data.

We have no conflict of interest to declare.