2.1.

Human Subjects and Drugs Administered

Tumor tissue samples were selected from two patients for noninvasive in vivo DRS measurements, and ex vivo EF5 analysis and MLu fluorescence studies. Two patients were diagnosed with recurrent intra-abdominal mesothelioma [Patient 1 (Pt 1)] and a gastrointestinal stromal tumor [Patient 2 (Pt 2)], respectively. DRS measurements were performed on peritoneal metastases. All patients signed a study-specific informed consent and all experimental procedures were approved by the Institutional Review Board of the University of Pennsylvania and the Clinical Trials Scientific Review and Monitoring committee of the University of Pennsylvania Cancer Center. EF5 (Cancer Therapy Evaluation Program, National Cancer Institute) was administered intravenously at a dose of $21\mathrm{mg}∕\mathrm{kg}$ via a peripheral intravenous catheter $24\mathrm{h}$ before surgical debulking. MLu (manufactured by Pharmacyclics, Incorporated, Sunnyvale, California, and supplied by the Cancer Therapy Evaluation Program, National Cancer Institute) was administered intravenously $3\mathrm{h}$ before surgical debulking at a dose of $1\mathrm{mg}∕\mathrm{kg}$ .

Tumor tissue samples for EF5 binding and MLu fluorescence were removed at surgery immediately after in vivo DRS measurements were obtained. Half of each resected tissue sample was labeled, placed in a specimen container, protected from light, and frozen at $-80°\mathrm{C}$ for later MLu fluorescence measurement. The remainder was placed in iced Excell 610 media (JRH Biosciences, Leneka, Kansas) with 15% fetal calf serum and frozen at $-80°\mathrm{C}$ in Tissue Tek OCT (Sakura Finetek USA, Incorporated, Torrance, California) for later EF5 binding assessment.

The tumors were spread over the peritoneum, thus the size of the tumor before excision cannot be described. The excised tissues were $8\times 8.5\times 8.5∕5.5\times 9\times 8\mathrm{mm}$ (Pt 1, site 1, two samples), $32\times 35\times 10\mathrm{mm}$ (Pt 2, site 1, one sample), and $23\times 19\times 6\mathrm{mm}$ (Pt 2, site 2, one sample). Two adjacent tumor samples were excised from patient 1. The optical measurement was performed over the area of these two tumor samples before excision. For patient 1, the results of the optical measurement were compared with averaged results derived from both ex vivo measurements of the excised tumor samples.

2.2.

Animal Tumor Model for Motexafin Lutetium Concentration Validation

Radiation-induced fibrosarcoma (RIF) tumors were propagated on the shoulders of C3H mice (Taconic, Germantown, New York) by the intradermal injection of $3\times {10}^5$ cells³⁶ approximately one week before MLu administration. At the time of the study, tumor size was $\sim 4\text{to}7\mathrm{mm}$ in diameter, with a maximum volume of $\sim 197{\mathrm{mm}}^3$ (tumor $\text{volume}=\text{diameter}\times {\text{width}}^2\times \pi ∕6$ ). Animal fur over the tumor area was removed with a shaver and a depilatory (Nair). MLu was administered via tail vein at 2.5, 5, or $10\mathrm{mg}∕\mathrm{kg}$ in three animals per dose at $\sim 3\mathrm{h}$ before the DRS measurements. During DRS measurements, animals were maintained under anesthesia via inhalation of isoflurane through a nosecone.

2.3.

Diffuse Reflectance Spectroscopy

Two continuous wave (cw), white light diffuse reflectance spectroscopy (DRS) systems were used to measure tissue optical and physiological properties in the spectral range of $\sim 600\text{to}800\mathrm{nm}$ , as described previously.³⁰ Briefly, one system was composed of a $250\text{-}\mathrm{W}$ quartz tungsten halogen lamp, a hand-held surface contact fiber optic probe, a spectrograph (SP-150, Acton Research, Acton, Massachusetts), and a liquid nitrogen cooled charge-coupled device (CCD) camera (LN/CCD-1100-PF/UV, Roper Scientific, Trenton, New Jersey). The other DRS system was similar, except that the CCD was thermoelectrically cooled and integrated with the spectrograph into a single unit (INS-150-252F, Roper Scientific, Trenton, New Jersey). The fiber optic probe consisted of a single source and ten detection fibers (each with a core diameter of $400\mu \mathrm{m}$ ) spaced nonuniformly. The source-detector separations $\rho$ were all less than $1\mathrm{cm}$ ( $\rho =0.6$ , 1.2, 1.8, 2.4, 3, 4, 5, 6, 8, and $10\mathrm{mm}$ ). Our analysis employed a multiwavelength algorithm to simultaneously fit all data in the optimal wavelength range and using optimal source-detector separations. The fits compared data to an analytical solution of the photon diffusion equation (P1 approximation) with semi-infinite boundary conditions.^{37, 38, 39, 40}

The tissue optical properties, i.e., the reduced scattering $\left({\mu }_s^{\prime }\right)$ and absorption $\left({\mu }_a\right)$ coefficients, were assumed to have the form: ${\mu }_s^{\prime }=A{(\lambda ∕{\lambda }_0)}^{-B}$ and ${\mu }_a=\sum c_i{\epsilon }_i\left(\lambda \right)$ , respectively. Here $\lambda$ is the light wavelength in nanometers, ${\lambda }_0$ is chosen to be $1\mathrm{nm}$ so that $A$ has the same units as ${\mu }_s^{\prime }$ (i.e., ${\mathrm{cm}}^{-1}$ ), and $c_i$ and ${\epsilon }_i$ are the concentration and extinction coefficient of the $i$ ’th chromophore, respectively. The primary chromophores in our analysis were oxyhemoglobin $\left(\mathrm{Hb}{\mathrm{O}}_2\right)$ , deoxyhemoglobin (Hb), MLu, and water. The extinction coefficients of $\mathrm{Hb}{\mathrm{O}}_2$ , Hb, and water were obtained from the literature,⁴¹ and the extinction coefficient of MLu was obtained by direct spectroscopic measurement in our laboratory (using a grating spectrometer from Ocean Optics, Dunedin, Florida). A nonlinearly constrained optimization function, FMINCON implemented in MATLAB (The MathWorks, Incorporated, Natick, Massachusetts), was used to globally fit the data. The multiwavelength algorithm extracts tissue scattering parameters ( $A$ and $B$ ) and chromophore concentrations ( $c_{\mathrm{MLu}}$ , $c_{\text{water}}$ , $c_{\mathrm{Hb}{\mathrm{O}}_2}$ , and $c_{\mathrm{Hb}}$ ). MLu concentration $\left(c_{\mathrm{MLu}}\right)$ was reported in ng/mg. Tissue total hemoglobin concentration (THC) and blood oxygen saturation $\left(\mathrm{St}{\mathrm{O}}_2\right)$ were calculated from these quantities (e.g., $\mathrm{THC}=c_{\mathrm{Hb}{\mathrm{O}}_2}+c_{\mathrm{Hb}}$ , $\mathrm{St}{\mathrm{O}}_2=c_{\mathrm{Hb}{\mathrm{O}}_2}∕\mathrm{THC}$ ).

Signals from the shortest source-detector separation distances $(\sim 0.6\mathrm{mm})$ were not used due to sporadic light leakage from the source fiber and to limitations of the photon diffusion model at small source-detector separation distances $(<1\mathrm{mm})$ . Between one and four DRS measurements were taken from each IP patient; more measurements could not be taken due to time constraints in the operating room. The optimal source-detector separations and wavelength ranges for calculation of patient physiological properties (see Materials and Methods in Sec. 2) were determined after the operation.

For each RIF tumor, 7 to 10 DRS measurements were taken over 7 to 10 tumor sites. Furthermore, due to the hemispheric tumor geometry (with minimal diameter $\sim 4\mathrm{mm}$ ), the signals at the larger source-detector separations were noisy and prone to systematic error. To be consistent for all nine animals, only signals from the detection fibers located 1.2 and $1.8\mathrm{mm}$ from the source fiber were used to determine the optical properties and chromophore concentrations in the wavelength range of $600\text{to}800\mathrm{nm}$ . The data acquisition time was $100\mathrm{ms}$ per measurement.

Figures 1a and 1b depict representative reflectance spectra $R_{\text{measured}}$ of an IP patient tumor [Fig. 1a] and a RIF tumor implanted on the mouse [Fig. 1b]. Note that corrections for the systematic spectral features of the source and detection system are accounted for in $R_{\text{measured}}$ ; i.e., after dark-count background subtraction, the raw spectra were divided by the spectrum (i.e., the dark-count subtracted spectrum) taken during a calibration run using an integrating sphere.^{28, 30}

Fig. 1

Representative reflectance spectra $R_{\text{measured}}$ of (a) an IP patient tumor and (b) a RIF tumor. The reflectance data $R_{\text{measured}}$ are calibrated; it is obtained by dividing the dark-corrected reflectance spectra from tissue sample by the dark-corrected reflectance spectra from an integrated sphere. Normalized reflectance spectra $\left(R_{\mathrm{norm}}\right)$ taken from the RIF tumor reflectance spectra in (b) is plotted (——) for two separations $\left(\rho \right)$ (c). The fit is based on analogous normalization of calculated spectra $R_{\text{calculated}}$ (------). See text for more details.

Then a source-detector-separation normalized reflectance $R_{\mathrm{norm}}$ is calculated. $R_{\mathrm{norm}}$ divides spectra taken at separation $\rho$ by that taken at a single separation distance, typically the minimum separation $\left({\rho }_0\right)$ ,³⁰ i.e.,

Fig. 2

The normalized reflectance spectra $\left(R_{\mathrm{norm}}\right)$ plotted (——) at the optimal separations $\left(\rho \right)$ for three tumor tissues: (a) patient 1 (Pt 1, tissue 1), (b) patient 2 (Pt 2, tissue 1), and (c) patient 2 (Pt 2, tissue 2). The fits based on $R_{\text{calculated}}\left(\rho \right)$ using the wavelength range between 600 and $800\mathrm{nm}$ (------), and using the wavelength range between $640∕650$ and $800\mathrm{nm}$ (-∙-∙-∙-) were plotted in the same figure to compare $R_{\mathrm{norm}}$ . The noisy signals at $\lambda \sim 600\mathrm{nm}$ in patient 1 (Pt 1, tissue 1) significantly distorted the fitted results when we used $600\text{to}800\mathrm{nm}$ for reconstruction. The fits were improved when we used $640∕650\text{to}800\mathrm{nm}$ for reconstruction (Pt 1, tissue 1 and Pt 2, tissue 2). The fitted results in patient 2 (tissue 1) were insensitive to the wavelength range used for reconstruction.

2.4.

2-(2-Nitroimida-Zol-1[H]-yl)-N-(2,2,3,3,3-Pentafluoropropyl)Acetamide Binding

Microscopic images of EF5 binding were obtained for each tissue section using methods previously described.⁴² In brief, $10\text{-}\mu \mathrm{m}$ frozen tumor sections were fixed in 4% paraformaldehyde, blocked, and stained with EF5 monoclonal antibody (ELK3-51) conjugated to Cy3 dye (regular stain). Multiple regions of the specimen, separated by at least $0.5\mathrm{mm}$ , were examined. In addition to the standard staining protocol, two tissue sections were stained without antibody to assess endogenous tissue fluorescence. A representative section from each patient was also stained with the standard anti-EF5 antibody mixed with an excess of authentic EF5 drug $\left(0.5\mathrm{mM}\right)$ . This control is termed “competed stain.” The excess drug binds to all specific sites on the antibody and prevents specific binding to tissue adducts. Accordingly, sections that received competed stain are used as an indication of nonspecific antibody binding. The competed stain values were subtracted from the absolute fluorescence values for the regular stain to obtain a final evaluation of EF5 binding. Additionally, in vitro studies were performed to determine the maximum binding level of each tissue.^{14, 15} This value is referred to as “cube reference binding” (CRB). The endpoint value used for EF5 binding as a surrogate for hypoxia is percent reference binding, which is the ratio of the in situ binding to the CRB, multiplied by 100. The final reported value is corrected for drug exposure,⁴³ camera exposure, and tissue thickness.¹⁷ This final value is percent maximum binding and is reported on a scale of 0 to 100. 1% of maximum binding represents physiologic conditions, which we define as $>10\%$ oxygen; 3% of maximum binding represents modestly hypoxic conditions, which we define as approximately 2.5% oxygen; 10% of maximum binding represents moderately hypoxic conditions, which we define as 0.5% oxygen; 30% of maximum binding represents severe hypoxia, which we define as 0.1% oxygen; and 100% of maximum binding represents anoxia. Finally, a single EF5 value was reported for each tissue section by averaging the pixels of the EF5 binding microscope image over its entire field of view of approximately $2.1\times 1.7\mathrm{mm}$ . This number was used for comparison to DRS-determined blood oxygenation.

2.5.

Ex Vivo Motexafin Lutetium Fluorescence

MLu was extracted from frozen samples of human or murine tissues using a procedure based on a previous report.⁴⁴ Tissues were thawed to room temperature, weighed, and divided into samples of $\sim 50\mathrm{mg}$ (macro) or $\sim 10\mathrm{mg}$ (micro). Samples were homogenized (Polytron 1200) in $1000\text{-}\mu \mathrm{l}$ (macro) or $400\text{-}\mu \mathrm{l}$ (micro) phosphate buffer ( $24\mathrm{mM}$ , pH 7.5) and mixed with chloroform and methanol, each at a volume equivalent to the phosphate buffer. Samples were centrifuged ( $3500\mathrm{rpm}$ , $15\min$ ), the organic layer was collected, and $600\mu \mathrm{l}$ (macro) or $200\mu \mathrm{l}$ (micro) was transferred to a cuvette. Fluorescence was measured by a spectrofluorometer (FluoroMax-3, Jobin Yvon, Incorporated, Edison, New Jersey) over an emission scan range from $650\text{to}850\mathrm{nm}$ . The excitation wavelength was ${\lambda }_{\mathrm{ex}}=474\mathrm{nm}$ and the peak MLu emission was at ${\lambda }_{\mathrm{em}}$ of $740\mathrm{nm}$ . MLu concentration in the tissue was calculated based on the increase in fluorescence resulting from the addition of a known amount of MLu to each sample after its initial reading. Data are presented as ng of MLu per mg of tissue.

2.6.

Clinical Data Analysis: Selection Criteria for Fitting Optimization

In this pilot study, we selected optimal source-detector separations based on the point at which fitting errors (see definition later) increased by a factor of 2. Thus we chose $\rho =2.4$ , 3, 4, and $5\mathrm{mm}$ for patient 1 (tissue 1) and $\rho =1.2$ , 2.4, and $3\mathrm{mm}$ for patient 2 (tissues 1 and 2).

Our choice of wavelength range was slightly more complex. In our previous publications,^{28, 29, 30} we empirically determined that the wavelength range from $\sim 600\text{to}800\mathrm{nm}$ was optimal for determination of tissue optical and physiological properties. However, in the current clinical study we experienced difficulty determining tissue oxygenation using this wavelength range. These difficulties are possibly due to poor signal-to-noise ratio, to limitations of the diffusion model in regions of high tissue absorption, to tissue surface homogeneity, or to uncontrolled variations in the operating room (OR). Figure 2 depicts the normalized reflectance $\left(R_{\mathrm{norm}}\right)$ in the spectral range between 600 and $800\mathrm{nm}$ . Also shown are the fits based on $R_{\text{calculated}}(\rho ,\lambda )$ using the wavelength range between 600 and $800\mathrm{nm}$ (dashed line), and using the wavelength range between $640∕650\mathrm{nm}$ (explained next) and $800\mathrm{nm}$ (dash dot line).

From Fig. 2 we see that the chosen wavelength range affects the fitting results. In patient 1 (Pt 1, tissue 1), the noisy signals at $\lambda \sim 600\mathrm{nm}$ significantly compromise the fitting (dash line). In patient 2, however, the fitted results using two different wavelength ranges ( $600\text{to}800\mathrm{nm}$ versus $650\text{to}800\mathrm{nm}$ ) showed no difference for bulk tissue 1 and only a small difference at $\sim 650\mathrm{nm}$ for bulk tissue 2.

To select the optimal wavelength range, we examined extracted tissue oxygenation ( $\mathrm{St}{\mathrm{O}}_2$ in percent), MLu concentration ( $c_{\mathrm{MLu}}$ in ng/mg), and fitting error ( $\xi$ in percent) using different fitting wavelength ranges from ${\lambda }_{\mathrm{lb}}\text{to}{\lambda }_{\mathrm{ub}}=800\mathrm{nm}$ , where ${\lambda }_{\mathrm{lb}}$ varied from $610\text{to}690\mathrm{nm}$ (lb is lower bound and $\mathrm{ub}=\text{upper}$ bound). In our previous studies,³⁰ we have defined a fitting error as $\xi =\sqrt{{\chi }^2}∕N$ , where

Fig. 3

The reconstructed $\mathrm{St}{\mathrm{O}}_2(\%)$ , $c_{\mathrm{MLu}}$ , and fitting error $\xi (\%)$ are plotted versus ${\lambda }_{\mathrm{lb}}$ corresponding to the DRS measurements [in Figs. 2a, 2b, 2c], in patients 1 (tissue 1) and 2 (tissues 1 and 2). The reconstruction was preformed using different fitting wavelength ranges from ${\lambda }_{\mathrm{lb}}$ to ${\lambda }_{\mathrm{ub}}$ , where ${\lambda }_{\mathrm{lb}}$ varied from $610\text{to}690\mathrm{nm}$ and ${\lambda }_{\mathrm{ub}}=800\mathrm{nm}$ . $\xi$ reached a minimum, then stabilized at ${\lambda }_{\mathrm{lb}}\sim 650\mathrm{nm}$ in patient 1 (tissue 1) and patient 2 (tissue 1). $\mathrm{St}{\mathrm{O}}_2$ did not vary much while ${\lambda }_{\mathrm{lb}}$ was varied in patient 1, but stabilized at ${\lambda }_{\mathrm{lb}}\sim 650\mathrm{nm}$ in patient 2 (tissue 2). In patient 2 (tissue 1), $\mathrm{St}{\mathrm{O}}_2$ and $\xi$ are insensitive to the change of ${\lambda }_{\mathrm{lb}}$ .

2.7.

Finite Element Method

To explore the role of tumor shape in affecting derived optical properties, we developed a finite element solver. The in-house finite element solver was written in MatLab to numerically solve the photon diffusion equation.⁴⁵ An isotropic point source of light was assumed to be located a distance $1∕{\mu }_s^{\prime }$ below the source at the air-tissue interface. Robin boundary conditions were used on all boundaries, accounting for the refractive index mismatch at diffuse/nondiffuse interfaces. We used software called GiD (see Ref. 46) to generate an unstructured tetrahedral mesh with varying element size. The element size, defined as the average length of six sides of tetrahedron, was set to be $\sim 0.1\mathrm{mm}$ near source and detector positions, and was gradually increased to $\sim 1\mathrm{mm}$ for the rest of the tumor volume.

The analytical solution of the photon diffusion equation for cw reflectance from a semi-infinite turbid medium with the extrapolated boundary conditions (EBC) has been well described and has been shown to agree with the Monte Carlo simulations in previous studies.³⁸ Here, we validate our FEM by comparing FEM generated reflectance from a $2.4\times 2.4\times 1\text{-}{\mathrm{cm}}^3$ turbid slab with analytical semi-infinite boundary condition solutions [Ref. 30, Eq. (3)] at multiple source-detector separations used in our clinical data analysis ( $\rho =0.6$ , 1.2, 1.8, 2.4, 3, 4, and $5\mathrm{mm}$ ). The slab was big enough to simulate a semi-infinite geometry and contained a scattering medium with the following predetermined optical properties: $A=1500.2{\mathrm{cm}}^{-1}$ , $B=0.8278$ , $c_{\mathrm{Hb}{\mathrm{O}}_2}=33.20\mu \mathrm{M}$ , $c_{\mathrm{Hb}}=67.77\mu \mathrm{M}$ , $c_{\mathrm{MLu}}=3.41\mu \mathrm{M}$ (or ng/mg considering the molecular weight of MLu is approximately 1000), and 90% water. The mesh we used in the FEM contained 11,804 nodes and 60,770 elements. We compared the FEM-simulated and analytically calculated reflectance at four wavelengths—620, 650, 700, and $800\mathrm{nm}$ —as an example. The corresponding optical properties, calculated based on ${\mu }_s^{\prime }\left(\lambda \right)=A{(\lambda ∕{\lambda }_0)}^{-B}$ and ${\mu }_a\left(\lambda \right)=\sum c_i{\epsilon }_i\left(\lambda \right)$ , were ${\mu }_s^{\prime }=7.32$ , 7.04, 6.62, and $5.93{\mathrm{cm}}^{-1}$ and ${\mu }_a=0.488$ , 0.288, 0.186, and $0.098{\mathrm{cm}}^{-1}$ at $\lambda =620$ , 650, 700, and $800\mathrm{nm}$ , respectively.

The reflectance generated from FEM simulation closely matched the analytical solution with extrapolated boundary conditions for all the source-detector separation distances except the smallest $(\rho =0.6\mathrm{mm})$ , where FEM simulations generated a higher reflectance. The deviation between two models was $\sim 10\%$ at $\rho =1.2\mathrm{mm}$ , but became larger ( $\sim 18$ to 28%) at $\rho =0.6\mathrm{mm}$ . Numerical error was negligible because we used a fine mesh. Different boundary conditions used in the analytical solutions and the FEM model may also induce small errors. The reflectance signal generated by FEM at $\rho ⩽1.2\mathrm{mm}$ was consistently higher than that of the analytical solution at all of four wavelengths, leading us to assume the systematic error between FEM and analytical solution is constant and to introduce a correction factor to account for the inherent discrepancy between two models in the shape simulations discussed next.

2.8.

Evaluation of Radiation-Induced Fibrosarcoma Tumor Boundary Effect Using Finite Element Method

A mesh as shown in Fig. 4 was generated for FEM simulation data from a hemispheric-like RIF tumor. The computational volume consisted of a rectangular-shaped base with a hemisphere of $4\text{-}\mathrm{mm}$ radius. The top of the hemisphere was truncated to account for the deformation due to the optical probe. The truncated circular plane had a diameter of $1.8\mathrm{mm}$ to accommodate a single source and two detector fibers at separations of 1.2 and $1.8\mathrm{mm}$ that were used in our MLu measurements of RIF tumors.

Fig. 4

The mesh, consisting of a hemisphere of $4\text{-}\mathrm{mm}$ radius and a rectangular base, was used to model the geometry of a radiation-induced fibrosarcoma (RIF) tumor for finite element method (FEM) simulations. The top of the hemisphere was truncated, leaving a radius of $0.9\mathrm{mm}$ to accommodate a source and two detector fibers at separations of 1.2 and $1.8\mathrm{mm}$ . Coordinate dimensions in the figure are in centimeters.

Because of the systematic error between FEM and analytical solution mentioned before, the correction factor $F(\rho ,\lambda )$ , derived from the semi-infinite normalized reflectance ratio, was used for evaluation of the RIF tumor boundary shape effect. We first generated reflectance data using the hemispheric mesh and the FEM code. Then we calculated $R{\left(\lambda \right)}_{\mathrm{norm}}$ , from the data and corrected it by multiplying $F\left(\lambda \right)$ and $R{\left(\lambda \right)}_{\mathrm{norm}}$ . We fit to this “simulation data” using the analytical semi-infinite model, and thereby extracted $A$ , $B$ , and $c_i$ of the underlying medium.

3.1.

Diffuse Reflectance Spectroscopy Measurement of Motexafin Lutetium Concentration in Radiation-Induced Fibrosarcoma Murine Tumors

Diffuse reflectance spectroscopy (DRS) of photosensitizer concentration has been validated in tissue phantoms.³⁰ To evaluate in vivo DRS measurements of photosensitizer concentration in the context of the present study, we performed DRS measurements of MLu concentration in nine radiation-induced fibrosarcoma (RIF) murine tumors.

Immediately after DRS measurements, the RIF tumors were excised, frozen, and stored for subsequent ex vivo spectrofluorometric assay. Figure 5 shows the MLu concentration (ng/mg) measured in vivo by DRS versus ex vivo by spectro-fluorometric assay in the same tumors. Due to their small size, each tumor was run as a single sample by ex vivo spectro-fluorometric assay, thus no error bars are presented for these data. The error bars for the DRS data points represent the standard error (STE) of 7 to 10 measurements taken randomly over various sites on the tumor of each animal.

Fig. 5

RIF tumor MLu concentration ( $c_{\mathrm{MLu}}$ in ng/mg) determined by in vivo DRS is plotted versus MLu concentration ( $c_{\mathrm{MLu}}$ in ng/mg) measured by ex vivo spectrofluorometric assay $(n=9)$ . The correlation coefficient $\left(r^2\right)$ and the slope of the data were 0.87 and 1.10, respectively. The standard error in in vivo $c_{\mathrm{MLu}}$ ranged from $0.15\text{to}0.79\mathrm{ng}∕\mathrm{mg}$ . There is no error bar plotted for ex vivo $c_{\mathrm{MLu}}$ , because only one tissue sample was analyzed for each animal.

Excellent agreement was found between the DRS measurements and those obtained by the spectrofluorometric assay. The correlation coefficient $\left(r^2\right)$ and the slope of the in vivo versus ex vivo data were derived by linear regression and were 0.87 and 1.10, respectively. In all nine animals, MLu uptake by the RIF tumor was shown to be approximately half of the MLu dosage administered. The administered MLu doses were 2.5, 5, and $10\mathrm{mg}∕\mathrm{kg}$ . The $\text{mean}\pm \mathrm{STE}$ of the ex vivo spectrofluorometric assays were $1.14\pm 0.17$ , $2.65\pm 0.21$ , and $4.79\pm 0.20\mathrm{ng}∕\mathrm{mg}$ , respectively.

3.2.

Ex Vivo Measured 2-(2-Nitroimida-Zol-1[H]-yl)-N-(2,2,3,3,3-Pentafluoropropyl)Acetamide Binding and Motexafin Lutetium Concentration

Table 1 summarizes the results of the pixel-averaged EF5 binding and MLu concentration determined by ex vivo assays of three intraperitoneal carcinoma samples from two patients. In patient 1, DRS measurements were performed at a single tumor site with two continuous captures (dual-measurement) before tumor excision. In patient 2, DRS measurements were performed at two different sites (tissue samples 1 and 2). At each site, four randomly selected adjacent locations were measured by DRS. After DRS measurements, each tumor was excised and divided equally for analysis of EF5 binding and MLu concentration.

The EF5 binding in six of the eight tissue sections corresponded to the physiologic range ( $\sim 1\%$ binding, corresponding to $\mathrm{p}{\mathrm{O}}_2\sim 76\mathrm{mmHg}$ ), and the modest hypoxia range ( $\sim 3\%$ binding, corresponding to $\mathrm{p}{\mathrm{O}}_2\sim 19\mathrm{mmHg}$ ), as previously defined.^{13, 47} Two out of the eight sections exhibited moderate EF5 binding, $\sim 4$ to 6%, which falls on the border of the modest and moderate hypoxia ranges ( $\sim 10\%$ binding, corresponding to $\mathrm{p}{\mathrm{O}}_2\sim 3.8\mathrm{mmHg}$ ). In patient 2, two separate tissue sections obtained from either sample 1 or 2 had almost identical EF5 binding (i.e., 4.60% versus 4.66% in sample 1, 1.55 versus 1.45% in sample 2); both of these regions had mild hypoxia. Four sections of the same sample from patient 1 showed a small degree of variation, i.e., from 0.09 to 2.35% EF5 binding; all of these regions were oxic. Regarding the EF5 heterogeneity, our patient 2 microscope images of EF5 binding showed a few small cell clusters ( $50\mu \mathrm{m}$ in diameter) that were very mildly hypoxic. The patient 1 samples were oxic, and no patterns could be described because of minimal binding.

MLu concentration, measured by spectroscopic assays, ranged from an average of $0.595\mathrm{ng}∕\mathrm{mg}$ in bulk tissue 1 of patient $2\text{to}0.882\mathrm{ng}∕\mathrm{mg}$ in bulk tissue 2 of patient 2. Generally, uptake was extremely consistent among samples and patients. The lowest and highest values of drug uptake were found in two samples taken from patient 1 (i.e., $0.24\mathrm{ng}∕\mathrm{mg}$ versus $1.299\mathrm{ng}∕\mathrm{mg}$ ).

3.3.

Diffuse Reflectance Spectroscopy Measurements in Intraperitoneal Carcinoma Patients

DRS data from patients 1 and 2 were analyzed to determine oxygenation and MLu concentration using our selection criteria described in Materials and Methods in Sec. 2. The results are presented in Table 1. In patient 1, two continuous DRS measurements were performed at the same site (i.e., the probe was placed once on the tissue surface and DRS data ware taken at two different time points). The so-determined oxygenation and MLu concentration were similar (i.e., $\mathrm{St}{\mathrm{O}}_2=88.98$ versus 88.88%, $c_{\mathrm{MLu}}=1.76$ versus $2.37\mathrm{ng}∕\mathrm{mg}$ ). In patient 2, we performed four DRS measurement at four random locations within the “bulk” tissue sample (i.e., four tissue contacts of the fiber optic probe). One measurement was made at each location. Substantial variation in tissue oxygenation was observed in this case. $\mathrm{St}{\mathrm{O}}_2$ ranged from 0 to 54.76% in sample 1 and from 39.83 to 73.40% in sample 2. MLu concentration among locations ranged from $0\text{to}1.97\mathrm{ng}∕\mathrm{mg}$ in sample 1 and from $2.10\text{to}10.27\mathrm{ng}∕\mathrm{mg}$ in sample 2.

3.4.

Correlation of Diffuse Reflectance Spectroscopy and Ex Vivo Measurements

Table 1 lists the mean and standard error (STE) of in vivo DRS and ex vivo tissue oxygenation (or EF5 binding) and MLu concentration measurements for each tumor tissue. The association between in vivo and ex vivo measurements from patients 1 and 2 is shown in Figs. 6 and 7 . Figure 6 shows tissue oxygenation ( $\mathrm{St}{\mathrm{O}}_2$ in percent) measured by DRS versus EF5 binding. Figure 7 shows MLu concentration ( $c_{\mathrm{MLu}}$ in ng/mg) measured by DRS versus MLu concentration measured by the ex vivo spectrofluorometric assay. The error bars on these plots represent the standard error (STE) of the mean. In patients 1 and 2, the number of measurements $\left(n\right)$ was 4 and 2, respectively, for EF5 studies, and 2 and 4, respectively, for $\mathrm{St}{\mathrm{O}}_2$ measurements. In patients 1 and 2, $n=6$ and 3, respectively, for the ex vivo $c_{\mathrm{MLu}}$ assay and 2 and 4, respectively, for the in vivo $c_{\mathrm{MLu}}$ assay.

Fig. 6

$\mathrm{St}{\mathrm{O}}_2$ (in percent) determined by diffuse reflectance spectroscopy (DRS) is plotted versus hypoxia marker EF5 binding (percent of maximum value). In patients 1 (data point with $\mathrm{St}{\mathrm{O}}_2=88.93\%$ ) and 2 (data point with $\mathrm{St}{\mathrm{O}}_2=17.81$ and 53.75%), the number of measurements $\left(n\right)$ was 4 and 2, respectively, for EF5 studies, and 2 and 4, respectively, for $\mathrm{St}{\mathrm{O}}_2$ measurements. The error bars represent the standard error (STE) of the mean. The dashed line shows an empirically fit through three data points (◇) that have been pre-established to represent the association of oxygen partial pressure $\mathrm{p}{\mathrm{O}}_2$ and EF5 binding in an in vitro calibration study: $\mathrm{p}{\mathrm{O}}_2=76$ , 19, and $3.8\mathrm{mmHg}$ correspond to $\approx 1$ , 3, and 10% of EF5 cube reference binding. Oxygen partial pressure $\mathrm{p}{\mathrm{O}}_2$ values were converted to $\mathrm{St}{\mathrm{O}}_2$ and plotted in the figure based on a published study,³⁹ where $\mathrm{St}{\mathrm{O}}_2=\mathrm{p}{\mathrm{O}}_2^n∕\mathrm{p}{50}^n+\mathrm{p}{\mathrm{O}}_2^n$ , $\mathrm{p}50=23.4$ , $n=2.44$ . The correlation of $\mathrm{St}{\mathrm{O}}_2$ and EF5 binding agrees well with the established association curve of $\mathrm{p}{\mathrm{O}}_2$ and EF5 binding.

Fig. 7

MLu concentration ( $c_{\mathrm{MLu}}$ in ng/mg) determined by in vivo DRS is plotted versus MLu concentration ( $c_{\mathrm{MLu}}$ in ng/mg) measured by ex vivo spectrofluorometric assay. In patients 1 (data point with in vivo $c_{\mathrm{MLu}}=2.06\mathrm{ng}∕\mathrm{mg}$ ) and 2 (data point with in vivo $c_{\mathrm{MLu}}=0.49$ and $6.52\mathrm{ng}∕\mathrm{mg}$ ), the number of measurements $\left(n\right)$ was 6 and 3, respectively, for the ex vivo $c_{\mathrm{MLu}}$ assay and 2 and 4, respectively, for the in vivo $c_{\mathrm{MLu}}$ assay. The error bars represent the standard error of the mean. A positive trend was seen between in vivo and ex vivo methods in measuring MLu concentration.

Table 1

Oxygenation (hypoxia) and photosensitizer values in human tumors, measured by in vivo versus ex vivo assays. Tissues were assigned an ID in the operating room at the time that DRS was performed. Oxygen saturation (%StO2) and MLu concentration (cMLung∕mg) were extracted from DRS measurement using a diffusion model (see Methods and Materials in Sec. 2). In patient 1(tissue 1), two continuous measurements were taken on the same site (single contact of the probe and tumor surface). In patient 2 (bulk tissues 1 and 2), four different measurements were taken on four adjacent sites (four fresh contacts of the probe and tumor surface).

As expected, $\mathrm{St}{\mathrm{O}}_2$ and EF5 binding were negatively correlated (Fig. 6). The dashed line (and diamonds) in Fig. 6 is derived from an independent in vitro determination of the association between tissue oxygen $\left(\mathrm{p}{\mathrm{O}}_2\right)$ and EF5 binding level.⁴⁷ In this study,⁴⁷ the relationship between $\mathrm{p}{\mathrm{O}}_2$ and percent EF5 cube reference binding was established: 10% oxygen $(\mathrm{p}{\mathrm{O}}_2=76\mathrm{mmHg})\approx 1\%$ of cube reference binding; 2.5% oxygen $(\mathrm{p}{\mathrm{O}}_2=19\mathrm{mmHg})\approx 3\%$ of cube reference binding; and 0.5% oxygen $(\mathrm{p}{\mathrm{O}}_2=3.8\mathrm{mmHg})\approx 10\%$ of cube reference binding. To obtain the dashed line in Fig. 6, we converted $\mathrm{p}{\mathrm{O}}_2$ to $\mathrm{St}{\mathrm{O}}_2$ based on published results,³⁹ wherein $\mathrm{St}{\mathrm{O}}_2=\mathrm{p}{\mathrm{O}}_2^n∕\mathrm{p}{50}^n+\mathrm{p}{\mathrm{O}}_2^n$ , $\mathrm{p}50=23.4$ , $n=2.44$ , and empirically fit the data (diamonds).

In Fig. 7, a positive correlation between DRS-measured $c_{\mathrm{MLu}}$ and spectrofluorometer-measured $c_{\mathrm{MLu}}$ was observed, although the DRS-measured concentrations were significantly higher than the concentrations found by the ex vivo assay; thus, the patient data are quite different from that of the murine tumors. Among the ex vivo measurements, the significant higher standard error of $c_{\mathrm{MLu}}$ in patient 1 $(c_{\mathrm{MLu}}=0.754\pm 0.164\mathrm{ng}∕\mathrm{mg})$ is due in part to large intratumor heterogeneity. The mean $c_{\mathrm{MLu}}$ $(\pm \mathrm{STE})$ of these two excised samples from the same tumor were $0.459\pm 0.122$ and $1.050\pm 0.178\mathrm{ng}∕\mathrm{mg}$ , respectively.