Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2023 Dec 11.
Published in final edited form as: Anal Bioanal Chem. 2010 Oct 15;398(7-8):2969–2978. doi: 10.1007/s00216-010-4259-6

Multivariate statistical differentiation of renal cell carcinomas based on lipidomic analysis by ambient ionization imaging mass spectrometry

Allison L Dill 1, Livia S Eberlin 2, Cheng Zheng 3, Anthony B Costa 4, Demian R Ifa 5, Liang Cheng 6, Timothy A Masterson 7, Michael O Koch 8, Olga Vitek 9, R Graham Cooks 10,11
PMCID: PMC10712022  NIHMSID: NIHMS1947726  PMID: 20953777

Abstract

Desorption electrospray ionization (DESI) mass spectrometry (MS) was used in an imaging mode to interrogate the lipid profiles of thin tissue sections of 11 sample pairs of human papillary renal cell carcinoma (RCC) and adjacent normal tissue and nine sample pairs of clear cell RCC and adjacent normal tissue. DESI-MS images showing the spatial distributions of particular glycerophospholipids (GPs) and free fatty acids in the negative ion mode were compared to serial tissue sections stained with hematoxylin and eosin (H&E). Increased absolute intensities as well as changes in relative abundance were seen for particular compounds in the tumor regions of the samples. Multivariate statistical analysis using orthogonal projection to latent structures treated partial least square discriminate analysis (PLS-DA) was used for visualization and classification of the tissue pairs using the full mass spectra as predictors. PLS-DA successfully distinguished tumor from normal tissue for both papillary and clear cell RCC with misclassification rates obtained from the validation set of 14.3% and 7.8%, respectively. It was also used to distinguish papillary and clear cell RCC from each other and from the combined normal tissues with a reasonable misclassification rate of 23%, as determined from the validation set. Overall DESI-MS imaging combined with multivariate statistical analysis shows promise as a molecular pathology technique for diagnosing cancerous and normal tissue on the basis of GP profiles.

Keywords: Ambient ionization, Kidney cancer, Lipidomics, Mass spectrometry, Molecular imaging, Phospholipids, Tissue analysis

Introduction

A substantial effort is underway to understand the biochemical processes occurring in cancerous cells in relation to those of normal cells. This research encompasses the field of lipidomics since cellular maturation and differentiation as well as histological cell type and state of cellular growth is related to lipid composition. For instance, it is known that alterations in the glycerophospholipid (GP) composition of tissues occur in various forms of cancer [1–3]. These GPs and their enzymatic by-products have been proven to be integrally associated with the malignant transformation process in tissue [4, 5]. In particular, increased levels of glycerophosphoinositols (PI) were observed in a number of transformed cells, suggesting that the cellular levels of PIs can be used as a biochemical marker of malignant transformation [6–8]. The expression of phosphatidylserine in the outer leaflet of cell membranes signals the recognition of altered cells, such as cancer cells, to macrophages which then destroy the cells [9–11]. Importantly, GPs are key compounds in the pathways of apoptosis and immune response [12]. Specifically, soluble glycerophosphoserines (PS) stimulate the release of anti-inflammatory mediators, supporting tumor-associated macrophages and blocking antitumor immune responses [13]. GPs can mediate signal transduction pathways in apoptosis, for example tumor-necrosis-factor-related apoptosis-inducing ligand alters the mitochondrial lipids that could be essential for apoptosis [14]. Additionally ether-linked analog GPs can induce selective apoptotic response in tumor cells [15]. While only a few examples are mentioned here of the key roles that lipids play in cellular processes leading to malignancy; lipids are involved in every cellular process and likely play key roles in cancer biochemistry.

Mass spectrometry (MS) is a powerful tool by which identification and characterization of lipids can be achieved. This is especially true in the area of imaging mass spectrometry, which combines the advantages of MS with the ability to determine the spatial distribution of chemical species within a sample. In imaging MS, the sample surface is continuously scanned while ionizing the sample and recording the resulting mass spectra. A mass spectrum is cataloged for each pixel and the data is then compiled to create a chemical image of the entire sample. Imaging MS is an excellent tool in the study of thin tissue sections, allowing the chemical nature of the tissue to be analyzed along with its morphological characteristics [16, 17]. The traditional ionization methods used in imaging MS are matrix-assisted laser desorption ionization [18–20] and secondary ion mass spectrometry [21, 22]. More recently ambient ionization methods have been introduced for tissue imaging. These include laser-ablation electrospray ionization [23–25], surface sampling probe [26], and laser-ablation flowing atmospheric pressure afterglow [27]. These ambient ionization methods have the long-term potential to be translated directly into the operating room to profile tissues during surgery, as has been shown by the technique of rapid evaporative ionization MS [28]. The study presented here employed the spray-based ambient ionization method of desorption electrospray ionization (DESI) [29–31].

In DESI-MS tissue imaging experiments the tissue is fresh and untreated and the most abundant species detected are free fatty acids and GPs. In the first DESI-MS imaging experiment reported on cancerous tissue, different lipid species were found to correlate with the normal, cancerous and transitional regions of the tissue as established by independent pathological examination [32]. This research was extended to bladder cancer (canine) where it was observed that the GP profiles of these tissues allowed cancerous and normal tissue to be distinguished in both negative and positive ion modes [33]. In the course of profiling human prostate cancer, a single molecule, cholesterol sulfate, was found to be elevated in cancerous and precancerous tissue relative to adjacent normal tissue [34]. In a very recent DESI experiment, the GP profiles of human brain gliomas were used to successfully discriminate between different grades of cancer [35].

Here, imaging DESI-MS is applied to human papillary renal cell carcinoma (RCC) and adjacent normal tissue as well as to clear cell RCC tissue and its adjacent normal tissue. The most readily detected ions were due to GPs and fatty acids as confirmed by tandem MS. These fatty acid and GP molecular ions were used to create DESI-MS molecular images showing the spatial distribution of these molecules throughout the cancerous and adjacent normal tissue sections. The samples were diagnosed as cancerous or normal (a rough classification given that normal tissue is found in regions of the tissue classified as cancerous because of the presence of diseased tissue) through the pathological examination of serial hematoxylin and eosin (H&E) stained tissue sections. Multivariate statistical analysis was applied to the DESI-MS data in order to visualize and classify the tissue sections, in line with the goal of developing clinically relevant DESI-MS methods. The DESI-MS and synthetic statistical images derived from the GP profiles correlate with the features of standard histological examination using H&E-stained serial sections. These profiles allow classification of disease; either papillary or clear cell RCC or normal tissue with high statistical accuracy.

Materials and methods

Tissue handling

All tissue samples were handled in accordance with approved Institutional Review Board protocols at Indiana University School of Medicine. All fresh tissue samples were flash frozen in liquid nitrogen and subsequently stored in closed containers at −80 °C. The tissue samples included 11 matched pairs of papillary RCC and adjacent normal tissue and nine matched pairs of clear cell RCC and adjacent normal tissue. Each set of tissue pairs was randomly divided into two sets; a training set which was used to develop the predictive models and a validation set used to test the predictive accuracy of these models. The training set was comprised of manually acquired spectra (excluding the non-informative background regions from the glass slide) from known tumor and normal regions of the tissue samples, as determined by pathological examination. This allowed the predictive models to be established using well-understood data. The validation set used the full mass spectral imaging data, with no manual selection, in order to construct synthetic images from the statistical models.

The tissue samples were sliced into 15-μm-thick sections using a Shandon SME Cryotome cryostat (GMI, Inc., Ramsey, MN, USA) and thaw mounted onto glass slides. They were individually sliced and mounted by hand onto the glass slides. Tissue orientation varied slightly between slides, and therefore H&E-stained optical images were rotated to match approximately the orientation of the DESI-MS images. The tissue sections mounted onto the glass slides were stored in closed containers at −80 °C until analysis. They were allowed to come to room temperature and then dried under nitrogen for approximately 20 min in a desiccator prior to analysis. Serial sections were formalin fixed and subsequently stained using H&E for pathological examination. H&E staining was chosen for diagnosis because examination of H&E-stained tissue sections is still considered the mainstay for pathological diagnosis. Immunostaining was not conducted as it is believed not to be reliable for distinguishing these renal tumors. Molecular genetics such as fluorescence in situ hybridization testing for 3p deletion (for clear cell RCC) and trisomy 7, 17 and loss of Y (for papillary RCC) are more reliable. However, these tests are not readily available in general practice and were not employed [36].

Mass spectrometry

The DESI ion source used was a lab-built prototype, configured as described previously [37]. Optimization involved mechanical adjustments to obtain a small (ca. 250 μm diameter) and uniform spray spot on the sample surface with minimal splashing [38]. The surface moving stage included an XYZ integrated linear stage (Newport, Richmond, CA) and a rotary stage (Parker Automation, Irwin, PA). The DESI spray was positioned 2 mm from the tissue surface at an incident angle of 52° (to the sample surface) and a low (ca 10°) collection angle was used for all of the experiments discussed. The spray solvent for MS acquisition was acetonitrile:water (50:50) with a 5 kV spray voltage applied. Acetonitrile was purchased from Sigma-Aldrich (St. Louis, MO, USA) and water (18.2 MΩ-cm) was from a PureLab ultra system by Elga LabWater (High Wycombe, UK). The nitrogen gas pressure was 150 psi and the solvent flow rate was 1.5 μL/min. In the tissue imaging experiments, the tissue was scanned using a 2D moving stage in horizontal rows, at approximately 219 μm/s, separated by a 250 μm vertical step until the entire tissue sample had been assayed [39]. For example, for sample UH9911–05, the dimension analyzed in the x axis was 19.75 mm, encompassing 78 pixels and the dimension in the y axis was 5.5 mm, encompassing 22 pixels. Although each tissue sample pair varied in size, each sample (approximate size 14 mm×10 mm) could be imaged in approximately 50 min. All experiments were conducted using a LTQ linear ion trap mass spectrometer controlled by XCalibur 2.0 software (Thermo Fisher Scientific, San Jose, CA, USA), with the automatic gain control turned off. An in-house program allowed the conversion of the XCalibur 2.0 raw files into a format compatible with the BioMap (freeware, http://www.maldi-msi.org) software. The individual spectra or pixels that were acquired were assembled into a spatially accurate image using the BIOMAP software. In the BIOMAP images, the same color scale was used within a set of ion images for a particular sample. Therefore, the 100% color shows the highest intensity for the most intense peak in the spectrum and other peaks are color scaled according to their intensity relative to the most intense peak in the spectrum. This allows for comparison of relative levels of species within one pixel as well as their relative intensities between multiple pixels within the tissue sections.

All tissue samples were subjected to DESI-MS imaging analysis. For multivariate statistical analysis representative spectra were acquired for the training set from known tumor and normal regions of each sample, excluding background regions of glass. These spectra were resampled to unit resolution, background corrected and scaled to the median area under the curve prior to statistical analysis. The entirety of the imaging data was used in the multivariate statistical analysis of those samples in the validation set.

Tandem MS product ion spectra, using collision-induced dissociation (CID), were recorded for tissue samples and compared to those of published product ion spectra to confirm molecular identity. For the MS/MS experiments, an isolation window of 1.5m/z units, collision energy of 20% to 40% (manufacturer’s units) and a Mathieu parameter qz value of 0.25 during collisional activation were used.

Multivariate statistics; PLS-DA analysis

Spectra from the training set of tissues were used as input to orthogonal projection to latent structures (O-PLS) [40] treated partial least square discriminate analysis (PLS-DA) [41], implemented in-house in the programming language R. O-PLS is a supervised multivariate procedure, which removes linear combinations of peaks in the spectra that do not contribute to the separation of the disease state of the samples, thereby, yielding a model with better interpretability compared with the regular PLS. In this analysis, the first orthogonal component was filtered out from the original data matrix using the reported eigenvalue approach [40]. The remaining components were used as input to the PLS-DA [42] to derive a formal classification rule of individual spectra (i.e., image pixels). Based on the pixel classification, entire sections were classified as cancerous or normal using a simple “majority rule” (i.e., if the spectra of a majority of the pixels lead to their being classified as cancerous, then the entire tissue sample is pathologically classified as cancer), allowing a determination of classification error rates based on the independent H&E results. The optimal number of components for prediction was obtained using cross-validation [43].

PLS-DA was then applied to the validation set for visualization. Scores of individual pixels were used to create synthetic images of the tissues, which combine the qualitative information from all the ions in the mass spectral GP profiles. The synthetic images were compared to the DESI-MS ion images and to the H&E-stained sections. PLS-DA was also applied to classify the disease status of tissues in the validation set to obtain an accurate estimate of classification error rate.

PLS-DA was selected over alternative multivariate statistical methods, such as principal component analysis due to its ability to better distinguish the characteristic features of disease states by taking into consideration the class labels for each sample as well as allowing for a formal classification rule to be assigned (see Electronic supplementary material for further explanation).

Results and discussion

Tissue sections from 20 different renal cancer patients were analyzed using DESI-MS imaging in the negative ion mode in order to examine the spatial distribution of particular lipids in the different tissue sections. The sample set contained both tumor and adjacent normal tissue for each patient, as well as encompassing two different types of renal cancer, clear cell (nine sample pairs) and papillary (11 sample pairs) RCC. The DESI-MS images representing different GP species were compared to the histological H&E-stained sections to seek correlations between the chemical and histochemical features of tumor and adjacent normal tissue sections. This was done using multivariate statistical analysis allowing for classification of the disease status of the samples.

Overall appearance of the spectra

The mass spectra from the papillary RCC and adjacent normal tissue samples from the training set exhibit clear differences within the GP and fatty acid profiles observed in the negative ion mode spectra in terms of absolute abundances and also within the pattern of relative intensities, as shown in Fig. 1. The same trends were confirmed in the validation set and were observed in the remaining ten tissue sample pairs, as can be observed for the tumor tissue data for four samples shown in Fig. S2. Representative negative ion mode mass spectra from clear cell RCC and adjacent normal tissue samples from the training set are shown in Fig. 2, with the remaining eight sample pairs again following similar trends. While identification of the chemical species present in the mass spectra is not necessary for an empirical correlation of the histopathology and mass spectrometry data, the GPs present at high intensity were identified. This was accomplished using CID tandem MS experiments to record product ion spectra for peaks detected in the tissue samples (see Table S1 for peaks identified and their fragmentation patterns). The data were compared to published fragmentation spectra [44–49] in order to identify the particular lipids involved. Because the internal energy deposition in DESI is similar to that in electrospray ionization, fragmentation patterns recorded using ion traps in conjunction with either of these ionization modes can be used for identification [29, 50]. The main species identified in negative ion mode were glycerophosphoserines (PS) and glycerophosphoinositols (PI). The fatty acids were tentatively identified based on the mass of the molecular anion, [M–H]−. The region from m/z 500 to m/z 600 is comprised of weakly bound dimers of the free fatty acids.

Fig. 1.

Fig. 1

Open in a new tab

Typical full scan negative ion mode mass spectra of human papillary RCC tissue and adjacent normal tissue of sample UH9911–05 from the validation set. a Negative ion mode spectrum of the tumor region. b Negative ion mode spectrum of the normal region

Fig. 2.

Fig. 2

Open in a new tab

Typical full scan negative ion mode mass spectra of human clear cell RCC tissue and adjacent normal tissue of sample MH0204–06 from the validation set. a Negative and ion mode spectrum of the tumor region. b Negative ion mode spectrum of the normal region

Single-ion DESI-MS images

It can be observed from the mass spectra that the lipid profiles are different between both types of RCC and normal tissue; therefore DESI-MS images were used to visualize the data. A DESI-MS image can be constructed from the individual mass spectra, giving a spatial representation of the abundance of ions of a particular m/z value and therefore displaying the distribution of the corresponding molecule within the tissue sample. Even with the modest resolution used for these studies, 250 μm, thousands of mass spectra are collected for each tissue pair; therefore yielding many replicate analyses from a small piece of tissue. The data presented include ion images from only a few molecular species although any ion in the mass spectrum can be used to generate an image. A series of negative ion mode DESI-MS images of papillary RCC and adjacent normal tissue from sample UH9911–05 of the validation set are shown in Fig. 3. The optical image of an adjacent tissue section stained with H&E is presented along with the ion images. Pathological examination was performed on the H&E-stained section and the morphological characteristics of the tissue sections were used to diagnose the tissue as cancerous or normal. Although H&E staining allows diagnosis, no information concerning the chemical nature of the tissue is obtained.

Fig. 3.

Fig. 3

Open in a new tab

Negative ion mode tissue imaging of kidney tissues including areas of papillary RCC and adjacent normal tissue of sample UH9911–05 from the validation set; a ion image of m/z 885.7, PI(18:0/20:4), b ion image of m/z 788.5, PS(18:0/18:1), c ion image of m/z 773.5, PG (18:1/18:1), d Ion image of m/z 810.5, PS(18:0/20:4), e Ion image of m/z 913.5, PI(22:4/18:0), f Ion image of m/z 215.3, FA(12:0), g H&E-stained tissue sections of the tumor tissue and normal and h PLS-DA synthetic image

As seen in the ion images of the papillary RCC tissue, with the cancerous tissue shown on the left and the normal tissue on the right side of each image, the cancerous tissue exhibits increased absolute intensities for the lipid species at m/z 885.7 (PI(18:0/20:4)), m/z 788.5 (PS(18:0/18:1)), m/z 773.5 (PG(18:1/18:1)) and m/z 913.5 (PI(22:4/18:0)). The fatty acid species at m/z 215.3 (FA(12:0)) shows increased absolute intensity in the normal tissue, correlating inversely with the cancerous tissue. Another lipid species at m/z 810.5 (PS(18:0/20:4)) is seen at approximately equal absolute intensities in both the cancerous and normal tissues. The homogenous distribution of the lipid PS(18:0/20:4) across both tumor and normal tissue implies that the intrinsic concentration in tissue is the major factor contributing to signal strength, and that changes in relative ion intensity are not artifacts due to changes in tissue texture, composition or surface topology. These ion images can be visually compared to the H&E-stained tissue sections and cancerous and normal tissue can be distinguished on the basis of these DESI-MS ion images. The trends observed in the training set were confirmed in the validation set, as shown here for one representative sample and are valid for all papillary RCC and adjacent normal tissue sections; additional examples can be seen in Electronic supplementary material Figs. S3–S4.

The same procedure performed for the papillary RCC was performed for the clear cell RCC. Figure 4 shows a series of negative ion mode DESI-MS images of clear cell RCC and adjacent normal tissue from sample MH204–06 of the validation set. The optical image of an adjacent tissue section stained with H&E is presented along with the ion images. Identical lipid species to those chosen for the papillary RCC samples were chosen for the negative ion images of the clear cell RCC samples, those as m/z 885.7 (PI(18:0/20:4)), m/z 788.5 (PS(18:0/18:1)), m/z 773.5 (PG (18:1/18:1)), m/z 913.5 (PI(22:4/18:0)), m/z 810.6 (PS(18:0/20:4)), and a fatty acid at m/z 215.3 (FA(12:0)). These images can be visually compared to the H&E-stained tissue sections. The trends shown here for one representative sample are true for all nine clear cell RCC and adjacent normal tissue sections; an additional example can be seen in Electronic supplementary material Fig. S5. Unlike the papillary RCC samples, there are no clear visual trends between the absolute or relative intensities of the lipid species for either the cancerous or normal tissue. On the basis of these ion images alone, a clear diagnosis cannot be obtained, highlighting the need for a multivariate statistical analysis using the entirety of the mass spectral data.

Fig. 4.

Fig. 4

Open in a new tab

Negative ion mode tissue imaging of kidney tissues including areas of clear cell RCC and adjacent normal tissue of sample MH0204–06 from the validation set; a ion image of m/z 885.7, PI (18:0/20:4), b ion image of m/z 788.5, PS(18:0/18:1), c ion image of m/z 773.5, PG(18:1/18:1), d ion image of m/z 810.5, PS (18:0/20:4), e ion image of m/z 913.5, PI(22:4/18:0), f Ion image of m/z 215.3, FA(12:0), g H&E-stained tissue sections of the tumor tissue and normal and h PLS-DA synthetic image

Classification of paired tissue samples using multivariate statistical analysis

While individual ions can be used to distinguish tumor from normal tissue, the use of multiple ions should improve the strength of the diagnosis. This is accomplished through the use of multivariate statistics to reduce the high-dimensional data that is acquired for each sample [51].

Multivariate statistical analysis was used to expand the analysis and seek a formal classification rule for disease diagnosis for both types of RCC. PLS-DA allows formal classification of the disease status of individual spectra (i.e., image pixels) to be achieved (in contrast to what is possible with principal component analysis, see discussion in the Electronic supplementary material). Figures S6a–f and S7a–f show the PLS-DA score plot and loading plots of the first three components for both papillary and clear cell RCC compared to matched normal tissues in negative ion mode of the training set indicating that the full spectra carry sufficient disease related information to discriminate between cancerous and normal tissues. The loading plots of the first three components for both papillary and clear cell RCC show the relative importance of various m/z values to the separation. The primary effects contributing to the discrimination between cancerous and normal tissues can be seen in the peaks present in the loading plots, with the presence of those peaks in the positive y-quadrant and with the absence of those peaks in the negative y-quadrant contributing to the separation.

In the case of papillary RCC, a PLS-DA model with ten components achieves the best prediction performance (tenfold cross-validation Q2=0.856), and produces a cross-validation error rate of 1.3% for classification between this type of RCC and normal samples. This model was then applied to the five tissue pairs in the validation set and a misclassification error rate of 14.3% was obtained for papillary RCC samples. In the case of clear cell RCC, a PLS-DA model with 12 components achieves the best prediction performance (tenfold cross-validation Q2=0.914), and produces a cross-validation error rate of 0% for classification. This model was then applied to the four tissue pairs in the validation set and a misclassification error rate of 7.8% was obtained for the clear cell RCC samples. A clear separation of both types of RCC from adjacent normal tissue was achieved using PLS-DA.

Visualization of paired tissue samples using multivariate statistical analysis

For complete visualization, the DESI-MS ion images were compared to the synthetic PLS-DA images from the validation set and to the H&E-stained sections. The synthetic images combine the qualitative information obtained from all of the ions seen in the mass spectral GP profiles. Figures 3 and 4 show the overall results of the DESI-MS, H&E-stained and PLS-DA visualization procedures for the validation set samples UH9911–05 and MH0204–06, respectively. Results for three other samples are shown in the Electronic supplementary material (Figs. S3–S5). PLS-DA outputs a continuous score with a larger value indicating stronger evidence for cancer, and the color scheme was calibrated so that “white” corresponds to the cutoff value, with red indicating a great probability of the presence of cancer and blue indicating a greater probability of normal. As seen in Fig. 3 for the papillary RCC sample particular lipid species present in DESI-MS ion images allow a distinction to be drawn between cancerous and normal tissue (results for two additional samples are shown in Figs. S3 and S4). This same distinction can be made using the PLS-DA synthetic image, which diagnoses cancerous and normal tissue on the basis of the full mass spectral information. This diagnosis is confirmed by pathological examination of the H&E-stained sections. In the case of clear cell RCC, the DESI-MS ion images did not allow for a clear diagnosis of tumor versus normal tissue, as shown in Fig. 4 (results for one additional validation set sample are shown in Fig. S5). However, when a multivariate statistical analysis was employed using the entire GP profiles from each sample, a clear distinction between tumor and normal tissue could be made. This is visualized in the synthetic PLS-DA images shown in Fig. 4h with the tumor tissue in red and the normal tissue in blue, a result which is confirmed by pathological examination of the H&E-stained tissue sections. In the case of clear cell RCC, the importance of a multivariate statistical analysis for use with DESI-MS imaging data is clearly demonstrated, without which this type of RCC could not have been distinguished from normal tissue.

Classification of carcinoma types using multivariate statistical analysis

In addition to the statistical analysis of the paired tumor and normal samples, multivariate analysis was also used to determine if a distinction could be drawn between papillary RCC, clear cell RCC, and the pooled normal tissue samples. To this end, a PLS-DA model was built using samples from all of the carcinoma types in the training set as input. As shown in the score plots in Fig. 5, the three sample classes are separated from each other on the basis of their DESI-MS lipid profiles. A model with 18 components gives the best Q2=0.835 and also produces a tenfold cross-validation error rate of 2.2%. This model was applied to the validation set resulting in a misclassification error rate of 23.7%. This demonstrates that on the basis of the GP profiles the two types of human RCC can be distinguished not only individually from their corresponding normal paired samples, but from each other and from the pooled normal tissue samples through statistical analysis.

Fig. 5.

Fig. 5

Open in a new tab

PLS-DA analysis of papillary and clear cell RCC and normal tissues. a–c Score plots for the first three components for PLS-DA of the training set. The axes LV1, LV2, and LV3 correspond respectively to the scores of the first three components for each sample after applying PLS. d–f Loading plots for the first three components in PLS-DA from the training set. The plots illustrate the m/z values and their relative importance in the PLS-DA analysis

Conclusion

It has been demonstrated that DESI-MS imaging of multiple GP and fatty acids along with multivariate statistics can be used to distinguish between human kidney cancers and adjacent normal tissue. While papillary RCC can be distinguished from paired normal tissue on the basis of the individual DESI-MS ion images, the use of multivariate statistical methods increases the confidence of this diagnosis. The multivariate statistical analysis played a critical role in distinguishing clear cell RCC from normal tissue since the differences between the tissues were not evident from the DESI-MS ion images alone. In addition, both types of kidney cancer are distinct both from each other and from the pooled normal tissues, indicating that specific lipid patterns are present in each type of cancer which are also different from the normal tissue. This is the first time any ambient ionization method has been used to discriminate between different types of cancer and between normal tissues. When compared to the potential of a single marker compound, the GP profiles determined from DESI-MS imaging data provides the sensitivity and specificity required for disease diagnosis. DESI-MS has the potential for use in a diagnostic capability based on the lipid profiles and intensities found in tissue samples, as shown here for human RCC. The predictive model generated through this research can be validated and in the future be applied to a larger set of unknown samples to further test its diagnostic capabilities. With further development and validation, DESI-MS imaging could provide a complementary technique to traditional H&E and immunostaining methods by providing a reproducible and faster diagnosis that currently has error rates similar to those obtained by staining methods [52].

Supplementary Material

SI

Acknowledgments

We thank the Purdue University Center for Cancer Research and its director, Timothy Ratliff, for assistance in obtaining the human kidney cancer samples. This work was supported by the National Institutes of Health (Grant 1 R21 EB00 9459–01).

Footnotes

Electronic supplementary material The online version of this article (doi:10.1007/s00216-010-4259-6) contains supplementary material, which is available to authorized users.

Contributor Information

Allison L. Dill, Department of Chemistry and Center for Analytical, Instrumentation Development, Purdue University, West Lafayette, IN 47907, USA

Livia S. Eberlin, Department of Chemistry and Center for Analytical, Instrumentation Development, Purdue University, West Lafayette, IN 47907, USA

Cheng Zheng, Department of Statistics, Purdue University, West Lafayette, IN 47907, USA.

Anthony B. Costa, Department of Chemistry and Center for Analytical, Instrumentation Development, Purdue University, West Lafayette, IN 47907, USA

Demian R. Ifa, Department of Chemistry and Center for Analytical, Instrumentation Development, Purdue University, West Lafayette, IN 47907, USA

Liang Cheng, Department of Pathology and Laboratory Medicine, Indiana University School of Medicine, Indianapolis, IN 46202, USA.

Timothy A. Masterson, Department of Urology, Indiana University School of Medicine, Indianapolis, IN 46202, USA

Michael O. Koch, Department of Urology, Indiana University School of Medicine, Indianapolis, IN 46202, USA

Olga Vitek, Department of Statistics, Purdue University, West Lafayette, IN 47907, USA.

R. Graham Cooks, Department of Chemistry and Center for Analytical, Instrumentation Development, Purdue University, West Lafayette, IN 47907, USA; Department of Chemistry, Purdue University, West Lafayette, IN 47907, USA.

References

  • 1.Athar M (2002) Oxidative stress and experimental carcinogenesis. Indian J Exp Biol 40(6):656–667 [PubMed] [Google Scholar]
  • 2.Sakai K, Okuyama H, Yura J, Takeyama H, Shinagawa N, Tsuruga N, Kato K, Miura K, Kawase K, Tsujimura T, Naruse T, Koike A (1992) Composition and turnover of phospholipids and neutral lipids in human breast-cancer and reference tissues. Carcinogenesis 13(4):579–584 [DOI] [PubMed] [Google Scholar]
  • 3.Shimma S, Sugiura Y, Hayasaka T, Hoshikawa Y, Noda T, Setou M (2007) MALDI-based imaging mass spectrometry revealed abnormal distribution of phospholipids in colon cancer liver metastasis. J Chromatogr B: AnalTechnol Biomed Life Sci 855(1):98–103 [DOI] [PubMed] [Google Scholar]
  • 4.Aboagye EO, Bhujwalla ZM (1999) Malignant transformation alters membrane choline phospholipid metabolism of human mammary epithelial cells. Cancer Res 59(1):80–84 [PubMed] [Google Scholar]
  • 5.Glunde K, Jie C, Bhujwalla ZM (2004) Molecular causes of the aberrant choline phospholipid metabolism in breast cancer. Cancer Res 64(12):4270–4276 [DOI] [PubMed] [Google Scholar]
  • 6.Alonso T, Morgan RO, Marvizon JC, Zarbl H, Santos E (1988) Malignant transformation by ras and other oncogenes produces common alterations in inositol phospholipid signaling pathways. Proc Natl Acad Sci USA 85(12):4271–4275 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 7.Alonso T, Santos E (1990) Increased intracellular glycerophosphoinositol is a biochemical marker for transformation by membrane-associated and cytoplasmic oncogenes. Biochem Biophys Res Commun 171(1):14–19 [DOI] [PubMed] [Google Scholar]
  • 8.Valitutti S, Cucchi P, Colletta G, Di Filippo C, Corda D (1991) Transformation by the k-ras oncogene correlates with increases in phospholipase a2 activity, glycerophosphoinositol production and phosphoinositide synthesis in thyroid cells. Cell Signal 3(4):321–332 [DOI] [PubMed] [Google Scholar]
  • 9.Yamaji-Hasegawa A, Tsujimoto M (2006) Asymmetric distribution of phospholipids in biomembranes. Biol Pharm Bull 29(8):1547–1553 [DOI] [PubMed] [Google Scholar]
  • 10.Utsugi T, Schroit AJ, Connor J, Bucana CD, Fidler IJ (1991) Elevated expression of phosphatidylserine in the outer-membrane leaflet of human tumor-cells and recognition by activated human blood monocytes. Cancer Res 51(11):3062–3066 [PubMed] [Google Scholar]
  • 11.Zwaal RFA, Comfurius P, Bevers EM (2005) Surface exposure of phosphatidylserine in pathological cells. Cell Mol Life Sci 62(9):971–988 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Chaurio RA, Janko C, Munoz LE, Frey B, Herrmann M, Gaipl US (2009) Phospholipids: key players in apoptosis and immune regulation. Molecules 14(12):4892–4914 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Kim R, Emi M, Tanabe K, Arihiro K (2006) Tumor-driven evolution of immunosuppressive networks during malignant progression. Cancer Res 66(11):5527–5536 [DOI] [PubMed] [Google Scholar]
  • 14.Sandra F, Esposti MD, Ndebele K, Gona P, Knight D, Rosenquist M, Khosravi-Far R (2005) Tumor necrosis factor-related apoptosis-inducing ligand alters mitochondrial membrane lipids. Cancer Res 65(18):8286–8297 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Mollinedo F, Gajate C, Martin-Santamaria S, Gago F (2004) Et-18-och3 (edelfosine): a selective antitumour lipid targeting apoptosis through intracellular activation of fas/cd95 death receptor. Curr Med Chem 11(24):3163–3184 [DOI] [PubMed] [Google Scholar]
  • 16.Wiseman JM, Ifa DR, Song Q, Cooks RG (2006) Tissue imaging at atmospheric pressure using desorption electrospray ionization (DESI) mass spectrometry. Angew Chem Int Ed 45(43):7188–7192 [DOI] [PubMed] [Google Scholar]
  • 17.Chaurand P, Cornett DS, Caprioli RM (2006) Molecular imaging of thin mammalian tissue sections by mass spectrometry. Curr Opin Biotechnol 17(4):431–436 [DOI] [PubMed] [Google Scholar]
  • 18.Cornett DS, Reyzer ML, Chaurand P, Caprioli RM (2007) MALDI imaging mass spectrometry: molecular snapshots of biochemical systems. Nat Meth 4(10):828–833 [DOI] [PubMed] [Google Scholar]
  • 19.Xu BJ, Gonzalez AL, Kikuchi T, Yanagisawa K, Massion PP, Wu H, Mason SE, Olson SJ, Shyr Y, Carbone DP, Caprioli RM (2008) MALDI-MS derived prognostic protein markers for resected non small cell lung cancer. Proteomics Clin Appl 2(10–11):1508–1517 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Doroshenko VM, Laiko VV, Taranenko NI, Berkout VD, Lee HS (2002) Recent developments in atmospheric pressure MALDI mass spectrometry. Int J Mass Spectrom 221(1):39–58 [Google Scholar]
  • 21.Fletcher JS, Lockyer NP, Vaidyanathan S, Vickerman JC (2007) TOF-SIMS 3D biomolecular imaging of Xenopus laevis oocytes using buckminsterfullerene (c-60) primary ions. Anal Chem 79(6):2199–2206 [DOI] [PubMed] [Google Scholar]
  • 22.Winograd N (2003) Prospects for imaging TOF-SIMS: from fundamentals to biotechnology. Appl Surf Sci 203:13–19 [Google Scholar]
  • 23.Nemes P, Barton AA, Vertes A (2009) Three-dimensional imaging of metabolites in tissues under ambient conditions by laser ablation electrospray ionization mass spectrometry. Anal Chem 81(16):6668–6675 [DOI] [PubMed] [Google Scholar]
  • 24.Nemes P, Barton AA, Li Y, Vertes A (2008) Ambient molecular imaging and depth profiling of live tissue by infrared laser ablation electrospray ionization mass spectrometry. Anal Chem 80(12):4575–4582 [DOI] [PubMed] [Google Scholar]
  • 25.Nemes P, Vertes A (2007) Laser ablation electrospray ionization for atmospheric pressure, in vivo, and imaging mass spectrometry. Anal Chem 79(21):8098–8106 [DOI] [PubMed] [Google Scholar]
  • 26.Kertesz V, Ford MJ, Van Berkel GJ (2005) Automation of a surface sampling probe/electrospray mass spectrometry system. Anal Chem 77(22):7183–7189 [DOI] [PubMed] [Google Scholar]
  • 27.Shelley JT, Ray SJ, Hieftje GM (2008) Laser ablation coupled to a flowing atmospheric pressure afterglow for ambient mass spectral imaging. Anal Chem 80(21):8308–8313 [DOI] [PubMed] [Google Scholar]
  • 28.Schafer KC, Denes J, Albrecht K, Szaniszlo T, Balog J, Skoumal R, Katona M, Toth M, Balogh L, Takats Z (2009) In vivo, in situ tissue analysis using rapid evaporative ionization mass spectrometry. Angew Chem Int Ed 48(44):8240–8242 [DOI] [PubMed] [Google Scholar]
  • 29.Takats Z, Wiseman JM, Gologan B, Cooks RG (2004) Mass spectrometry sampling under ambient conditions with desorption electrospray ionization. Science 306(5695):471–473 [DOI] [PubMed] [Google Scholar]
  • 30.Dill AL, Ifa DR, Manicke NE, Zheng OY, Cooks RG (2009) Mass spectrometric imaging of lipids using desorption electrospray ionization. J Chromatogr B Anal Technol Biomed Life Sci 877(26):2883–2889 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31.Kertesz V, Van Berkel GJ (2008) Improved imaging resolution in desorption electrospray ionization mass spectrometry. Rapid Commun Mass Spectrom 22(17):2639–2644 [DOI] [PubMed] [Google Scholar]
  • 32.Wiseman JM, Puolitaival SM, Takats Z, Cooks RG, Caprioli R (2005) Mass spectrometric profiling of intact biological tissue by using desorption electrospray ionization. Angew Chem Int Ed 44(43):7094–7097 [DOI] [PubMed] [Google Scholar]
  • 33.Dill AL, Ifa DR, Manicke NE, Costa AB, Ramos-Vara JA, Knapp DW, Cooks RG (2009) Lipid profiles of canine invasive transitional cell carcinoma of the urinary bladder and adjacent normal tissue by desorption electrospray ionization imaging mass spectrometry. Anal Chem 81(21):8758–8764 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Eberlin LS, Dill AL, Costa AB, Ifa DR, Cheng L, Masterson T, Koch M, Ratliff TL, Cooks RG (2010) Cholesterol sulfate imaging in human prostate cancer tissue by desorption electrospray ionization mass spectrometry. Anal Chem 82(9):3430–3434 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Eberlin LS, Dill AL, Golby AJ, Ligon KL, Wiseman JM, Cooks RG, Agar NYR (2010) Discrimination of human astrocytoma subtypes by lipid analysis using desorption electrospray ionization imaging mass spectrometry. Angew Chem Int Ed 49(34):5953–5956 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36.Cheng LA, Williamson SR, Zhang SB, MacLennan GT, Montironi R, Lopez-Beltran A (2010) Understanding the molecular genetics of renal cell neoplasia: implications for diagnosis, prognosis and therapy. Expert Rev Anticancer Ther 10(6):843–864 [DOI] [PubMed] [Google Scholar]
  • 37.Manicke NE, Kistler T, Ifa DR, Cooks RG, Ouyang Z (2009) High-throughput quantitative analysis by desorption electrospray ionization mass spectrometry. J Am Soc Mass Spectrom 20(2):321–325 [DOI] [PubMed] [Google Scholar]
  • 38.Eberlin LS, Ifa DR, Wu C, Cooks RG (2010) Three-dimensional visualization of mouse brain by lipid analysis using ambient ionization mass spectrometry. Angew Chem Int Ed 49(5):873–876. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 39.Ifa DR, Wiseman JM, Song QY, Cooks RG (2007) Development of capabilities for imaging mass spectrometry under ambient conditions with desorption electrospray ionization (DESI). Int J Mass Spectrom 259:8–15 [Google Scholar]
  • 40.Trygg J, Wold S (2002) Orthogonal projections to latent structures (O-PLS). J Chemom 16(3):119–128. [Google Scholar]
  • 41.Team RDC (2009) R: a language and environment for statistical computing (trans: Book citations: Books without editor: E. Wingender GRiE, VCH, Weinheim, 1993, p. 215. Books with editor: T. D. In: Atwood JL, Davies JED, MacNicol DD, Vögtle F, Suslick KS (eds) Tullius in Comprehensive supramolecular chemistry, vol. 5, Pergamon, Oxford, 1996, pp 317–343.). Vienna, Austria [Google Scholar]
  • 42.Wehrens RM, Mevik BH (2007) PLS: partial least squares regression (PLSR) and principal component regression (PCR) R package versiion 21–0 http://meviknet/work/software/plshtml [Google Scholar]
  • 43.De Meyer T, Sinnaeve D, Van Gasse B, Tsiporkova E, Rietzschel ER, De Buyzere ML, Gillebert TC, Bekaert S, Martins JC, Van Criekinge W (2008) NMR-based characterization of metabolic alterations in hypertension using an adaptive, intelligent binning algorithm. Anal Chem 80(10):3783–3790 [DOI] [PubMed] [Google Scholar]
  • 44.Pulfer M, Murphy RC (2003) Electrospray mass spectrometry of phospholipids. Mass Spectrom Rev 22(5):332–364 [DOI] [PubMed] [Google Scholar]
  • 45.Manicke NE, Wiseman JM, Ifa DR, Cooks RG (2008) Desorption electrospray ionization (DESI) mass spectrometry and tandem mass spectrometry (MS/MS) of phospholipids and sphingolipids: ionization, adduct formation, and fragmentation. J Am Soc Mass Spectrom 19(4):531–543 [DOI] [PubMed] [Google Scholar]
  • 46.Hsu FF, Turk J (2000) Characterization of phosphatidylinositol, phosphatidylinositol-4-phosphate, and phosphatidylinositol-4, 5-bisphosphate by electrospray ionization tandem mass spectrometry: a mechanistic study. J Am Soc Mass Spectrom 11(11):986–999 [DOI] [PubMed] [Google Scholar]
  • 47.Hsu FF, Turk J (2008) Structural characterization of unsaturated glycerophospholipids by multiple-stage linear ion-trap mass spectrometry with electrospray ionization. J Am Soc Mass Spectrom 19(11):1681–1691 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 48.Hsu FF, Turk J (2005) Studies on phosphatidylserine by tandem quadrupole and multiple stage quadrupole ion-trap mass spectrometry with electrospray ionization: structural characterization and the fragmentation processes. J Am Soc Mass Spectrom 16(9):1510–1522 [DOI] [PubMed] [Google Scholar]
  • 49.Murphy RC (1993) Mass spectrometry of lipids. Plenum, New York [Google Scholar]
  • 50.Nefliu M, Smith JN, Venter A, Cooks RG (2008) Internal energy distributions in desorption electrospray ionization (DESI). J Am Soc Mass Spectrom 19(3):420–427 [DOI] [PubMed] [Google Scholar]
  • 51.Clarke R, Ressom HW, Wang AT, Xuan JH, Liu MC, Gehan EA, Wang Y (2008) The properties of high-dimensional data spaces: implications for exploring gene and protein expression data. Nat Rev Cancer 8(1):37–49 [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 52.Lechevallier E (2007) Core biopsy of solid renal masses under ct guidance. Eur Urol Suppl 6(8):540–543 [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

SI
NIHMS1947726-supplement-SI.pdf (869.8KB, pdf)

RESOURCES