EPR Oximetry of Cetuximab-Treated Head-and-Neck Tumours in a Mouse Model

Head and neck squamous cell carcinoma (HNSCC) tumours are associated with high mortality despite advances in therapy. The monoclonal antibody cetuximab (Erbitux®) has been approved for the treatment of advanced HNSCC. However, only a subset of HNSC patients receiving cetuximab actually responds to treatment, underlining the need for a means to tailor treatments of individual patients. The aim of the present study was to investigate the effect of cetuximab treatment on tumour growth, on tumour partial oxygen pressure as measured by LiPc electron paramagnetic resonance oximetry and on the expression of proteins involved in tumour growth, metabolism and hypoxia. Two HNSCC cell lines, UT-SCC-2 and UT-SCC-14, were used to generate xenografts on female BALB/c (nu/nu) nude mice. Mice with xenografts were given three injections of intraperitoneal cetuximab or phosphate-buffered saline, and the tumour volume was recorded continuously. After treatment the tumour partial oxygen pressure was measured by LiPc electron paramagnetic resonance oximetry and the expression of epidermal growth factor receptor (EGFR), phosphorylated EGFR, Ki-67, MCT1, MCT4, GLUT1, CAIX and HIF-1α were investigated by immunohistochemistry. In xenografts from both cell lines (UT-SCC-2 and UT-SCC-14) cetuximab had effect on the tumour volume but the effect was more pronounced on UT-SCC-14 xenografts. A higher tumour oxygenation was measured in cetuximab-treated tumours from both cell lines compared to untreated controls. Immunocytochemical staining after cetuximab treatment shows a significantly decreased expression of EGFR, pEGFR, Ki67, CAIX and nuclear HIF-1α in UT-SCC-14 tumours compared to untreated controls. MCT1 and GLUT1 were significantly decreased in tumours from both cell lines but more pronounced in UT-SCC-14 tumours. Taken together, our results show that cetuximab treatment decreases the tumour growth and increases the tumour partial oxygen pressure of HNSCC xenografts. Furthermore we found a potential connection between the partial oxygen pressure of the tumours and the expression of proteins involved in tumour growth, metabolism and hypoxia.


Introduction
Several methods have been developed in the past to quantify tumour oxygenation for applications in radiation therapy [1]. Procedures relying on electrochemical and enzymatic reactions have been employed for measuring tumour hypoxia [2,3]. Physical measurements have also been used [4]. Electron paramagnetic resonance (EPR) oximetry has recently been proposed as an alternative method for the determination of oxygen concentration in vivo [5][6][7][8]. EPR oximetry is based on the observation that the EPR linewidths of certain paramagnetic species are broadened in the presence of molecular oxygen, which itself is paramagnetic [9]. The partial pressure of oxygen can be measured non-invasively in vivo by means of the EPR line width of a paramagnetic probe present in the tissue. EPR oximetry is currently evolving towards clinical use both in the form of low frequency in vivo EPR spectroscopy and in vivo imaging (Electron Paramagnetic Resonance Imaging, EPRI). Quantitative measurements and/or images of the oxygen partial pressure in the tissue of interest have been obtained. Initial clinical applications have been reported [7]. Lithium phthalocyanine (LiPc) probes are robust, sensitive and minimally invasive. The EPR spectra in initial studies [8] had, however, a poor signal to noise ratio (SNR). The spectrometer parameters were therefore a trade-off between spectral resolution and sufficient SNR to determine the oxygen concentration. Overmodulation has been employed to improve the SNR, and the true linewidth was obtained by post-treatment of the data using methods described in [10]. An adaptive filter method has also been developed to improve the SNR [11]. These methods require special software or longer measurement times, however. Freely available filter functions [12] and tools in Matlab were applied for the post-processing of the signals in the present work. The obtained peak-to-peak line widths were subsequently refined by parabolic fits to the experimental spectra. The software was applied in an ongoing study [13] to investigate the effect of cetuximab treatment on tumour growth. In this work the original Matlab program [14] has been updated and a version employing signal smoothing with moving average filters has been developed. The performance of the software has been investigated. The spectral resolution obtained with this software, was increased in comparison with the original one.

EPR measurements
Lithium phthalocyanine (LiPc) (Clin-EPR, NH, USA) crystals/aggregates were loaded into 23 G needles and compressed to obtain LiPc oximetry probes as described by Khan et al [15]. Two probes were usually employed in each EPR measurement. EPR measurements were performed using a Bruker Elexsys E540 L band EPR spectrometer equipped with an E540 R36 L band Resonator (36 mm sample access), an E540 GCL Triple axis coil set (gradient field strength up to 40 G/cm) and an EPR 066L-AMC L band Microwave Bridge. Typical EPR spectrometer parameters were: microwave power 36 mW, modulation amplitude 0.1 G, time constant 20.48 ms, sweep time 10 s (512 measurement points), centre field (B0) approximately 384 G, sweep width 3 G. 20 sweeps were added, yielding a total measurement time of approximately 200 s. The two probes were separated by a gradient of 1 G/cm along B0.

EPR analysis
The recorded EPR spectra were imported into Matlab employing the Easyspin toolbox [12] and analysed using in-house developed software. The signal obtained after applying the "rcfilt" or "datasmooth" functions implemented in Easyspin was analysed to obtain the number and positions of the peaks in each spectrum. An automatic procedure was applied using the Matlab function "findpeaks" to detect lines with an amplitude exceeding an adjustable threshold value. EPR line widths were calculated after the peak positions had been refined by parabolic fits to the experimental spectra. The obtained linewidths and amplitudes were used to display theoretical spectra, with Gaussian, Lorentzian or Voigt shapes [16], together with the experimental ones. The line width of the EPR signal was converted to pO 2 (mm Hg) by calibration measurements at different partial oxygen pressures as described in [15]. Filter functions implemented in Easyspin.

b)
An average over 2m+1 = 11 data points was made. c) The signal was filtered with a time constant Tc = 5 ms by an RC-filter.

d)
Post-processing was not applied.
The results in Table 1, obtained from the analysis of data obtained in an in vivo study [13], indicate that the pO 2 values may differ with up to 10% depending on the filter type at fixed settings. A value of Tc = 5 ms was employed to conform with the conditions in [13], while averaging over 11 data points (m=5) was used for the three types (b) of moving average smoothings. The width obtained from the unfiltered signal (d) may be difficult to estimate accurately due to the low SNR. Probe II has a narrow linewidth, which might become broadened by the RC filter as indicated in the Table. This possibility was investigated by applying a number of filter settings, Fig.1. The results indicate that the pO 2 values obtained in initial studies [13] might have been somewhat overestimated, especially at hypoxic conditions with narrow EPR line widths. The "EPR_width_RC" software may still be applied, by using an option in the input/output dialogue to interactively decrease the Tc value. The SNR may become low, however. The line widths obtained with the Easyspin functions "savgol" and "binom" were nearly unaffected by the filter settings (m) over a large interval and had a satisfactory SNR at m = 15. The functions have been implemented in EPR_width_RCM" by a slight revision of the original code. The "flat" option was also included, although not employed due to the unsatisfactory performance observed in Fig.1 and in previous works [12(b)].

Applications
The initially developed software, "EPR_width_RC", has so far been employed in a few studies [8,13]. The pO 2 level was determined after applying an RC-filter, typically with a time constant (Tc) of 5 ms. The results obtained in this work demonstrate that a slight broadening of the line widths may then occur, yielding ca 10 % too high pO 2 levels. Broadening can be avoided by decreasing Tc, but the SNR may become low. Moving average filter functions were implemented in the "EPR_width_RCM" software, in an attempt to avoid line broadening. The results indicated that the options "binom" and "savgol" available in Easyspin [12(b)] could be used over a wide range of filter settings, with averaging over 2m+1 field points. The measured line widths remained constant up to m = 15, allowing determination of the linewidths at a satisfactory SNR value. Fig.1 indicates that even higher m values might be used with the binomial filter. This software has not yet been used in practical applications, however.

Input/Output dialogue using software "EPR_width_RCM"
The input/output dialogue with "EPR_width_RCM", is similar to that of the original program [14] with the difference that the filter type has to be specified. An example is given below.

Performance
The errors in the pO 2 values given in the output are solely attributed to the differences of the line width values, calculated from the distance between the peak marks of each line, and by a fitting procedure. Those errors are usually small. As discussed above, larger errors may occur by unsuitable data processing, e.g. by excessive filtering. Inaccuracies in the calibration of the EPR probes can give rise to errors of the oxygen concentrations that at present are difficult to estimate.

Software
The software used in earlier applications [8,13] is available at http://www.liu.se/simarc/downloads/?l=en. The folder EPR_oximetry.zip contains a manual, the Matlab code, and example data. Software prepared for this work will be available at the same site, containing the following items in the folder EPR_oximetry_RCM.zip: Manual, "EPR_width_RCM.pdf". The analysis of the data is performed interactively by following the instructions on the screen. The results of this work indicate that the moving average filters "binom" and "savgol" implemented in "EPR_width_RCM.m" are well suited for linewidth measurements of narrow EPR signals. As shown in Fig.1 an RC-filter may broaden the EPR lines significantly, while the line width is nearly independent of the settings of the "binom" filter. A satisfactory SNR value was also obtained.