Effects of erlotinib therapy on [¹¹C]erlotinib uptake in EGFR mutated, advanced NSCLC

Patients with histologically proven EGFRmut NSCLC who were either planned to initiate erlotinib therapy or to stop erlotinib therapy due to disease progression while on erlotinib were asked to participate. Key inclusion criteria were age above 18 years, life expectancy of at least 12 weeks, at least one tumor lesion with a diameter of at least 1.5 cm in the chest region as measured by CT, Karnofsky index >60 %, and a written informed consent. Exclusion criteria were claustrophobia, pregnancy or lactating patients, metal implants in the thorax (e.g., pacemakers) that could interfere with PET/CT imaging, and concurrent treatment with experimental drugs. The study was approved by the Medical Ethics Review Committee of the VU University Medical Center. All patients provided written informed consent prior to inclusion.

The aim was to include ten patients with EGFRmut NSCLC, who underwent two PET scan sessions. Patients who were on erlotinib therapy (E+) stopped therapy on the day of their first scan. Patients who were erlotinib naïve (E−) started therapy, immediately following the scanning procedure on day 1. For all patients, a second PET scan session was performed after 7 to 14 days. All PET scans were planned to start at the same time of the day, i.e., at 1:00 p.m. Patients on erlotinib therapy were asked to take their last medication, i.e., erlotinib 150 mg, at 8:00 a.m.

One cannula was inserted into the radial artery for arterial blood sampling and another one into a contralateral arm vein for tracer injection and venous blood sampling. Scans were performed on a Gemini TF-64 PET/CT scanner (Philips Medical Systems, Best, the Netherlands), which is a high performance, time-of-flight (TOF), fully three-dimensional PET scanner combined with a 64-slice Brilliance CT scanner [9]. First, 370-MBq [¹⁵O]H₂O was injected intravenously, simultaneously starting a 10-min emission scan. Next, a low-dose CT scan (30 mAs, without contrast) was performed for attenuation correction. Subsequently, 349 ± 46 MBq of [¹¹C]erlotinib (synthesized as previously described and corresponding to a non-pharmacological dose of approximately 16.2 μg “cold” erlotinib with ≥18.5 GBq/μmol specific activity) was injected intravenously, simultaneously starting a 60-min emission scan [1]. [¹⁵O]H₂O and [¹¹C]erlotinib emission scans were acquired in list-mode and reconstructed into 26 frames with progressive increase in frame duration (1 × 10, 8 × 5, 4 × 10, 2 × 15, 3 × 20, 2 × 30, and 6 × 60 s) and 36 frames (1 × 10, 8 × 5, 4 × 10, 2 × 15, 3 × 20, 2 × 30, 6 × 60, 4 × 150, 4 × 300, and 2 × 600 s), respectively. All appropriate corrections were applied for dead time, decay, random, scatter, and attenuation. Reconstruction of PET data was performed using the 3D row-action maximum-likelihood algorithm (RAMLA) with CT-based attenuation correction. The final voxel size was 4 × 4 × 4 mm³ and the spatial resolution 5–7 mm full-width at half-maximum. No corrections for patient motion were applied.

Arterial and venous samples (7 mL) were taken at six time points (i.e., at 5, 10, 20, 30, 40, and 60 min) after injection of [¹¹C]erlotinib. For both arterial and venous samples, plasma polar [¹¹C]erlotinib metabolites and whole blood and plasma radioactivity concentrations were measured, as described previously [1].

For each patient, the primary tumor was identified on the low-dose CT scan, and the tumor contours were delineated visually at the margins of the tumor on all planes where the primary tumor was visible, to generate a three-dimensional tumor volume of interest (VOI). Large blood vessels and the liver were avoided as much as possible. We did not delineate tumors on PET, as [¹¹C]erlotinib PET uptake depends strongly on tumor characteristics (e.g., EGFR mutation). CT-based contour delineation was performed using an in-house software, developed within the interactive data language (IDL Virtual Machine 6.2, RSI Inc., Boulder, CO, USA) environment. Then, tumor VOIs were projected onto the dynamic [¹¹C]erlotinib PET scan to generate tumor [¹¹C]erlotinib time activity curves (TACs). In addition, metabolite-corrected image-derived plasma input functions (IDIFs) were derived from VOIs drawn on ten subsequent slices within the descending aorta (approximately 7 mL). Then, this arterial whole blood TAC was calibrated by the whole blood activity concentrations measured from the six manually drawn arterial blood samples. Next, the data was multiplied by the multi-exponential function that best fitted the plasma-to-whole blood ratios, again derived from the manual samples, to generate a plasma TAC. Then, the plasma TAC was corrected for metabolites using a sigmoid function derived from the best fit to the measured parent fractions of the arterial samples. Finally, a correction for delay was applied; this metabolite-corrected plasma TAC was used as IDIF [10‐12].

A distinction was made in processing the kinetic data from patients on and off therapy. Previously, in patients off therapy, the optimal model for tumor [¹¹C]erlotinib pharmacokinetics was found to be the reversible two-tissue model (2T4k) [1]. In patients off therapy, all tumor TACs were analyzed using this model. It was unknown, however, whether the same model was also valid for patients on erlotinib therapy. Therefore, in the latter patients, first, the optimal model was identified by fitting tumor [¹¹C]erlotinib TACs to three conventional compartment models (i.e., single tissue, irreversible two-tissue, and reversible two-tissue) [10]. Subsequently, the optimal model was chosen on the basis of the Akaike information criterion [13]. After establishing the optimal model for patients on erlotinib therapy, all tumor TACs were analyzed using the corresponding preferred model. Pharmacokinetic analysis and modeling of tumor TACs and IDIF was performed using in-house software, developed within MATLAB (MathWorks, Inc).

In order to understand the effects of metabolism on V _T values under erlotinib therapy, the change in tumor V _T was correlated with the level of metabolism. As the level of metabolism, the parent fraction measured at 60 min post injection was used.

For [¹⁵O]H₂O, same VOIs were drawn on the CT scans accompanying the [¹⁵O]H₂O PET scans, and then projected onto the [¹⁵O]H₂O data. All tumor TACs were analyzed with the standard single tissue compartment model (1T2k) for [¹⁵O]H₂O [14], resulting in estimates of tumor blood flow (TBF) as calculated by rate constant of influx (K1) of [¹⁵O]H₂O [15].

Accuracies of several simplified static approaches were evaluated. SUVs, normalized for patient weight and injected dose, were evaluated in the interval 40–50 and 50–60 min. In addition, TBR values were evaluated using both arterial and venous whole blood activity concentrations in the time interval 40–50 and 50–60 min. These intervals were chosen, as unpublished analysis of previous scans showed that TBR using whole blood activity between 40 and 60 min correlated best with V _T [1].

Statistical analysis was performed using SPSS software (SPSS for Windows 20.0, SPSS, Inc., Chicago, USA) and GraphPad (GraphPad Prism version 5.00 for Windows, GraphPad Software, San Diego, CA, USA). Spearman’s correlation coefficient (r _s) and simple linear regression were used for correlations. A two-tailed probability value of P < 0.05 was considered to be significant. Bland-Altman analysis was performed to assess agreement between venous TBR and arterial TBR, between TBR_40–50 and TBR_50–60, and between venous and arterial tracer parent fractions. The Wilcoxon matched-pairs signed-rank test was used to test differences between scans with and without erlotinib therapy regarding V _T values, parent fraction values, whole blood SUVs, and TBF values. This test was also used to assess differences between whole blood SUVs obtained with arterial and venous samples.

Patient characteristics are shown in Table 1. As three patients could only be scanned once, a total of 13 patients were recruited in order to obtain ten patients who were scanned twice using [¹¹C]erlotinib. In nine out of these ten patients, both quantitative kinetic analyses could be obtained. In the remaining patient, this was not possible due to technical problems with blood sampling. In seven out of nine evaluable patients, the first [¹¹C]erlotinib PET scan was without erlotinib therapy. In the remaining two out of nine patients, the first scan was performed while receiving erlotinib therapy. In only five out of nine evaluable patients, both [¹⁵O]H₂O TBF(E−) and TBF(E+) could be derived. In the remaining patients, data from both [¹⁵O]H₂O PET scans could not be performed or analyzed due to technical problems, as indicated in Table 1.

Parent fractions of [¹¹C]erlotinib, as measured in arterial plasma samples, were higher in patients on therapy at all time points (all P values <0.05, see Fig. 1a). Change in tracer metabolism during erlotinib therapy did not correlate with changes in V _T (r _s = 0.33, P = 0.385), as shown in Fig. 1b.

Arterial blood activity, normalized to injected dose and patient weight, was also higher in patients on therapy at all time points (all P values <0.05, Fig. 1c). Detailed results are shown in Additional file 1: Tables S1 and S2.

According to the Akaike information criterion, the reversible two-tissue compartment model (2T4k) was the preferred model in all patients(E+). This was also the case for patients(E−), confirming previous findings [1, 5, 6]. In all nine evaluable patients, tumor [¹¹C]erlotinib V _T(E+) was significantly lower than V _T(E−) with a mean (±SD) intrapatient decrease of 38 ± 13 % (median V _T(E−) = 1.61, range 0.77–3.01; median V _T(E+) = 1.17, range 0.53–1.74; P = 0.004; see Fig. 2a). There was a good correlation between V _T(E+) and V _T(E−) (r _s = 0.82; P = 0.011), as shown in Fig. 2b. See Additional file 1: Table S3 for detailed results.

Tumor [¹⁵O]H₂O perfusion did not change between patients on (N = 8) and off (N = 8) erlotinib therapy (with a mean ± SD TBF of 0.475 ± 0.194 and 0.622 ± 0.397 mL/cm³/min, P = 0.813, respectively, see Fig. 3a). There was no correlation between TBF and [¹¹C]erlotinib V _T in patients(E−) and patients(E+) (r _s = −0.452, P = 0.268 and r _s = −0.167, P = 0.703, respectively, see Fig. 3b). Tumor rate constant of [¹¹C]erlotinib influx, i.e., K1, showed a trend towards positive correlation with TBF(E−) and TBF(E+); however, this was not statistically significant (r _s = 0.714, P = 0.058 and r _s = 0.405, P = 0.327, respectively, see Fig. 3c) (see Additional file 1: Table S4 for detailed results).

SUVs did not correlate with V _T values, both on and off erlotinib therapy (r _s = 0.39, P = 0.260 and r _s = 0.30, P = 0.342, respectively). However, TBR_40–50 values showed good correlation with V _T values, both on and off therapy (r _s = 0.97, P < 0.001 and r _s = 0.96, P < 0.001, respectively), as shown in Fig. 4. TBR_50–60 also showed good correlation (with r _s = 0.92, P < 0.001 and r _s = 0.99, P < 0.001, respectively). The mean (±SD) difference between TBR_40–50 and TBR_50–60 in patients off therapy was 4 ± 7 % and 1 ± 7 % in patients on therapy.

Representative parametric [¹¹C]erlotinib images, using TBR_50–60, of a typical patient off and on erlotinib therapy are shown in Fig. 5.

The present study demonstrated that tumor [¹¹C]erlotinib V _T decreases significantly during erlotinib therapy.

To our best knowledge, this is the first clinical study that investigated the change of radiolabeled EGFR TKI uptake in patients off and on treatment using the same EGFR TKI. In the presence of therapeutic concentrations of erlotinib, tumor [¹¹C]erlotinib uptake decreased. This was presumably caused by occupancy of EGF receptors by abundantly present non-labeled erlotinib, i.e., due to a decrease in available binding sites. Blocking studies in xenograft models provide support for this mechanism. Using [¹¹C]erlotinib, Petrulli et al. showed that NSCLC xenografts with activating EGFRmut (HCC827) in mice had lower tracer uptake when cold erlotinib was given along with the tracer [16]. In addition, Abourbeh et al. showed in mice-bearing HCC827 xenografts more than 50 % reduction in tumor [¹¹C]erlotinib uptake after administration of excess non-labeled erlotinib [17]. Similar results were obtained with other radiolabeled TKI, such as [¹⁸F]afatinib [18] and [¹¹C]PD153035 [19]. The fact that there was consistent decrease in [¹¹C]erlotinib uptake in the present study supports the notion that uptake of [¹¹C]erlotinib is, at least in part, due to specific binding. Furthermore, in the presence of therapeutic concentrations of erlotinib, obtained by taking a fixed oral dose of 150 mg erlotinib daily, there was still residual tumor tracer uptake. Interestingly, from a pharmacokinetic perspective, this may indicate that there may be room for increasing the therapeutic concentration of erlotinib, as at maximal concentration, the specific binding would be absent.

Erlotinib therapy is known to induce metabolizing enzymes, such as CYP1A, CYP3A4, and CYP3A5 [20]. Also, in vitro data suggest that erlotinib stimulates the metabolism of midazolam in human microsomes, suggesting that erlotinib could induce its own metabolism and thus also increase the clearance of [¹¹C]erlotinib [21]. However, this was not observed in the present study. On the contrary, parent fractions at 60 min post injection were significantly higher during erlotinib therapy. Possibly, the presence of abundant non-labeled erlotinib also saturated the metabolizing enzymes, thereby slowing down the metabolism of [¹¹C]erlotinib. Moreover, patients on therapy had higher blood activity concentrations, normalized to injected dose and patient weight. This may also be caused by higher concentrations of circulating parent tracer due to the blocking of receptors and enzymes by high concentrations of non-labeled erlotinib.

Among the nine evaluable patients, two were scanned first under erlotinib therapy and stopped therapy immediately thereafter. In these two patients, the abovementioned findings were also true, i.e., V _T(E+) was lower than V _T(E−) and metabolites(E+) were lower than metabolites(E−). This supports the notion that the presence of non-labeled erlotinib determined these pharmacokinetic changes by the abovementioned mechanism.

High tumor sensitivity to erlotinib could potentially cause a large decrease in V _T. Namely, in the absence of cold erlotinib, EGFR-TKI-sensitive tumors are expected to have high [¹¹C]erlotinib V _T values as compared to resistant tumors [1]. Once cold erlotinib is added, EGF receptors become blocked causing V _T to drop. The results of this study confirmed that the patients with the largest decrease in V _T did have responsive tumors; however, there was no clear association between decrease in V _T and tumor response. To illustrate, three patients (patients 8, 2, and 6) had a large (i.e., approximately 50 %) decrease in V _T. Patient 8 was treated with erlotinib therapy for a few weeks only. Erlotinib was stopped, as he refused to continue therapy due to a pneumonia that he ascribed to erlotinib. He did have some tumor regression with erlotinib during these few weeks, however, not enough to be declared a partial response. Patient 2 had a complete response to erlotinib therapy after 3 months. Patient 6 had a slow disease progression prior to erlotinib scanning; she stopped erlotinib therapy after her first scan but developed a severe flare of her disease within 1 week. Her second scan showed increased tumor volume and increased V _T; this illustrates that her tumor still had significant amount of sensitive clones. These cases demonstrate that high decrease in V _T can occur in sensitive tumors; however, there was no clear association, that is, responders did not exclusively show high decrease, as there were two other cases with partial tumor response to erlotinib therapy who showed moderate decrease in V _T of 29 and 17 %. On the other hand, as a result of erlotinib therapy, changes can occur in the size of the tumor, its concentration of vital tumor cells and possibly its EGFR density. These changes can occur in a period as short as 7 to 14 days after initiation or discontinuation of therapy and may also influence V _T. Therefore, any tumor response to erlotinib therapy may also influence the decrease of V _T during therapy. However, the limited number of patients scanned does not allow for extensive elaboration. Future studies including more patients should investigate the correlation between response and change in uptake.

The decrease in V _T varied between 17 and 58 %. This high level of variability disqualifies V _T(E+) as substitute for V _T(E−). Any quantitative comparison between patients or between different time points in a single patient should be performed using V _T(E−). However, for intrapatient interlesional comparison at a single time point, V _T(E+) may still be considered. Whether tumor TKI sensitivity can be predicted by [¹¹C]erlotinib V _T(E+) remains to be investigated.

Tumor perfusion was not changed by erlotinib treatment. Also, tumor perfusion showed no association with [¹¹C]erlotinib V _T(E−) nor with V _T(E+). However, there was a positive trend between tumor perfusion and the delivery of [¹¹C]erlotinib to the tumor, which is in accordance with the 2T4k model. These findings suggest that the extraction of [¹¹C]erlotinib remains unchanged during erlotinib therapy.

For simplification of future scanning protocols, SUV was not found to be a suitable uptake parameter, as it did not correlate with V _T, both on and off therapy. SUV normalizes on the basis of injected dose and patient weight, which is less accurate as compared to TBR that normalizes on the basis of the blood pool activity itself. For example, during erlotinib therapy, the blood tracer concentrations were higher. Due to this increased tracer availability, the absolute amount of tumor tracer binding may have changed in varying extent. Tumor SUV does not take this variable into account, whereas TBR does. Contrary to SUV, arterial and venous TBR showed an excellent correlation with arterial V _T, supporting the use of both arterial and venous TBR in the time interval of 40 to 60 min post injection in future whole body static scanning protocols.

Interestingly, venous TBR values were lower than the arterial TBR values, especially in patients off therapy. This was due to higher plasma activity in venous samples than in arterial samples. As no difference in metabolism was observed between venous and arterial samples, the higher venous plasma activity values were only caused by a higher venous concentration of parent molecules. The reason for this finding is unclear. Possibly, the interstitial compartment together with EGFR molecules that are highly expressed at the epidermal tissue compartment act as a capacitator, by reversibly binding [¹¹C]erlotinib molecules. So, venous plasma collects not only the unbound [¹¹C]erlotinib molecules coming from the arterio-capillary route but also the [¹¹C]erlotinib molecules being released from the interstitial and peripheral tissue compartments. This can also explain why patients on therapy, who have more EGFR saturation, have less veno-arterial activity difference. Another cause that may be considered is the fact that a single venous cannula was used for tracer injection and blood withdrawal, which implies that venous activity may increase due to the presence of tracer molecules that remained sticking to the cannula wall during injection. However, this mechanism is unlikely as it cannot explain why the veno-arterial difference was higher in off therapy than on therapy. Nevertheless, venous sampling was found to be suitable for interlesional quantitative comparison using TBR, as long as venous values are not interchanged together with arterial values.

This study was limited by the fact that not all uptake values of [¹¹C]erlotinib and [¹⁵O]H₂O and not all arterial and venous sampling values were present or evaluable in all patients; this was due to practical limitations as mentioned in Table 1. The small number of patients did not allow to establish the clinical role of [¹¹C]erlotinib PET during erlotinib therapy. Larger studies are needed to explore the clinical benefits of scanning during therapy, e.g., for evaluating interlesional differences within a single patient, preferably using TBR as validated in the current study.