Gliomas represent 40% of all primary brain tumors, and treatment of glioma poses a challenge due to their tendency to infiltrate the surrounding normal brain. Astrocytoma and glioblastoma are the two most common types of gliomas. The current survival rate of glioblastoma patients treated with surgery for biopsy only, partial resection, and complete resection is 15%, 25%, and 34% respectively, making it one of the most lethal tumors.1 The standard treatment for brain tumor patients is surgery followed by radiation or chemotherapy. The extent of surgical resection is a primary determinant of outcome, and it was shown that the extent of tumor resection is a single most important factor for longer survival.2, 3 The surgical resection is complicated by location of these tumors, as they may be embedded in or near the eloquent areas of the brain such as motor cortex, sensory cortex, and speech centers. Determining the functional areas and the resectable tumor during the surgery has presented a great challenge.
In order to overcome such problems, currently several techniques are employed by surgeons in their quest for near complete removal of the tumor. This includes stereotactic image-guided surgery that delineates the tumor and intraoperative ultrasound or electric mapping to determine the functional areas from the tumor. Stereotactic image-guided surgery based on pre operative magnetic resonance imaging (MRI) scans could be used to determine the position of excision with respect to the position of tumor. However, this technique has intrinsic confounding factors such as the brain shift that occurs during the surgery. This shift may be due to loss of cerebral spinal fluid or due to release of pressure; this can result in significant inaccuracies in resection of the tumor. More recently, in order to eliminate these limitations, many major medical centers have employed intraoperative MRI (iMRI). This technique requires, however, a major investment and modification of operating room settings as well as surgical instrumentation. Moreover, the actual pathological diagnosis can be provided only by biopsy and frozen section, which requires and can be used only in limited capacity. Even in the hands of the most skilled surgeon, complete excision is limited by the tactile and visual feedback. Thus, in order to enhance the ability of the surgeon to excise completely the tumor with more accuracy without sacrificing safety, new technologies need to be developed. Such new technologies should be able to distinguish accurately between the tumor and normal brain and guide the tumor resection in near-real time.
Laser-induced fluorescence spectroscopy (LIFS) represents a promising new adjunctive technique in the diagnosis of tissue pathologies.4, 5, 6, 7, 8 Fluorescence spectroscopy involves the use of lasers to excite endogenous autofluorescence (label-free) within tissues. Either the steady-state5, 9, 10 or the time-resolved6, 11, 12 fluorescence spectroscopy technique can be used to measure or monitor these changes. The time-resolved measurement resolves fluorescence intensity decay in terms of excited-state lifetimes and thus provides additional information about the underlying fluorescence intensity decay dynamics and improves its specificity. Earlier work has demonstrated that label-free steady-state LIFS of endogenous fluorophores (autofluorescence) can be used as a tool for diagnosis of neoplasm in the brain. Several types of brain tumors have been investigated, both in vivo and ex vivo, using this technique. These include studies of glioblastoma, 5, 6, 9 astrocytoma,9 oligodendroglioma,9 and metastatic carcinoma.9 In addition, our group has previously reported results on the utility of time-resolved LIFS (TR-LIFS) for distinguishing glioblastoma from uninvolved cerebral cortex and white matter and on its increased specificity relative to steady-state fluorescence spectroscopy.6, 11, 12
The overall objective of this study is to investigate the use of TR-LIFS as a rapid diagnostic tool for intraoperative delineation of tumor from surrounding normal tissue. Such a tool has a potential to enhance the ability of the neurosurgeon-neuropathologist to distinguish between tumor and normal brain during surgery. Extending on our earlier work conducted in ex vivo specimens,6, 12 The current study is designed (1) to determine the spectrum- and time-dependent autofluorescence emission characteristics of gliomas (high- and low-grade), normal cortex (NC), and normal white matter (NWM) measured in vivo in patients undergoing surgical resection of brain tumors; and (2) to identify the main autofluorescence signatures that provide means of discrimination between tumor and normal tissue. In addition, this study evaluates the possible differences between fluorescence signatures measured in vivo versus those measured ex vivo in tumor specimens as reported earlier.6, 12
Materials and Methods
A total of 17 patients diagnosed with a primary brain tumor (glioma) were included in this study. The patients were scheduled for surgical removal of the brain tumor and underwent the planned operation. During this procedure, the TR-LIFS fiber-optic probe was positioned above the areas of interest using a Greenburg retractor, and the brain tissue was spectroscopically investigated (total: 34 locations). While for this study, we recorded the fluorescence emission for multiple sites , the biopsy was performed at the surgeon’s decision based on the suitability of each location for biopsy. Thus, we report here only data from the sites where biopsy was available. Areas with distinct pathologic features were selected based on the gross visual evaluation by the neurosurgeon who-performed the surgery and preoperative MRI scans typically used to guide the surgery. For future validation, intraoperative biopsy was conducted at each point of spectroscopic investigation, except for areas considered normal or not considered suitable for physical biopsy due to risks posed to the patient. The study was carried out with approval of the Cedars-Sinai Medical Center Institutional Review Board.
Experiments were conducted with an instrumental setup, which allowed for spectrally resolved fluorescence lifetime measurements. A schematic of the optical and electronic layout of the apparatus is shown in Fig. 1 . A detailed account of this apparatus and its performance is given elsewhere.13 Briefly, it consisted of (1) a pulsed nitrogen laser (Lasertechnik Berlin, Berlin, Model MNL200-ATM205, , pulse FWHM), which was used as the excitation source; (2) a custom-made sterilizable bifurcated fiber-optic probe (CeramOptec, East Longmeadow, Massachusetts); (3) an imaging spectrometer/monochromator (Chromex, Inc., Albuquerque, New Mexico, Model 250is/sm, F/4.4, grating, blazed at ); (4) a gated multichannel plate photomultiplier tube (MCP-PMT; Hamamatsu, Bridgewater, New Jersey, Model R5916-50, rise ) with a fast preamplifier (Hamamatsu, Bridgewater, New Jersey, Model C5594, ); (5) a digital phosphor oscilloscope (TDS5104, Tektronix, Beaverton, Oregon, ); and (6) a computer workstation and peripheral electronics. The instrument allowed for mobility, as it was contained in a standard endoscopic cart , internally modified to accommodate the individual devices.
Light delivery and collection were implemented with a custom-made bifurcated sterilizable probe. The probe consisted of nonsolarizing silica/silica step index fibers of 0.11 numerical aperture (NA; CeramOptec, East Longmeadow, Massachusetts). It had a central excitation fiber of core diameter, surrounded by a collection ring of twelve -core-diameter fibers. The collection fibers were beveled at a angle in order to improve excitation/collection overlap for small tissue-to-probe distances. The center-to-center separation between the excitation and collection fibers was . The probe was flexible throughout its entire length except for a distal part consisting of a rigid stainless steel tube. This facilitated the mounting and micromanipulation of the probe. The laser light was coupled into the illumination channel of the probe with a standard subminiature version A (SMA) connector, while the distal end of the collection channel was formed into a straight line in order to facilitate coupling to the spectrograph. After sample excitation, the emitted fluorescence light was collected and directed into the entrance slit of the spectrometer via the collection channel of the probe. A long-pass filter (WB360, Optima, Tokyo, ) was placed before the entrance slit of the monochromator to prevent contamination of the fluorescence signal by the excitation light. The signal was then detected by the MCP-PMT, amplified by the fast preamplifier, and finally digitized at resolution by the digital oscilloscope. The overall time resolution of the systems was approximately .
The fiber-optic probe was positioned at above the exposed brain tissue specimen by the Greenburg retractor to optimize the probe light collection efficiency as previously reported14 and to steady the probe over the tissue. Time-resolved emission of each sample was recorded in the spectral range and scanned at intervals (Fig. 2 ). The total amount of time required to obtain a complete time-resolved spectra by scanning through all the wavelengths is approximately . Five consecutive measurements of the fluorescence pulse emission at 390 and were performed to assess the reproducibility of the fluorescence lifetime measurement. The energy output of the laser (at the tip of the fiber) for sample excitation was adjusted to /pulse. The area illuminated by the probe (defined by the NA of the excitation fiber and the probe-to-tissue distance) was approximately ; thus, the total fluence per pulse at the tissue level was approximately , which is very well within the safety limits. After the spectroscopic analysis, the tissue was biopsied at the exact site and sent for pathological investigation.
Except for the normal cortex (NC), tissue was biopsied from the areas where fluorescence was recorded. Each biopsy sample was fixed in 10% buffered formalin. The tissue samples were then cut and fixed on the slides; the slides were stained by the standard hematoxylin & eosin (H&E) stain. All slides were studied by the pathologist and correlated with original fluorescence spectroscopy measurements results. Histologically, gliomas are categorized in low-grade (grade I), intermediate-grades (grades II and III), and high-grade gliomas or glioblastoma multiforme (grade IV), which is based on the WHO grading (Kernohan and Sayred).15 In the current study, however, we grouped the gliomas in two sets: low-grade glioma (LGG), including grades I and II, and high-grade glioma (HGG), including grades III and IV.
TR-LIFS Data Analysis
In the context of TR-LIFS, the intrinsic fluorescence impulse response functions (FIRFs), , describe the real dynamics of the fluorescence decay. The IRFs were recovered by numerical deconvolution of the measured input laser pulse from the measured fluorescence response pulse transients. The Laguerre expansion of kernels technique.16 was used for deconvolution. This method allows a direct recovery of the intrinsic properties of a dynamic system from the experimental input-output data. The technique uses the orthonormal Laguerre functions to expand the IRFs and to estimate the Laguerre expansion coefficients (LECs) . Once the fluorescence IRFs were estimated for each emission wavelength, the steady-state spectrum was computed by integrating each intensity decay curve as a function of time. Further, to characterize the temporal dynamics of the fluorescence decay, two sets of parameters were used: (1) the average lifetime computed as the interpolated time at which the IRF decays to 1/e of its maximum value; and (2) the normalized value of the corresponding LECs. Thus, a complete description of the fluorescence emission from each sample as a function of emission wavelength, , was given by the variation of a set of spectroscopic parameters ( , , and LECs). This analytical approach for characterization of fluorescence decay was recently developed by our research group, and its advantages over other deconvolution methods are described in detail elsewhere.16, 17
Parameters derived from the time-resolved fluorescence spectrum were used for the discrimination of distinct types of tissue. These parameters were based on the discrete fluorescence intensity , average lifetime , and Laguerre coefficient values at all recorded wavelengths. A univariate statistical analysis (one-way ANOVA) was used to identify a set of spectroscopic parameters that best discriminate between distinct tissue types as defined by the histopathologic analysis. This was achieved by comparing time-resolved derived parameters ( and LECs) at every and intensity ratios at discrete wavelengths. A -value of was assumed to indicate statistically significant differences. A systematic comparison of the -values obtained for each parameter at every allowed us to identify a set of spectroscopic parameters ( , and LECs at specific yielding the lowest -values) likely to provide means of discrimination among different compositional features of the brain tissue. All statistical analysis was conducted using SPSS (SPSS, Inc., Chicago, Illinois).
The histological examination revealed the presence of the following tissue types: cortex ; white matter ; -grade glioma with WHO grade I or II ; and -grade glioma with WHO grade III or IV . In addition, the histological analysis revealed that the cells underlying the areas of laser excitation were not visibly altered post-excitation and had a well-preserved morphology.
Figure 2 shows the typical fluorescence response IRF from which the fluorescence emission spectra and the average lifetime were recovered. The NC (Fig. 3 ) autofluorescence was characterized by a broad emission spectrum with two well-defined peaks, one within the wavelength range, and the other at shorter emission wavelengths at approximately . The former peak is consistent with the fluorescence emission of enzyme cofactor nicotinamide adenine dinucleotide (NADH). Attenuation of fluorescence intensity was observed at about that corresponds to the hemoglobin absorption band. The lifetime or the intensity decay characteristics (Fig. 3) in the normal cortex showed that the emission was significantly longer lasting at blue-shifted when compared with the red-shifted wavelengths. The lifetimes observed at are consistent also with the NADH lifetime reported in the literature.4, 18, 19 The NWM was characterized by a similar broad fluorescence emission spectrum (Fig. 3) with two well-defined peaks around 390 and , the latter corresponding to emission enzyme cofactor NADH, respectively. The attenuation of fluorescence intensity observed at about corresponds to the hemoglobin absorption band.10 The average fluorescence lifetime (Fig. 3) showed similar characteristics as NC with longer lasting emission values in the blue-shifted when compared with the red-shifted wavelengths . The fluorescence emission spectra of LGG samples showed a relatively narrow broadband emission characterized by a single well-defined peak at wavelength [Fig. 3] lasting for less than . Across the entire spectrum, the fluorescence emission decay was short-lasting, with an average lifetime mainly at about . As clearly observed, the LGG shows a very distinct average lifetime when compared with NC , NWM , and HGG at . HGG demonstrated similar fluorescence emission spectrum to NC with a well-defined peak between wavelengths, which is consistent4, 18, 20 with emission of enzyme cofactor NADH, and a second smaller peak at . Similar attenuation of fluorescence intensity at about that corresponds to the hemoglobin absorption band was also observed. The fluorescence emission closely followed the trend observed in normal cortex at and at .
The results of the ANOVA demonstrated that the following parameters were within the statistical significance ( -value ): , , , , , , , , , LEC- , LEC- , LEC- , LEC- , LEC- , LEC- , and LEC- . This set of parameters including , , LECs were found able to discriminate between the four type of tissues of interest (LGG, HGG, NC, and NWM). Figure 4 demonstrates the mean with the dispersion and skewness. Spectroscopic parameters derived from both spectral domain and time domain ( , LECs) were able to distinguish LGG from healthy brain tissue (NC, NWM). These parameters were generally characterized by a narrow distribution around the mean value. In contrast, the parameters derived for HGG exhibited a broad distribution. A few parameters were able to distinguish HGG tissue from NWM, but not from NC. Overall, based on this analysis of variance test, we determined that most of the time-resolved fluorescence parameters that help in distinguishing between the four tissue types were derived from three wavelengths: , , and . In addition, intensity ratios including intensity values derived from the main peak emissions ( and ) are important for tissue discrimination.
By examining the ex vivo brain tumor tissue using TR-LIFS, we demonstrated previously6, 12 that this technique has the potential to characterize and diagnose brain tumors. Extending on these early results, in this study, we examined for the first time in vivo the fluorescence emission of low- and high-grade gliomas and compare this with that of normal brain tissue. We demonstrated that TR-LIFS could be used to distinguish between the glioma tissue versus NWM and NC based on specific spectral and temporal fluorescence characteristics.
TR-LIFS Features in the Characterization of Brain Tumors versus Normal Brain Tissue
We determined that when the tissue is excited using ultraviolet light (e.g., ), the emission from two wavelength ranges appeared important for diagnosing gliomas from normal brain tissue: and (the region of main peak emission for NC and LGG and HGG). Striking and distinct fluorescence parameters were observed in LGG when compared with the NC or NWM. One important observation was that LGG shows mainly a single peak of fluorescence emission at of wavelength, whereas the NC and NWM demonstrate two fluorescence emission peaks centered at and . This is interpreted as the presence of a single fluorophore in LGG. This is further reinforced by inspecting the fluorescence lifetimes, which clearly show fast-intensity decay and nearly constant average lifetime values across the fluorescence spectrum. This particular emission with peak at and lifetime of around is attributed to NADH, which demonstrates spectral10 and temporal18, 19, 21 characteristics as described in previous studies (Table 1 ).
Summary of fluorescence properties listing all the fluorophores significant in the brain.
In contrast to these findings for LGG, normal tissue (i.e., NC and NWM) and HGG demonstrated additional fluorescence emission peak and longer fluorescence lifetimes at . This longer lasting emission at blue-shifted wavelengths when compared to that at red-shifted emission wavelengths suggests that this additional emission peak originates from a distinct fluorophore, other than NADH. While relatively scarce information on the origin of the fluorescence emission of brain tissue at on UV has been reported, we identified two possible fluorophores known to emit fluorescence within this wavelength band: Pyridoxamine-5-phosphate (PMP) and glutamate decarboxylase (GAD).22 Both of these fluorophores are metabolically affected in gliomas and are briefly discussed in the following.
Pyridoxine or vitamin is required by the body to convert carbohydrates to glucose. In addition, the function of pyridoxine is the formation of RNA and DNA, and it plays a role in the removal of the toxic chemical homocysteine. Pyridoxamine phosphate (PMP) along with Pyridoxine phosphate (PN) is converted by pyridoxine- -phosphate oxidase (PNPO) into pyridoxal phosphate (PLP). PMP fluoresces with peak emission at , along with PLP, which fluoresces with emission peak between 315 and (Ref. 23). But in hydrated form at neutral pH, it shows similar fluorescence characteristics as PMP. The PLP mechanism is known to be disturbed in cancers.24, 25, 26 Our TR-LIFS studies of gliomas and normal brain tissue show that there is a decrease/absence of the emission peak at of wavelength in glioma when compared with normal cortex tissue. We hypothesize that this may be due to altered metabolism in the PLP mechanism that leads to decrease in the PMP levels in the brain.
The second fluorophore possibly responsible for changes observed at emission peak at is glutamate decarboxylase, which is an enzyme essential in the formation of neurotransmitter GABA ( -amino butyric acid) from amino acid L-glutamate in a single step. Glutamate decarboxylase shows fluorescence emission maxima at of wavelength when excited at at neutral pH.27 GABA is a most important inhibitory neurotransmitter found in abundance in the brain.28 The absorption maxima of GAD are and . GABAA-benzodiazepine receptors are present in of brain, and the subunit of a GABAA receptor is expressed throughout the brain.29 Most of the GAD is present in the neurons at the synapses. It is known that the activity of this enzyme increases with the differentiation of the neuroblastoma cells. It has been shown that activities of GAD are much lower in the C6 glioblastoma and C1300 neuroblastoma cells than in normal mouse cerebral cortex.30 C6 glioma cells may represent a neuroglial precursor at an early developmental.31 Moreover, it is known that malignant glioma cells in rats actively secrete glutamate in vitro and that the extracellular levels of glutamate have been shown increased both in and around glioma implants in vivo.32 This may suggest an absence of glutamate decarboxylase in its active form in neoplastic tissue.
The ANOVA test performed on various TR-LIFS-derived parameters demonstrated that a small subset of spectral and time-resolved parameters (total: 16) retrieved primarily from a few wavelengths (390, 440, and ) provide means of discrimination of distinct brain tissue types. This finding is important, as it allows for future design of TR-LIFS devices that collect fluorescence decay transients at a limited number of spectral bands, as recently reported.33 In turn, such design will allow for fast data acquisition time (less than for each point measurement). HGG samples demonstrated a high degree of variability of both fluorescence emission intensity and average lifetime at . This variability may be due to a high degree of variation in the genetic as well as metabolic makeup of HGGs. This finding suggests that future studies focused on HGG have to also evaluate the changes in fluorescence parameters as a function of tumor heterogeneity.
While the in vivo measurements of fluorescence intensity data are often affected by various factors such as the presence of endogenous absorbers (e.g., blood hemoglobin), variation in fiber-optic probe excitation collection-collection geometry (e.g., moving artifacts due to arterial pulsation or breathing) and photobleaching, we determined that the use of time-resolved (lifetime) information has the potential to account for factors that nonlinearly affect the fluorescence measurement. Figure 5 , for example, demonstrates that while in some cases the emission spectrum was significantly altered by breathing artifacts and blood absorption, the average fluorescence lifetime is virtually unaffected. This underscores the importance of using fluorescence decay information as an additional means for tissue discrimination.
Ex Vivo versus In Vivo TR-LIFS Measurements
Current results demonstrate that the brain tumor fluorescence signatures obtained in vivo (in patients undergoing brain tumor surgery) are distinct from those we previously reported for ex vivo measurements (brain tumor specimens).6, 11, 12, 34 When the fluorescence emission from the ex vivo NC tissue was compared with the in vivo samples, we noticed that the ex vivo samples show a single peak of emission. This suggests that the additional fluorophore emitting at the wavelength band is absent in the ex vivo samples. The peak emission of NADH also appeared blue-shifted to . The difference in the emission characteristics becomes even more evident when the average lifetime at and are taken into consideration. While the tissue fluorescence measured in ex vivo samples demonstrated a uniform average lifetime value across the emission spectrum, the fluorescence measurements in the in vivo samples showed a decrease (approximate at versus at ) of lifetime values as a function of emission wavelength. The emission spectrum of ex vivo NWM showed similar trends (two distinct emission peaks at and ) to the in vivo samples. However, the lifetime information at these peaks was found to have slightly longer lifetime at in ex vivo tissue when compared with the in vivo samples . The lifetime at is similar in both ex vivo and in vivo samples of normal white matter . LGG fluorescence emission (Fig. 6 ) in both ex vivo and in vivo tissue is characterized by a single peak at and a uniform lifetime along the emission wavelengths.
While the exact cause of differences in the brain tissue fluorescence emission ex vivo versus in vivo is still to be determined, we hypothesize that such changes are determined by difference in the environment and/or metabolic changes that may occur after tissue excision. These differences, however, are expected to provide additional insight for future studies targeted at understanding the origin of fluorescence emission from brain tissue.
This study has demonstrated for the first time the ability of a time-resolved fluorescence spectroscopy technique to characterize in vivo the biochemical features of primary brain tumors that are clinically relevant. Although this is a pilot study conducted on a small number of patients, we have shown that normal brain tissue (NC, NWM) and LGG are well-characterized by distinct fluorescence emission features. Fluorescence temporal features including average lifetime and Laguerre coefficients derived from three spectral emission bands centered at 390, 440, and can be used to distinguish LGG from normal brain tissue. Thus, future optimization of TR-LIFS data acquisition at these predefined wavelength bands33 centered around these wavelengths combined with the fast analysis of the fluorescence intensity decay facilitated by the Laguerre expansion of kernels deconvolution approach,16, 17 will enable near-real time (less than ) tissue diagnosis. We also determined that while fluorescence of LGG is distinct and can be used to clearly delineate from normal tissue, HGG fluorescence characteristics have a large variability, which appears to be correlated with heterogeneous composition of high-grade glioma, a factor that is not taken into account in the conventional gross histopathological diagnosis or classification of the brain biopsy sample. We also observed differences in the emission characteristics between the ex vivo and in vivo tissues that indicate a possible change in the biochemical composition of tissue after excision. This finding suggests that studies conducted in brain tissue specimens cannot be fully translated to the in vivo situation. In turn, this emphasizes the need for further validation of fluorescence-based diagnostic devices directly in vivo in a large number of patients and development of optimized TR-LIFS devices that allow for clinical investigations. Current study has provided the base of knowledge that will enable further development of such TR-LIFS devices.
This work was supported in part by the Whitaker Foundation (RG-01-0346, PI: L. Marcu) and the Maxine-Dunitz Neurosurgical Institute at Cedars-Sinai Medical Center and by the WISE Fund at the University of Southern California.
https://doi.org/10.1002/1097-0142(197506)35:6<1551::AID-CNCR2820350612>3.0.CO;2-V 0008-543X Google Scholar
https://doi.org/10.1562/0031-8655(2003)077<0309:DPOAFA>2.0.CO;2 0031-8655 Google Scholar
https://doi.org/10.1562/0031-8655(2001)073<0396:IVBTDU>2.0.CO;2 0031-8655 Google Scholar
https://doi.org/10.2741/1878 1093-4715 Google Scholar
https://doi.org/10.1364/AO.43.002846 0003-6935 Google Scholar
https://doi.org/10.1007/s00401-007-0243-4 0001-6322 Google Scholar
https://doi.org/10.1016/1011-1344(95)07178-4 1011-1344 Google Scholar
https://doi.org/10.1016/S1011-1344(96)07327-7 1011-1344 Google Scholar
https://doi.org/10.1073/pnas.0832308100 0027-8424 Google Scholar
https://doi.org/10.1021/bi972983r 0006-2960 Google Scholar
https://doi.org/10.1007/s00429-003-0374-x 0340-2061 Google Scholar
https://doi.org/10.1161/hq0701.092091 1079-5642 Google Scholar