Real Time Analysis of Brain Tissue by Direct ... - ACS Publications

Sep 14, 2011 - Direct combination of cavitron ultrasonic surgical aspirator (CUSA) and sonic spray ionization mass spectrometry is presented. A commer...
0 downloads 0 Views 4MB Size
ARTICLE pubs.acs.org/ac

Real Time Analysis of Brain Tissue by Direct Combination of Ultrasonic Surgical Aspiration and Sonic Spray Mass Spectrometry Karl-Christian Sch€afer,† Julia Balog,‡ Tamas Szaniszlo,‡ Daniel Szalay,‡ Geza Mezey,§ Julia Denes,† Laszlo Bognar,§ Matthias Oertel,|| and Zoltan Takats*,†,^ †

Institute for Inorganic und Analytical Chemistry, Justus-Liebig-University, Giessen, Germany Medimass Ltd., Budapest, Hungary § Neurosurgery Clinic, University of Debrecen, Debrecen, Hungary Neurosurgery Clinic, University Hospital Giessen and Marburg, Giessen, Germany ^ 1st Department of Pediatrics, Semmelweis University, Budapest, Hungary

)



bS Supporting Information ABSTRACT: Direct combination of cavitron ultrasonic surgical aspirator (CUSA) and sonic spray ionization mass spectrometry is presented. A commercially available ultrasonic surgical device was coupled to a Venturi easy ambient sonic-spray ionization (V-EASI) source by directly introducing liquified tissue debris into the Venturi air jet pump. The Venturi air jet pump was found to efficiently nebulize the suspended tissue material for gas phase ion production. The ionization mechanism involving solely pneumatic spraying was associated with that of sonic spray ionization. Positive and negative ionization spectra were obtained from brain and liver samples reflecting the primary application areas of the surgical device. Mass spectra were found to feature predominantly complex lipid-type constituents of tissues in both ion polarity modes. Multiply charged peptide anions were also detected. The influence of instrumental settings was characterized in detail. Venturi pump geometry and flow parameters were found to be critically important in ionization efficiency. Standard solutions of phospholipids and peptides were analyzed in order to test the dynamic range, sensitivity, and suppression effects. The spectra of the intact tissue specimens were found to be highly specific to the histological tissue type. The principal component analysis (PCA) and linear discriminant analysis (LDA) based data analysis method was developed for real-time tissue identification in a surgical environment. The method has been successfully tested on post-mortem and ex vivo human samples including astrocytomas, meningeomas, metastatic brain tumors, and healthy brain tissue.

I

ntraoperative identification of tissues is critically important in various fields of cancer surgery. Since the visual and tactile localization of a solid tumor is not sufficiently sensitive and selective, surgeons have to rely on preoperative medical imaging (computerized tomography (CT), magnetic resonance (MR), positron emission tomography (PET), etc.) and intraoperative histology.1 7 This latter method is a simplified and thus accelerated version of traditional histological analysis, which is capable of providing results within 20 40 min during a surgical intervention. However, this time frame is extremely long from the point of view of the intervention and the reliability of the results is far behind that of traditional histopathology. A number of alternative methods have been developed for in situ, real-time identification (or differentiation) of tissues in a surgical environment. Several fluorescent or radioactive tags including nanoparticles have been described recently for in situ brain tumor visualization.8 10 These methods all involve the administration of a fluorescent or radioactive compound, which is selectively accumulated in the tumor cells. The disadvantages of these techniques include the specificity of the labeling, the specific operation environment, and in some cases, the defined time range in which the operation can take r 2011 American Chemical Society

place after the drug administration. For example, 5-aminolevunilic acid induces the accumulation of fluorescent protoporphyrin IX in glioblastoma cells allowing in vivo, in situ observation of tumor tissue during brain surgery.11 However this approach requires the administration of labeling drugs and also requires a modified surgical instrumentation and setup for the detection of fluorescence. The use of imaging methods (particularly sonography and MRI) and infrared spectroscopy during surgery was also proposed in the past decades, but the complete differentiation of healthy and cancerous tissue has not yet been demonstrated in a human surgical environment.12 14 It has been demonstrated recently that several mass spectrometric ionization methods have the potential to improve the specificity of intrasurgical histology examinations.15 17 Furthermore, dedicated mass spectrometric methods were also developed for intraoperative tissue identification.18 20 Rapid evaporative ionization mass spectrometry (REIMS) is based on the Received: May 17, 2011 Accepted: September 14, 2011 Published: September 14, 2011 7729

dx.doi.org/10.1021/ac201251s | Anal. Chem. 2011, 83, 7729–7735

Analytical Chemistry electrosurgical evaporation of the tissues, where the ions formed during dissection are subsequently analyzed. While REIMS analysis yields tissue-specific data in real time, its applications are limited to interventions using thermal evaporative dissection, i.e., electrosurgery or laser surgery. However, applicability of these techniques is limited in a number of cases, especially in the field of brain surgery. While the originally described REIMS method is based on monopolar electrosurgery, in the case of brain surgery, bipolar electrosurgery is used almost exclusively. This method (i.e., bipolar electrosurgery) provides poor ion yield in an identical setup, thus the resulting mass spectra are insufficient for direct real-time tissue identification. In brain surgery, though, another group of surgical methods is also utilized which employ mechanical disintegration of the tissues and vacuum suction to remove resulting tissue debris. These methods include simple surgical aspiration and ultrasonic aspiration methods.21,22 In both cases, the technique is applied in order to avoid excessive tissue damage and incidental bleeding, since the aspirator device leaves the blood vessels intact. The cellular components are disintegrated by the ultrasound and aspirated, leaving the vasculature visible. At this point the blood vessels are either cauterized or surgically blocked before the intervention proceeds. Further advantages of ultrasonic surgery include the complete removal of irreversibly damaged tissue from the surgical site and the lack of electric current. These features are particularly important in the case of brain surgery, where scar formation can lead to epilepsy and stray electric current may induce functional damage to the brain (e.g., memory loss).20 23 During neurosurgical interventions, it is generally not possible to make a tumor resection in the intact, healthy brain tissue due to the functionality of the area. Hence, it is crucial to identify the margin of the tumor precisely, and it is often left to the neurosurgeon’s experience to decide on the position of the actual tumor border and thus the resection line. While the importance of coupling ultrasonic aspiration with mass spectrometric tissue identification has already been recognized, only off-line combinations of the two techniques were demonstrated.16,17 In this study, the ultrasonic aspirator was used as a sampling tool combined with neuronavigation for accurate determination of the origin of samples. Millimeter-sized tissue particles were filtered out from the effluent of the surgical device, followed by freezing, transportation to mass spectrometric laboratory, cryosectioning, and imaging mass spectrometric analysis by desorption electrospray ionization-mass spectrometry (DESI-MS). Although this approach is highly innovative, it allows analysis of individual tissue specimens in the time range of ∼10 min. In contrast to the recently described off-line approaches, in the present study the feasibility of online coupling of cavitron ultrasonic surgical aspiration (CUSA) with mass spectrometric analysis is demonstrated. Online coupling was implemented by introducing the effluent of the CUSA device directly into a Venturi easy ambient sonic-spray ionization (V-EASI) source.24 The resulting technique is demonstrated to allow in situ, quasi real-time, continuous tissue identification throughout ultrasonic surgical interventions.

’ EXPERIMENTAL SECTION CUSA Instrument and Ion Transfer. A commercially available ultrasonic surgical unit (Selector, Erbe Elektromedizin GmbH, T€ubingen, Germany) was used for tissue disintegration and removal. Ultrasonic disintegration was performed at 24 kHz, with a 60% energy setting. PTFE tubing (1/16 in. o.d., 1.2 mm i.d.)

ARTICLE

Figure 1. (a) Experimental setup for the coupling ultrasonic surgical handpiece to a mass spectrometer. Inset shows tip of surgical handpiece. Arrows indicate water flow (down) and evacuation of tissue debris (up). (b) Detailed scheme of V-EASI-type setup mounted in front of mass spectrometer inlet. (c) The scheme represents the surgical handpiece.

was placed into the drain line of the commercial surgical handpiece in order to minimize memory effects. This tubing was connected to a custom-built Venturi pump through a 2 m long (1/8 in. o.d., 2 mm i.d.) tube. The Venturi pump was driven by nitrogen with the inlet pressure set to 10 bar. The Venturi pump was mounted on the mass spectrometer atmospheric interface orthogonal to the heated capillary inlet (Figure 1b). An identical atmospheric interface setup was described earlier by Santos et al. termed Venturi easy ambient sonic-spray ionization (V-EASI).24 High-resolution mass spectrometry was performed using a Thermo LTQ Orbitrap Discovery instrument (ThermoScientific, Bremen, Germany). Low-resolution measurements were performed using an LCQ Deca XP Max quadrupole ion trap mass spectrometer (ThermoFinnigan, San Jose, CA). During ultrasonic tissue resection, a vibrating handpiece is brought into the proximity of the tissue surface and the vibration (∼20 40 kHz) is transmitted to the tissue by an annular water jet (for details, see inset of Figure 1a). The liquid is continuously removed from the tissue surface by vacuum suction through a hollow handpiece (Figure 1c) with typical flow rates in the range of 2 10 mL/min. Aerosol Inhalation Safety Regulations. An appropriate breathing mask (M7500 series mask with 6055 A2 filter, 3 M Deutschland GmbH, Neuss, Germany) was worn throughout all experiments. Samples. Food grade porcine organs were used for the systematic characterization of experimental parameters. Human brain samples including tumor samples were obtained from the Institute of Neurosurgery, University of Debrecen (Debrecen, Hungary) and Klinik f€ur Neurochirurgie, Universit€atsklinikum Giessen and Marburg. Human brain slices and liver cancer samples were obtained from the Institute of Pathology, University of Debrecen. Complete ethical clearance was obtained for the collection and analysis of human samples. Data Analysis. Mass spectra were collected in single stage MS, negative and positive ion mode, in the mass range 150 1000 m/z, 7730

dx.doi.org/10.1021/ac201251s |Anal. Chem. 2011, 83, 7729–7735

Analytical Chemistry

ARTICLE

Figure 2. (a) Positive ion mode CUSA/SSI spectrum of porcine brain cortex. (b) Negative ion mode CUSA/SSI spectrum of human brain tumor sample. The insets show multiple charged ion signals, which have been associated with peptides.

unless otherwise stated. Spectral data was binned using a 0.01 m/z bin size in the case of high-resolution experiments and a 1.0 m/z bin size in the case of low-resolution experiments and stored in an SQL database (Oracle, Redwood City, CA) containing the full known classification of every tissue specimen corresponding to the spectra, including WHO tumor type and grade. On a chosen, normalized subset of spectra, principal component analysis (PCA) was carried out in order to reduce the dimensions to 60. PCA was followed by linear discriminant analysis (LDA) without further reduction of the dimensions. The number of components was selected based on a previously described cross-validation study,20 where the misclassification rate was tested as a function of the number of components. The resulting PCA/LDA model was saved in the database and used in the tissue identification process or, alternatively, the points were plotted as function of the first 3 PCA or LDA components for demonstration purposes (see Figure 5). The real-time classification of new spectral data is based on the saved PCA/LDA model, the new spectrum is first transformed to the 60-dimensional PCA space then to the 60-dimensional LDA space, and subsequently a squared Mahalanobis distance

is calculated to every class average in the model. The new spectrum is classified to the closest class average, if the spectra is in the range of (5  standard deviation from the class average in every dimension. If the spectrum is outside of the above specified range for all included tissue classes, the spectrum is marked as “outlier”.

’ RESULTS AND DISCUSSION In the present study, we investigated the possibility of coupling ultrasonic surgical techniques with online mass spectrometric analysis. Ultrasonic aspirator devices disintegrate tissues and evacuate the resulting tissue debris. Because of the continuous flushing of the surgical site, the tissue debris is continuously drained away in form of a dilute suspension in physiological saline. Theoretically, the online mass spectrometric analysis of this drainage could provide sufficient data for the identification of the disintegrated and aspirated tissues. Although spray ionization of the drainage seems to be a straightforward approach, the presence of macroscopic tissue particles effectively hinders the utilization of any traditional spray setup that employs capillaries as a mean of 7731

dx.doi.org/10.1021/ac201251s |Anal. Chem. 2011, 83, 7729–7735

Analytical Chemistry

ARTICLE

Table 1. Identified Lipids with CUSA/SSI in the Negative and Positive Ion Modes compound

negative ion mode

fatty acids

16:0, 16:1, 18:0, 18:1, 18:2, 20:3, 20:4, 22:6

phosphatidyl ethanolamines phosphatidyl serines

16:0, 18:0, 18:1, 18:2, 20:4, 20:2, 20:3 22:6 16:0, 18:1, 18:0, 20:5, 22:6

phosphatidyl inositols

16:0, 18:0, 20:4

plasmalogens

16:0, 18:1

phosphatidic acids

16:0, 18:1, 18:0

sulfatides

18:1, 22:1, 22:2, 24:0, 24:1, 24:2, 24:3, 26:2

compound

positive ion mode

phosphatidyl cholines

16:0, 18:0, 18:1, 18:2, 20:3, 20:4, 22:6

phosphatidyl ethanolamines triglycerides

16:0, 18:0, 18:1, 18:2, 20:4, 22:4, 22:6 18:2, 20:0, 20:4, 18:1

liquid transfer. The recently described Venturi easy ambient sonic spray ionization technique,24 however, allows the use of tubing with diameters exceeding 1 mm. A further advantage of the Venturi device is that it provides sufficient suction force for the liquid transfer. The experimental setup (Figure 1) was tested using porcine brain samples as a model system. Resulting spectra are shown in Figure 2. As it was expected upon theoretical considerations, the spray ionization of tissue debris produces predominantly ions corresponding to various lipid species ranging from simple fatty acids to glycerophospholipids or sulfatide glycosphingolipids. The list of identified lipid species is given in Table 1. Unlike other “ambient” mass spectrometric methods like REIMS, the cavitron ultrasonic surgical aspiration/sonic spray ionization (CUSA/SSI) method produces ions of metabolic constituents, carbohydrates, and peptides besides the highly intensive lipid-related ion population. Negative ion mode mass spectrum of porcine brain tissue in the lower mass range (m/z 150 500) is shown in Figure S-1 in the Supporting Information. As it is shown in the insets in Figure 2b, multiply charged peptide anions have also been observed in the negative mode. The ion signals indicate the presence of species A with a molecular mass of 3788.3 Da and peptide B with molecular mass of 4960.5 Da. Species A was associated with calcitonine gene related peptide (CGRP), which is an abundant peptide in the CNS and has been described as a diagnostic marker in the case of lung tumors.25,26 Species B was tentatively associated with thymosine β 4 peptide.27 Thymosine β peptide has also been described as a prognostic marker in the case of nonsmall cell lung cancer (NSCLC). Our tentative identification is based on the facts that the sample was a brain metastasis of NSCLC, and the corresponding peptides have been associated with metastasis formation;28,29 however, further experiments are needed for the validation of these identifications. The effect of various instrumental parameters on spectral intensity was investigated, including the power setting of the ultrasonic surgical device, nitrogen inlet pressure of the Venturi-pump, relative geometry of the atmospheric interface, voltage settings of the ion optics, and inlet capillary temperature. Signal intensity is strongly dependent on the ultrasound power (Figure 3a). Results clearly indicate that ultrasonic disruption of the cells is a prerequisite for obtaining high concentrations of cell components in the draining liquid and thus sufficiently good spectral quality. The data also show that the presented strategy cannot be used in the absence of ultrasonic disintegration, i.e., in the case of simple surgical aspirators which are also widely used in neurosurgery.

Figure 3. (a) Dependence of signal intensity on ultrasonic power setting and (b) dependence of signal intensity on nitrogen inlet pressure.

Investigation of the dependence of signal intensity on the nitrogen inlet pressure of the Venturi pump shows a dynamic relationship, with a clear saturation phenomenon in the 0 30 bar range (Figure 3b). Because of the clear optimum value, a 10 bar nitrogen inlet pressure was used throughout the experiments described below. Both the dependence of signal intensity on the nebulizing gas flow rate and the overall experimental setup suggest a sonic spray-like ionization phenomenon, since the sample undergoes a solely pneumatic spraying, in the absence of electric potential gradients or thermal effects. This assumption is also in agreement with the observation that only species undergoing dissociation in the liquid phase are detected. In order to provide experimental support for this hypothesis of sonic spray-like ionization mechanism, filtered tissue homogenate was analyzed by using a traditional sonic spray ion source.30 The resulting spectra (Figure 4a) were highly similar to the CUSA/SSI results shown in Figure 4b. Further experimental evidence was provided by spraying amino acid glutamine at different pH values. The pH dependence of the intensity of [M + H]+ and [M H] ions were almost identical using the Venturi device and a more traditional sonic spray source and both followed well the theoretical speciation of this amino acid, similar to earlier studies.30 The relative position of the Venturi pump and the atmospheric interface also has a dramatic influence on signal intensity. 7732

dx.doi.org/10.1021/ac201251s |Anal. Chem. 2011, 83, 7729–7735

Analytical Chemistry

Figure 4. (a) Negative ion mode sonic spray ionization spectrum of porcine liver homogenizate using a traditional sonic spray ion source. (b) Negative ion mode CUSA/SSI spectrum of porcine liver.

Optimal signal was obtained when the central axis of the Venturi pump pointed toward the inlet opening. The distance was found to have an optimum at ∼20 25 mm. Although the angle of the Venturi pump and the ion optics only moderately affects the actual signal intensity, the orthogonal setup was chosen in order to minimize the contamination of the instrument. The influence of the inlet capillary temperature on the signal intensity of phospholipid compounds was also investigated. Under 100 C, the intensity of phosphatidyl inositol 38:4 was negligible; however, from 100 to 400 C an increasing tendency was found (data are shown in Figure S-2 in the Supporting Information). Since ionization efficiency improves with increasing temperature, the result is also in agreement with the sonic spray-like ionization phenomenon. The probability of obstruction of the inlet capillary by sticking tissue parts also increases with higher temperature; therefore, 250 C was chosen as a compromise. Optimization of the instrumental parameters (i.e., relative position of the Venturi pump, inlet capillary temperature) allows the continuous operation of the CUSA/SSI-MS system for longer than 4 h. In comparison, during a 2 3 h long surgical intervention, the ultrasonic aspirator is used overall for 20 30 min. Possible obstruction of the CUSA transfer tubes is already wellknown in the surgical practice; therefore, the assistance personnel is prepared for the flushing of the system. The memory effect of the ion source was also tested, and no significant carry over was observed (for details see the Supporting Information). Optimal settings of the ion optics for the Venturi-sonic spray ionization technique were considerably different from the optimal settings for electrospray ionization (ESI) or atmospheric pressure chemical ionization (APCI) ionization. Direct current potentials applied onto the inlet capillary or the tube lens had negligible influence on the signal intensity; however, floating the skimmer electrode at 90 V resulted in an approximately 5-fold signal enhancement. (The skimmer electrode is grounded in the commercial setup.) Detailed testing of the dc ramp profile of ion optics revealed that the dc potential difference between the

ARTICLE

skimmer electrode and the first ion guide has a key importance regarding the described signal enhancement. Because of the cylindrical geometry of the skimmer, there is practically no electric field gradient within the electrode, unless an extremely high potential difference is applied between the skimmer and the following square quadrupole or octapole ion guide. These results suggest that the ions are formed within the skimmer electrode from charged droplets, with close to zero axial kinetic energy. The general sensitivity of the method was determined regarding both full spectral information and the detection of individual compounds. In the former case, various amounts of porcine brain white matter tissue were suspended in 1 mL of saline by the ultrasonic surgical handpiece and aspirated by the Venturi pump device. The resulting spectra were integrated, and the signal-tonoise ratio was plotted for various peaks as a function of suspended tissue weight. The results are shown in Figure S-3 in the Supporting Information. It is concluded that full spectral information can be obtained using 500 μg of tissue or more, while the spectral pattern remains unchanged. Major peaks, however, were detected using as small as 50 μg of tissue sample suspended in 1 mL of liquid. Sensitivity to individual compounds was tested using Substance P and phosphatidyl ethanolamine (15:0/16:0). Because of the presence of an odd-numbered acyl chain, the natural concentration of this lipid species is very low. The limit of detection and dynamic range were tested at three different brain tissue concentrations. The data are shown in Figure S-4 in the Supporting Information. The observed LOD values (corresponding to a 3:1 signal-to noise ratio) were in the low nanogram/milliliter range for the peptide species and in the 100 ng/mL range for the phospholipid compound. These sensitivity values are comparable to those of conventional sonic spray ionization31,32 and allow the detection of tissue components at trace concentration levels. The main objective of the present study was to develop a mass spectrometry-based technique for the intraoperative identification of tissues, which aids the surgeons in removing all malignant tumor tissues while the healthy tissues are preserved. Providing tissue specific data is a prerequisite in this regard for the method. The CUSA/SSI spectra show considerable tissue specificity even in the case of healthy brain samples. Spectra collected from the gray and white matters show characteristic differences (Figure 5d). Similar to the complex lipid data collected by matrix-assisted laser desorption ionization (MALDI) imaging,33 DESI imaging,34 or REIMS techniques, the tissue specificity of the spectra is generally not associated with tissue-specific biomarkers, rather due to the different distribution pattern of a similar set of lipid species. Since individual ion intensities are less informative, we chose to utilize full spectral information for identification. Obviously, identification using mass spectral “fingerprint” requires a database of histologically assigned, authentic spectra. The spectral database was constructed using fresh, ex vivo samples and fresh post-mortem samples. Spectral differences between ex vivo and post-mortem samples (within 24 h following exitus) were found negligible, using rat, canine, and porcine models. A database comprising 284 histologically assigned spectra of human brain tissues and brain tumors was used to develop an identification algorithm. Mass spectra recorded in the m/z 600 1000 were binned and normalized to an integrated total ion count. Appropriate subsets of the database were subjected to principal component analysis in order to decrease the number of dimensions to 60, similar to previously reported applications.20 Figure 5c demonstrates the separation of tissue-specific data groups in the space 7733

dx.doi.org/10.1021/ac201251s |Anal. Chem. 2011, 83, 7729–7735

Analytical Chemistry

ARTICLE

Figure 5. (a) Optical image of healthy human brain tissue with sampling areas (same color code like in part c). (b) Plot of first three LDA parameters of different human brain tissue samples. (c) Plot of the first three PCA dimensions of human brain tissue samples. (d) Typical negative ion mode spectra of human brain gray matter (top) and white matter (bottom).

determined by the first three principal components, and spectra are shown in Figure S-5 in the Supporting Information. Although the data already shows complete separation, supervised linear discriminant analysis was also performed to enhance the segregation of the data groups. The result of the LDA, obviously using only the first 3 LDA parameters, is depicted in Figure 5b. Hence, the proposed quasi-real time data analysis comprises the localization of the actual spectrum in the 60-dimensional LDA space and its classification into the closest histology-specific data group. Since the LDA space is nonorthogonal, the squared Mahalanobis distance function is used as a metric. The time demand for the identification of unknown spectra is