Introduction

Neuroendocrine tumours (NETs) are rare, but their incidence is rising [1]. They are frequently metastatic at diagnosis and require systemic treatments. Several strategies are available, including targeted therapies as everolimus [2–4], but were never compared to each other. Three large phase III placebo-controlled randomized trials [2–4] recently reported the efficacy of everolimus in patients with advanced pancreatic NETs (RADIANT-3), gastrointestinal (GI)-NETs and pulmonary NETs (RADIANT-2 and RADIANT-4).

Only a subset of patients draws a clear benefit from mTOR inhibition; for instance, the progression free survival (PFS) at 18 months was 34 % under everolimus as compared with 9 % under placebo in pancreatic NETs [4]. However, no predictive factor of response to everolimus has been identified in NETs. In other solid tumours, several possible markers of response to everolimus or other mTOR inhibitors with the same mechanism of action, such as temsirolimus, have been proposed. They include clinical and biological factors, including the occurrence of side effects under everolimus treatment, such as metabolic disorders, lymphopenia or stomatitis, which have been associated with higher efficacy [5–7]. They also include molecular markers, selected on the concept that a high activation of the mTOR pathway would improve tumor response to mTOR inhibitors. Such markers include genetic alterations such as PTEN loss of function [8], activating mutations of PIK3CA [9] and loss-of-function mutations of TSC1 or TSC2 [10]. They also include protein alterations such as loss of PTEN protein expression [8] and presence of high amounts of phosphorylated forms of mTOR and its targets p70S6K and 4EBP1 [11–14].

Our aim was therefore to analyze whether the response to everolimus could be predicted by i) conventional clinical, biological and histological tumour characteristics, ii) occurrence of side effects such as hyperglycemia, hyperlipemia or haematological toxicity, or iii) functional in situ markers of mTOR activity, such as phosphorylated p70S6K, which we selected because of the availability of a robust immunohistochemical test working in the archival formalin-fixed paraffin-embedded tissue material available for this retrospective study, and already validated within our laboratory in several clinical and preclinical studies [ 11 ].

Subjects and methods

Patients

We retrospectively included all patients who i) had histologically proven advanced well differentiated NET, ii) received everolimus for more than 14 days in our institution (ENETs center of excellence), between 2006 and 2013, iii) gave informed consent to clinical research studies, and iv) adhered to complete follow-up in our institution. Patients could be treated with everolimus during a clinical trial (RADIANT-1, −2, −3, or COOPERATE-2) or according to clinical practice, with or without concurrent somatostatin analogs.

The following clinical, biological and histological data were collected: age at diagnosis, gender, MEN1 syndrome, primary tumour location, functional status, Ki67 index, World Health Organization 2010 classification, TNM stage, location and number of metastatic sites, prior antitumour treatments and progressive disease status at beginning of everolimus.

Efficacy assessment

The efficacy of everolimus was assessed from clinical, biological and morphological parameters. When tumours were functioning, clinical response was defined as a > 50 % decrease in their clinical syndrome (daily numbers of flushes and of stools for carcinoid syndrome, number of stools for VIPoma, number of hypoglycemic episodes for insulinoma, disappearance of cutaneous lesions for glucagonoma). Biological response in plasma chromogranin A (CgA) levels was evaluated among patients with baseline CgA above twice the upper limit of normal and was defined as a > 30 % decrease in plasma CgA as compared to pre-treatment value. We analyzed the best tumour response obtained in accordance with Response Evaluation Criteria In Solid Tumors (RECIST) criteria v1.1, stable disease being defined as a tumour stable after at least 2 months of treatment. Patients were assessed at 2–3 month intervals of treatment according to clinical trial protocols or to clinicians’ appreciation, or earlier if clinically indicated, using clinical examination and spiral computed tomography scan of the thorax, abdomen and pelvis and/or magnetic resonance imaging of the liver when appropriate. Patients who stopped everolimus within 2 months for another reason than disease progression were considered unavailable for best tumor response, but were censored for calculation of progression free survival (PFS).

Safety assessment

All grade 3 and 4 adverse events according to the National Cancer Institute’s Common Terminology Criteria (NCI-CTC) for Adverse Events version 4.0 were collected. We reported dyslipidemia and hyperglycemia when they required the introduction or the increase of a medical treatment according to the protocol for patients in clinical trial or according to our current recommendations for other patients [15, 16]. Briefly, Grade 2 or higher hypercholesterolemia (>300 mg/dL or 7.75 mmol/L) or Grade 2 or higher hypertriglyceridemia (>2.5 x ULN) were treated with a statin (HMG-CoA reductase inhibitor), fibrate, or appropriate lipid-lowering medication. We recorded the time to initiation or increase of hypolipidemic treatment, calculated from the start of everolimus. For hyperglycemia, metformin was used in first intent.

In situ functional analysis of mTOR activation

We used one of the main downstream effector of the mTOR pathway, phospho-p70S6 kinase, as a marker of mTOR activation. The apparent expression levels of the phosphorylated protein were evaluated by immunohistochemistry applied to representative formalin-fixed, paraffin-embedded tumor tissue samples. Tissue samples from either primary tumor or metastatic lesions were available in 43 patients. Four-μm-thick sections were prepared according to conventional procedures; after deparaffinization and dehydration, tissue sections were immersed in citrate buffer pH 6 using a water bath at 98 °C for 35 min; endogenous peroxidase activity was blocked by incubation in 3 % hydrogen peroxide in sterile water. Immunostaining was performed by using a polyclonal antibody directed to phospho-p70S6 kinase at 1/200 dilution (Santa Cruz Biotechnology, Dallas, TX). The apparent intensity of expression of p-p70S6K was evaluated by comparison with internal controls, usually intra-tumoral endothelial cells. Expression levels were defined as high when they were higher than in internal controls. They were defined as low when they were comparable to, or lower than, in internal controls. Patients presenting tumours with high p-p70S6K expression levels were compared to patients presenting tumours with normal or low expression levels.

Statistical analysis

Categorical variables were expressed as percentages, and compared by the Chi-square test or with Fisher’s exact test when appropriate. Continuous variables were expressed as median with range. PFS was calculated from initiation of everolimus to the date of disease progression according to RECIST criteria or death from any cause, whichever occurred first. Overall survival (OS) was calculated from initiation of everolimus to the date of death or last follow-up. PFS and OS were estimated by the Kaplan–Meier method and comparisons were performed using the log-rank test. Cox proportional hazard models were developed using relevant clinico-pathological variables to determine the association of each parameter with OS or PFS. Relative risks were expressed as hazard ratios (HRs) with 95 % confidence intervals (CIs). All testing was two-tailed and p < 0.05 was considered to be statistically significant. Statistical analysis was performed with R software version 3.2.2.

Results

Patient and tumour characteristics

Fifty-three patients with advanced NETs treated with everolimus between 2006 and 2013 were included: 68 % of them were treated in clinical trials and 75 % had a pancreatic primary location (Table 1). All patients had metastatic and progressive disease at time of everolimus treatment, which was started 44 (3–237) months after diagnosis, after a median of 2 prior systemic treatments.

Table 1 Patient and tumour characteristics at initiation of everolimus treatment
Full size table

Histological data were obtained from either the primary tumour (n = 34) or from one metastasis (n = 34); in 8 patients, specimens from both primary and metastasis were available. All tumours were well-differentiated. Median Ki67 was 10 % (0–50). In primary tumours, histological grade was G1 in 5 (24 %), G2 in 10 (50 %), and G3 in 6 (26 %, but with a median Ki67 of 29 (22–50)), whereas, in metastases, histological grade was G1 in 6 (26 %), G2 in 13 (56 %), and G3 in 4 (18 %). In patients for whom both the primary and a metastasis were available, histological grades of metastasis were recorded.

Functional in situ study of mTOR activation

The results of phospho-p70S6K immunostaining were interpretable in 42 (20 primary tumours and 22 metastases) out of the 43 specimens studied. The staining was nuclear in all cases; it could be nuclear and cytoplasmic in the cases with the highest expression levels. The apparent expression levels were high in 50 % of cases studied, and low in the other 50 % (Fig. 1a-b). The two groups had similar clinical and histological characteristics (data not shown) except that high p-p70S6K immunostaining was more frequent when patients had two or more metastatic sites (57 % vs 19 %, p = 0.01) and in metastatic specimens (71 % vs 33 %, p = 0.01).

Fig. 1
figure 1

Immunostaining for p-p70S6K: two pancreatic neuroendocrine tumours with high (a: ×20) and low (b: ×20) staining in comparison with endothelial cells (arrow)

Full size image

Response and tolerance to everolimus treatment

Clinical, biological and morphological responses are shown in Table 2. Objective response and stable disease were achieved in 9 % and 68 %, respectively. The median PFS and OS were 8.1 (4.7–11.5) and 27.5 (16.5–38.5) months.

Table 2 Efficacy of everolimus (n = 53)
Full size table

Safety is shown in Table 3. Grade 3–4 toxicities occurred in 25 (47 %) patients. Grade 2 metabolic disorders, which were treated before becoming grade 3, occurred in 21 (40 %) patients: 23 % for hypercholesterolemia and 25 % for diabetes mellitus. The median time between everolimus start and cholesterol-lowering treatment was 3.7 (0.5–14.7) weeks.

Table 3 Safety (n = 53)
Full size table

Prediction of the response to everolimus

Factors associated with progression-free survival in patients treated with everolimus at univariate analysis are presented in Table 4. Occurrence of hypercholesterolemia under everolimus was associated with longer PFS (HR = 0.13, p < 0.001) (Fig. 2a), whereas presence of bone metastasis, number of metastatic sites, high tumour grade and a high tumoral expression of p-p70S6K (HR = 2.51, p = 0.013) (Fig. 2b) were associated with shorter PFS under everolimus at multivariate analysis. Among other biological markers, plasma CgA levels ≥5 ULN, but not CgA ≥ 2 ULN, was associated with a worse PFS (2.03 (1.05–3.95), p = 0.04). A biological response at 3 months and need of treatment for grade 2 diabetes mellitus were not predictive of longer PFS in our study (Table 4). Results after multivariate analysis are similar (data not shown), but with wide ranges of CI due to the low number of patients by groups.

Table 4 Factors associated with progression-free survival in patients treated with everolimus after univariate analysis
Full size table
Fig. 2
figure 2

Progression-free survival according to occurrence of hyperlipidemia (a) and expression of p-p70S6K (b)

Full size image

We also studied the association of these factors with overall survival (data not shown), showing similar results than on PFS: more than one metastatic site, bone metastasis, plasma CgA level ≥ 5 ULN, no surgery of primary tumour, no occurrence of hypercholesterolemia under everolimus and a high tumoral expression of p-p70S6K were associated with shorter overall survival from the beginning of everolimus treatment after univariate analysis.

Discussion

Everolimus is already approved for pancreatic NETs [4]. Data from Radiant-2 and Radiant-4 studies suggest that the drug might be of interest also in intestinal and pulmonary NETs [2, 3]. We here report the outcome of 53 patients treated with everolimus for a metastatic NET, including pancreatic and non-pancreatic tumors. Our population also includes some patients with grade 3 tumors, but well-differentiated, not include in Radiant-2, −3, or −4 [2–4]. The median Ki67 index was 10 %. Efficacy of everolimus (9 % of objective response and 8.1 months of PFS) was similar close to that published in randomized phase III studies, but with worse outcome. Indeed, in Radiant-2, −3, and − 4, response rate was less than 5 % and median PFS was 11–16.4 months [2–4]. Median overall survival was also shorter in our study (27.5 months) than in published phase 3 studies (23.7–44 months). This can be explained by a selection of fit patients in phase 3 trials. For instance, in Radiant-3 study, patients had better performance status (PS) (97 % of PS 0), less prior lines of treatments (previous chemotherapy was reported in 50 % of patients in Radiant-3 vs 74 % in our patients), and no grade 3 tumor (vs 21 % in our study). Therefore, some patients do not benefit from everolimus, at least at the time of treatment initiation. Predictive factors of response to everolimus are clearly needed because other systemic treatments are available, and that adverse events and cost of everolimus are not negligible.

We first aimed at studying the role of conventional clinical, biological or histologic markers to predict the response to everolimus. We found several factors associated with a better PFS under everolimus treatment, such as only one metastatic site or the absence of bone metastases, but these variables have been identified previously as independent prognostic factors [2, 17]; as we have no control group, we cannot conclude that the corresponding patients get a higher benefit from everolimus treatment. Similarly, a high histological grade was associated with shorter PFS under everolimus, but this factor is known to be associated with worse outcome independently of everolimus treatment [2, 18]. The present results are thus in agreement with those of phase III studies vs placebo, where no clinico-biological parameters clearly predicted response to everolimus [2–4].

High levels of phosphorylated p70S6 kinase (p-p70S6K) in tumor samples were proposed to identify patients with glioblastoma and NETs more likely to derive benefit from temsirolimus, another mTOR inhibitor [5, 13]. p-p70S6K is a direct effector of the activated mTOR complex and targets the S6 ribosomal protein to induce protein synthesis. The main location of the protein is nuclear. p-p70S6K can be demonstrated by a robust immunohistochemical test which might be applied to routinely processed and archival tissues, making this test particularly suitable for our retrospective study. We explored whether this mTOR pathway effector could help to predict tumor response under everolimus. In contrast with prior results, our patients with high p-p70S6K have an even worse outcome under everolimus than patients with low p-p70S6K. These results are in keeping with some recent studies suggesting that high expression levels of mTOR pathway proteins are associated with worse overall survival in NETs [19–21]. Therefore, again because we have no control arm, we cannot demonstrate from our study whether high activation of mTOR pathway would be a predictive factor of response to everolimus; its predictive weight may actually be much lower than its poor prognostic value.

The occurrence of metabolic disorders induced by everolimus treatment looks more promising. Metabolic effects have been reported to yield early markers of response under everolimus in other solid cancers [5, 7]. Metabolic disorders are well-known adverse events of everolimus, but are often underestimated in prospective studies (5 % grade 3 and 4 hyperglycemia, no severe dyslipidemia) because grade-2 metabolic disorders required treatment that prevented them from reaching grade 3 [16]. In our study, 40 % of patients developed grade-2 metabolic disorders, 23 % were treated for hypercholesterolemia within 4 weeks of treatment by everolimus and 25 % of patients needed the initiation or the increase of an anti-diabetic treatment. The occurrence of hypercholesterolemia under everolimus is explained by the decrease of the expression of LDL-receptors by mTOR inhibitors, inhibiting cholesterol endocytosis and increasing its serum level [ 22 ]. Therefore, hypercholesterolemia could be a reliable surrogate evidence of everolimus exposure, in the absence, for most of the patients, of direct measurements of drug levels in the serum. We here show that occurrence of hypercholesterolemia was associated with much higher PFS under everolimus. In glioblastoma, Galanis et al. also reported that development of grade-2 or higher hyperlipidemia in the first two treatment cycles of temsirolimus was associated with a higher percentage of radiological response (71 % v 31 %; p = 0.04) [5]. Similar results were reported in renal cell carcinoma treated by temsirolimus [6]. At the 2015 ENETS meeting, Custudios et al. reported on the predictive value of everolimus-induced hyperglycemia in patients with NETs [23]. The risk of confounding predictive versus prognostic factors does not exist here, because we cannot speculate that metabolic disorders could be a good prognostic factor independently of everolimus treatment. Ravaud et al. reported the relationship between everolimus exposure, safety and efficacy [7]. In patients with pancreatic NETs treated by everolimus, the probability of 12-months PFS was higher in patients with average minimum concentration (Cmin) between 10 and 35 ng/mL than in those with an average Cmin between 0 and 10 ng/ML (54 % and 26 %, respectively). In a meta-analysis on more than 900 patients treated by everolimus, they demonstrated that pulmonary, stomatological and metabolic events were associated with increase in everolimus Cmin and efficacy. We did not study serum everolimus concentration in our patients. However, we can hypothesize that stomatological [24] and metabolic disorders, as already reported for lymphopenia [25], could be indirect markers of everolimus exposure.

In conclusion, this retrospective study of 53 patients with NETs treated with everolimus, shows that conventional clinical and pathological markers are not useful to predict response to everolimus. However, occurrence of hypercholesterolemia under treatment may be an early marker of response. Prospective studies are required to confirm these results.