Correspondence to: Hartmut Schlüter, PhD, Institute of Clinical Chemistry, University Medical Center Hamburg-Eppendorf, Hamburg Germany, Tel.: +49-40741058795, Fax: +49-405741040097, E-mail: email@example.com
To identify molecular features associated with clinico-pathological parameters and TMPRSS2-ERG fusion status in prostate cancer, we employed MALDI mass spectrometric imaging (MSI) to a prostate cancer tissue microarray (TMA) containing formalin-fixed, paraffin-embedded tissues samples from 1,044 patients for which clinical follow-up data were available. MSI analysis revealed 15 distinct mass per charge (m/z)-signals associated to epithelial structures. A comparison of these signals with clinico-pathological features revealed statistical association with favorable tumor phenotype such as low Gleason grade, early pT stage or low Ki67 labeling Index (LI) for four signals (m/z 700, m/z 1,502, m/z 1,199 and m/z 3,577), a link between high Ki67LI for one signal (m/z 1,013) and a relationship with prolonged time to PSA recurrence for one signal (m/z 1,502; p = 0.0145). Multiple signals were associated with the ERG-fusion status of our cancers. Two of 15 epithelium-associated signals including m/z 1,013 and m/z 1,502 were associated with detectable ERG expression and five signals (m/z 644, 678, 1,044, 3,086 and 3,577) were associated with ERG negativity. These observations are in line with substantial molecular differences between fusion-type and non-fusion type prostate cancer. The signals observed in this study may characterize molecules that play a role in the development of TMPRSS2-ERG fusions, or alternatively reflect pathways that are activated as a consequence of ERG-activation. The combination of MSI and large-scale TMAs reflects a powerful approach enabling immediate prioritization of MSI signals based on associations with clinico-pathological and molecular data.
Prostate cancer is a leading cause of cancer-related mortality in males. Worldwide, more than 600,000 men are diagnosed with prostate cancer each year. Although the majority of prostate cancers are detected at early stages as a result of prostate specific antigen (PSA) screening, many patients present with advanced and metastatic cancer at the time of diagnosis. It is hoped that a better understanding of the molecular biology of prostate cancer will help to improve prostate cancer diagnosis and therapy.
Screening methods such as DNA, RNA, and protein arrays have led to the identification of multiple candidate gene alterations that occur in usually small fractions of prostate cancers.[3-5] Most of these alterations have not been further analyzed for clinical significance, partly because the necessary tissue samples with clinical annotation were lacking in the laboratories where the screening methods were utilized. Matrix-assisted laser desorption/ionization (MALDI) mass spectrometric imaging (MSI) on a tissue microarray (TMA) is an interesting concept for screening for clinically relevant biomarkers.[6-8] This approach combines a systematic search for novel biomarkers by mass spectrometry with a simultaneous analysis of the newly found parameters on hundreds of tumors with clinical follow-up data.
In this study, we applied MSI to a TMA containing 1,044 prostate cancers and identified multiple signals associated with clinical outcome in this disease.
Material and Methods
Two TMA sections containing one sample each from 1,044 radical prostatectomy specimens obtained from patients undergoing surgery between 2003 and 2004 at the Department of Urology, University Medical Center Hamburg-Eppendorf were used. Presence of tumor cells in the tissue spots was confirmed in 729 tissue spots by cytokeratin 5/6 analysis using the diagnostic antibody 34βE12 as previously described. 34βE12 (Cytokeratin-903) detects high molecular weight cytokeratins expressed by prostatic basal cells. This antibody is routinely used as a basal cell specific marker active in paraffin-embedded tissue. Complete absence of a basal cell layer in prostatic acinar proliferations is one important diagnostic feature of invasive carcinoma. The remaining 315 samples were excluded from the analysis because the corresponding tissue spot was either missing on the TMA section, lost during slide pretreatment or lacked tumor cells. The pathological parameters of the 729 TMA spots that were eventually included in this study are described in Table 1. Clinical follow-up data were available for all arrayed tumors. Median follow-up time was 63 months ranging from 1 to 109 months. None of the patients received neoadjuvant or adjuvant therapy. PSA values were measured following surgery and recurrence was defined as a postoperative PSA of 0.1 ng/mL and increasing at first of appearance. Immunohistochemical data on ERG and Ki67 labeling index (LI) were available from previous studies analyzing the same TMA block.
Table 1. Pathological and clinical data of the arrayed prostate cancers
No of Patients (%)
Study cohort, on TMA (n = 729)
Biochemical relapse among categories (n = 126)
Note: Number do not always add up 729 in the different categories because of cases with missing data. Abbreviation: AJCC, American Joint Committee on Cancer.
Pretreatment PSA (ng/ml)
pT category (AJCC 2002)
Sample preparation and tryptic digestion
The formalin-fixed paraffin embedded (FFPE) TMA blocks were sliced into 6-µm thick sections using a microtome (Microm HM 355). The analysis samples were deposited onto indium-tin-oxide (ITO)-coated glass slides (Bruker Daltonik, Bremen, Germany), and then dried for 1 hr inside the incubator at 37°C. Paraffin removal was performed by incubating the slides at 65°C for 60 min in the oven. Subsequently, the slides were washed by stepwise immersions xylene of 5-min duration each, following slide immersion in 100% ethanol twice, and once each in 95% ethanol, 70% ethanol and 30 % ethanol. Slides were then incubated at pH 2 (Retrievit TM2, Target Retrieval Solution 10×, Biogenex) at 65°C for 4 hr for antigen retrieval. Trypsin solution was prepared by dissolving 1× 20 µg porcine Trypsin (Promega, Fremont, California) in 200 µL of 40 mM ammonium bicarbonate. The solution was sprayed onto the samples using an ImagePrep device (Bruker Daltonik). Subsequently, the slides were incubated for 3 hr at 37°C to perform tryptic digestion of the proteins.
To obtain digital images of the TMA, the slides were scanned using a flatbed scanner before matrix deposition. A droplet of a peptide standard (Peptide Calibration Standard for MALDI, Bruker Daltonik) was placed at a blank position on the slide prior to matrix application to calibrate the mass spectrometer. The matrix solution was prepared by dissolving 400 mg 2,5-dihydroxybenzoic acid (DHB, Bruker Daltonik) in 13.3 mL 50% methanol/water and 1% trifluoroacetic acid to obtain a final matrix concentration of 30 g/L. The matrix was applied using an ImagePrep device performing the following steps: (i) Initialization 0.2 V, (ii) 30 sec drying, (iii) 6× two cycles 0.05 V, (iv) 6× three cycles 0.1 V, (v). 13× four cycles 0.15 V. The spraying technique enabled full matrix coverage over the entire tissue surface and facilitated co-crystallization of matrix and biomolecules.
Analysis of tissue sections by MALDI mass spectrometry imaging
Mass spectra were acquired using an autoflex speed MALDI-TOF mass spectrometer (Bruker Daltonik) equipped with a smartbeam 2 laser. The mass spectrometer was controlled with the flexControl 3.3/flexImaging 3.0 software package (Bruker Daltonik). Spectra were acquired in positive reflectron mode in a mass per charge (m/z) range of 500–3,680 and a sampling rate of 1.0 GS/sec. Five hundred single spectra per pixel at a laser shot frequency of 1,000 Hz at constant laser power were accumulated. Each spot position was systematically analyzed row by row to sample the entire spot area at a lateral resolution of 50 µm. Data analysis and image generation were done with flexImaging 3.0 and SCiLS Lab (Steinbeis Innovation Center SCiLS, Bremen, Germany). The total ion current (TIC) of each spectrum was used for signal intensity normalization. Following MSI analysis, the slides were rinsed in 100% xylene for 20 min in order to remove the matrix. The slides were subsequently stained with hematoxylin and eosin (H&E) and scanned with an automated slide scanner (Mirax Scan, Zeiss, Oberkochen, Germany) in order to spatially relate the peak intensities to histological structures. Relevant peaks were identified manually, using flexImaging 3.0 and flexAnalysis 3.3 software, taking into consideration adequate distribution over the entire area of the TMA and peptide-specific isotopic signal structure. Only peaks that were present in at least 5% but less than 90% of all tumors were considered for further analyses. This selection was based on the assumption that masses occurring in <5% of prostate cancers represent very rare events without a general role for prostate cancer only, while masses occurring in >90% of samples may most likely represent abundant molecules potentially unrelated to cancer. In addition to the manual search, peaks were found automatically by determining all masses which are co-localized with selected tissue spots. This step has been done using the SCiLS Lab software by calculating the Pearson correlation between the spatial mask given by the selected tissue spots and the intensities of an m/z-signal image of all tissue spots. For each peak found by either method, TMA spots were grouped into “positive” and “negative” for statistical analysis. A TMA spot was classified “positive” for a given peak if the average peak intensity in the individual tissue spot exceeded 45% of the average peak intensity across all tissue spots. Moreover, the peak distribution was related to histological structures on single TMA spots. Only peaks that could repeatedly be linked to epithelial structures were used for statistical analysis. A peak was considered epithelial-related if most (>80%) tissue spots shows that peak had extended areas with epithelial (cancer) cells (example shown in Fig. 1).
Identification of peptides underlying individual m/z signals
To identify tryptic peptides that relate to individual m/z peaks, another section of the TMA block was analyzed by MALDI-FTICR mass spectrometry to obtain highly precise mass values of the peptides. Then, sections of a cancer tissue block, from which a needle core had been taken to initially build the TMA, were digested with trypsin and peptides were extracted to perform MALDI-TOF-TOF and ESI-Q-TOF tandem mass spectrometry, each after separation of the tryptic peptides by liquid chromatography (LC) (detailed description in Supporting Information methods).
For statistical analysis, the JMP 9.0.2 software (SAS Institute Inc, NC) was used. Contingency tables were calculated to study association between presence or absence of selected signals and ERG status as well as other clinico-pathological variables, and the chi-square (Likelihood) test was used to find significant relationships. ANOVA was applied to compare average Ki67LI between groups. Kaplan–Meier curves were generated for PSA recurrence free survival. The log-rank test was applied to test the significance of differences between stratified survival functions. Cox proportional hazards regression analysis was performed to test the statistical independence and significance between pathological, molecular and clinical variables. To perform multivariate statistics, Mathematica 8.0 software (Wolfram Research Inc., IL) was used. A binomial logistic regression model for predicting the ERG status of each tumor was built using the seven peaks found to be significantly correlated with the ERG status. For all tests, p < 0.05 was assumed as the level of significance.
m/z signals discriminating tissue samples
Our MSI analysis identified 32 distinguishable m/z signals that occurred in 5–90% of the spectra of the 729 analyzable tissues samples. Examples of mass spectra are shown in Supporting Information Figure 1. The percentages of peak presence in the spectra are given in Supporting Information Table 1.
A total of 15 of these signals appeared to be associated with epithelial structures, based on the comparison with the H&E stain of the slide. These signals correspond to the following m/z values 605, 616, 644, 678, 700, 899, 976, 1,014, 1,044, 1,199, 1,275, 1,502, 3,071, 3,086 and 3,577. Representative images of MALDI analysis are shown in Figure 1.
Associations with ERG status, clinical phenotype and clinical outcome
Associations with Gleason grade, pT status, Ki67 and ERG results are shown in Table 2 and Figures 2-4. These data show significant associations with ERG status for 7 of 15 analyzed signals (m/z 644, 678, 1,013, 1,044, 1,502, 3,086, 3,577; Fig. 2). These signals were used to build a multivariate regression model for predicting the ERG status of each tumor. The regression model yielded a prediction of the ERG status with an area under the receiver operator characteristics curve (AUC) of 0.74, whereas single-signal prediction yielded receiver operator characteristics curves with AUC values between 0.38 and 0.63 (Supporting Information Fig. 2). Comparison with Gleason grade showed significant associations with 2 out of 15 signals (m/z 1,502, m/z 3,577). Lack of the signal m/z 1,502 was related to early PSA recurrence after prostatectomy (Fig. 3a). This association was independent from the established prognostic parameters tumor stage and Gleason grade (p = 0.0484, Supporting Information Table 2). Two signals were linked to altered cell proliferation as measured by the Ki67LI (m/z 1,013, 1,199; Fig. 4).
Table 2. Associations between m/z signals and prostate cancer phenotype in 729 arrayed prostate cancers. Significant p-values are indicated by bold face type
Reproducibility of MALDI measurements
In order to test whether the outcome of MALDI-IMS is reproducible, we analyzed another section of one of the two TMA blocks that was taken 6 months after the initial analysis and re-calculated the Kaplan–Meier plot for m/z 1,502. Again, we found a significant prognostic difference between tumors with presence and absence of m/z 1,502 (Supporting Information Fig. 4).
Identification by MS/MS analysis
For the identification of the candidate marker peptides, the TMA section was analyzed with MALDI-FTICR yielding highly accurate masses, which were used to verify the peptides identified by MS/MS. Four candidate marker peptides were identified: m/z 976 as AGFAGDDAPR, m/z 1,199 as AVFPSIVGRPR, both occurring in multiple isoforms of actin, m/z 1,502 was identified as IWHHSFYNELR, a specific peptide from aortic or gamma-enteric smooth muscle actin and m/z 1275 was identified as a peptide from a hemoglobin subunit with the sequence LLVVYPWTQR. A summary of the identified peptides and sample MS/MS spectra are given in Supporting Information Table 3 and Supporting Information Figure 3, respectively.
We successfully analyzed a prostate cancer TMA containing samples from 729 patients by MSI. Among the numerous peaks detected on the TMA sections there were 15 signals that could be assigned to epithelial structures. The large number of patients included in this study allowed us to compare the presence or absence of these signals with various clinical, pathological and molecular patient characteristics. Remarkably, 7 out of 15 signals showed a significant association with ERG status, including two signals that were linked to positivity and five signals related to ERG-negative cancer. These signals include m/z 644, 678, 1,013, 1,044, 1,502, 3,086 and 3,577. Using a multivariate regression model, a decent prediction of the ERG status for each tumor was made on the basis of those seven signals. These masses have not been linked to prostate cancer before, but a molecule corresponding to m/z 1,013 has been recently identified as an oligosaccharide in formalin-fixed samples from small intestine.
These findings suggest that molecular differences between ERG-positive and -negative cancers may be quite considerable. This notion is in line with various recent studies demonstrating indeed substantial differences between ERG-positive and -negative cancers in various molecular features, including phosphatase and tensin homolog (PTEN) alterations,[14-17] deletions of 6q or 5q, and 3p3 androgen receptor expression and many other parameters. Given the relationship of many of the identified signals with ERG status it is not surprising that most of these parameters were unrelated to clinical or pathological features. Several earlier studies have demonstrated that ERG status has no impact on tumor-phenotype and clinical outcome, at least in patients that have not undergone anti-androgen treatment. Exceptions to this rule were m/z 1,502 and 3,577, which were equally linked to negative (m/z 3,577) or positive (m/z 1,502) ERG status and, in case of m/z 1,502, also to favorable prognosis. However, given the overall low fraction of samples positive for these markers (<25%) and the inverse associations, these alterations did not overall influence prognosis in the subsets of ERG-positive or ERG-negative cancers (Fig. 3b).
Employing additional high-precision mass spectroscopy analysis we were able to identify the molecules underlying the peaks corresponding to m/z 967, 1,199, 1,502 and 1,275. Since these masses had yielded strong m/z peaks, it was not entirely surprising that they belonged to abundantly expressed proteins including isoforms of actin (m/z 976, 1,199 and 1,502) and hemoglobin (m/z 1,275). However, the prognostic significance observed for m/z 1,502 can hardly be explained by down-regulation of actin. Possible explanations include that the decrease of the signal intensity of the actin peptide with an m/z 1,502 results from an increased conversion into a chemically modified peptide by an increased posttranslational modification reaction or an increase of the concentration of a molecule, which suppresses the signal intensity of the actin peptide with an m/z 1,502 and which escape MALDI-MSI detection. Alternatively, deregulation of actin could possibly result from disturbed activity of actin bundling proteins, which have been suggested to play a role in cancer biology. For example, Di Modugno et al. reported overexpression of a splice variant of the actin bundling protein mena in breast and colon cancers, and Goicoechea et al. found that knockdown of the actin-associated protein palladin inhibits invasive migration of breast cancer cells.
Other observed but yet unidentified signals in this study may characterize molecules that play a role in the development of TMPRSS2-2-ERG fusions or alternatively reflect pathways that are activated as a consequence of ERG activation. It is currently well known that androgen receptor (AR) signaling induces chromatin movements that are one important prerequisite for chromosomal breakage events leading to TMPRSS2-ERG fusions.[15, 16] Although the relevant molecular factors causing chromosomal proximity, DNA breakage and subsequent translocation are largely unknown, it has been reported that AR promotes these events by recruiting proteins that are induced by genotoxic stress, including the activation-induced cytidine deaminase (AID) and LINE-1 repeat-encoded ORF2 endonuclease. As none of the signals associated with ERG status showed a one to one relationship with the presence or absence of ERG protein, it is tempting to speculate that these signals may rather reflect environments facilitating or preventing the development of gene fusions. In case of signals related to pathways activated by ERG an even stronger association to ERG status could be expected. It cannot be excluded, however, that the sensitivity of detecting these signals is limited in formalin-fixed tissues.
Significant associations between MSI signals and prostate cancer phenotype were expected given that a multitude of markers of progression and/or prognosis has been identified before in our prostate prognosis TMA using immunohistochemistry or fluorescence in situ hybridization (FISH) analysis, including for example p53, epidermal growth factor receptor (EGFR), CD10, HER2, PSMA or copy number changes of 8p/8q.
The data of this study demonstrate that the combination of MSI and large-scale TMAs has strong potential. This approach combines screening for unknown molecules in cancers with simultaneous clinical validation in one experiment. Finding associations with clinical outcome, tumor phenotype or important molecular features enables prioritization of specific signals for subsequent protein identification. If dozens of signals are found in MSI of a specific cancer type, it is intuitive that these signals with strongest clinical pathological associations may be most clinically relevant and biologically interesting. Such data have the highest likelihood to find diagnostic applications either using MSI or by other methods after characterization of the signals in question.
Earlier MSI studies have generally used frozen tissue sections as a substrate.[6, 27-42] Several of these studies have demonstrated the utility of MSI for finding and characterizing signals related to histological structures. For example, these studies identified patterns of MSI signals that were associated with overexpression of cancer marker proteins including Her2[27, 28] and MEKK2, or found MSI signatures suitable to distinguish between cancerous and normal[30, 31, 33] or premalignant tissues.
The major disadvantage of using frozen tissues is the limited availability of such tissues for which clinical follow-up data are available. To overcome these problems, various attempts have been made and protocols have been established to expand MALDI imaging to FFPE tissues. Most groups have applied various antigen retrieval strategies to make the fixed tissue accessible for MALDI MSI, including incubation of the slides in buffers with varying pH ranging from 6 to 9 at temperatures between 95°C and 121°C for 20–40 min.[8, 43, 44] In our laboratory, we found that low pH (pH 2) pretreatment at 65°C, if applied for 4 hr, was decisive for reproducibly obtaining distinct peaks in formalin-fixed tissues.
A spatial resolution of 50 µm was used for the MALDI imaging experiment, allowing for approximately 100 data points per 0.6 mm tissue spot. Given that the diameter of a single tumor cell is markedly lower (about 10 µm), and that prostate cancer often consists of small glands with single layers of tumor cells surrounded by stromal cells, the MALDI-MSI measurement cannot always be unequivocally assigned to cancer cells only. To compensate for this shortcoming, we have manually inspected the MALDI spectra of all individual tissue spots and only included m/z signals into further analysis if the tissue spot contained abundant amounts of cancer cells that co-localized with individual m/z peaks. It can be expected that future generations of spectrometers will provide higher resolutions allowing for highly precise measurements. In order to estimate whether the spot architecture impacts MALDI measurement, we analyzed a second TMA section that was taken several months after the initial measurement was performed. In the meantime, the TMA block had been sectioned several times so that the spot composition with respect to tumor and stroma cell had changed markedly. Nevertheless, we could reproduce the prognostic impact of m/z 1,502, demonstrating that not only the MALDI measurements were reproducibly but also that m/z signals can be successfully linked to cancer cells based on a manual inspection of the MALDI image.
Other groups have already demonstrated that biologically relevant data can be obtained by MSI from FFPE tissues. For example, Casadonte et al. and Groseclose et al. reported MALDI imaging signal patterns that were suitable to separate squamous cell cancers from adenocarcinomas of the lung.[7, 8] Lazova et al. described distinct signatures for Spitz naevi and Spitzoid skin cancers. Others identified specific proteins including actin and collagen in ovarian cancers or Grp78 in pancreatic cancers.
In summary, the results of our study demonstrate that MALDI MSI, in combination with the TMA technology and clinical databases, is a powerful tool for simultaneously screening for molecular alterations and assessing their clinical value.
Authors thank Bruker Daltonik (Bremen, Germany) for providing them the MALDI mass spectrometric imaging system. D.T. gratefully acknowledges the financial support of the Bremen Economic Development (WFB, project “3D MALDI-Imaging basic experiment”, grant FUE0485B) and of the European Union Seventh Framework Programme (project “UNLocX”, grant 255931).