• Open Access

Proteomic analysis of formalin-fixed paraffin-embedded renal tissue samples by label-free MS: Assessment of overall technical variability and the impact of block age

Authors


  • Colour Online: See the article online to view Figs. 1 and 4 in colour.

Correspondence: Professor Rosamonde E. Banks, Cancer Research UK Centre, Leeds Institute of Molecular Medicine, St. James's University Hospital, Beckett Street, Leeds LS9 7TF, UK

E-mail: r.banks@leeds.ac.uk

Fax: +44-113-2429886

Abstract

Purpose

Protein profiling of formalin-fixed paraffin-embedded (FFPE) tissues has enormous potential for the discovery and validation of disease biomarkers. The aim of this study was to systematically characterize the effect of length of time of storage of such tissue blocks in pathology archives on the quality of data produced using label-free MS.

Experimental design

Normal kidney and clear cell renal cell carcinoma tissues routinely collected up to 10 years prior to analysis were profiled using LC-MS/MS and the data analyzed using MaxQuant. Protein identities and quantification data were analyzed to examine differences between tissue blocks of different ages and assess the impact of technical and biological variability.

Results

An average of over 2000 proteins was seen in each sample with good reproducibility in terms of proteins identified and quantification for normal kidney tissue, with no significant effect of block age. Greater biological variability was apparent in the renal cell carcinoma tissue, possibly reflecting disease heterogeneity, but again there was good correlation between technical replicates and no significant effect of block age.

Conclusions and clinical relevance

These results indicate that archival storage time does not have a detrimental effect on protein profiling of FFPE tissues, supporting the use of such tissues in biomarker discovery studies.

Abbreviations
FFPE

formalin-fixed paraffin-embedded

RCC

renal cell carcinoma

1 Introduction

The success of initial studies exploring the possibility of using formalin-fixed paraffin-embedded (FFPE) tissue samples for proteomic analysis has made the use of tissues available in hospital archives in biomarker discovery and validation experiments, a realistic strategy with the potential for significant clinical impact. A number of groups have extracted proteins from FFPE tissues and carried out global protein profiling, most frequently using bottom-up proteomic approaches where proteins are digested with enzymes such as trypsin and analyzed by LC-MS/MS. Label-free approaches and stable isotope tags have been used to facilitate quantification. Good correlation has been reported between datasets obtained from fresh frozen and FFPE tissues in many different tissue types indicating that proteome coverage is not compromised by using FFPE tissue (for a review see [1, 2]). For example, in analysis of 30 000 cells from mouse liver tissue, 776 and 684 proteins were identified from fresh frozen and FFPE tissues, respectively [3]. Furthermore, preliminary comparative analyses have identified differentially expressed proteins that frequently are supported by previously published data or could be confirmed in validation studies using immunohistochemistry [4-10]. Although the number of studies is small and often involving very few samples, overall the data in the literature supports the huge potential of such an approach.

In any biomarker discovery study, the importance of both technical and biological preanalytical variables is now well acknowledged [11]. Understanding the impact of such factors on downstream results and hence incorporating such considerations into experimental design is essential. Studies that systematically evaluate the potential effects of various aspects of sample collection, processing, and storage such as delay to fixation, tissue block size, and fixation length are needed. For FFPE tissue, very few such studies exist (for a review see [12]) but one potentially important factor is the time in storage of the FFPE blocks. This will be critical for many studies, particularly those involving rare diseases or disease subtypes, where the full value of the ability to use archived blocks will be realized.

In the area of renal cancer, two studies using LC-MS/MS analysis of FFPE tissue have shown good correlation of datasets between fresh frozen and FFPE tissues. The first study compared normal kidney tissue and clear cell tumors. A total of 777 proteins were identified, 105 of which showed differential expression between different grades and normal kidney tissue and validation studies generally confirmed the MS data [13]. A second study carried out by our group yielded 250–300 proteins per sample and showed preservation of several known tumor-normal differences [14]. The aim of this study was to build on these findings, carrying out a systematic analysis of normal and malignant FFPE renal tissues to examine the effect of block age/time in storage and the level of overall technical and biological variability. This will provide further information that will be important in future study design and interpretation of results.

2 Materials and methods

2.1 Tissue samples and study design

FFPE tissue blocks were selected from pathology archives in Leeds from patients diagnosed with renal cancer and who had previously provided ethically approved informed consent to tissue samples being used in research. The study adopted a phased approach. In the first part of the study, 16 normal kidney tissue samples were selected with four cases in each of four time periods to assess the effect of block storage: T1–2001, T2–2006, T3–2008/2009, and T4–2010/2011. In all cases, tissue sections were reviewed by a pathologist and samples with no apparent inflammation and good representation of cortex were selected for analysis. Biological variation was minimized by matching age and gender between the groups. In the second part of the study, conventional (clear cell) renal cell carcinoma (RCC) tissue was used, selecting blocks free from necrosis, hemorrhage, or inflammation. Based on the results using normal tissue, the two extreme time periods of approximately 10 years (T1) and less than 18 months (T4) block age were used for RCC tissue with four cases in each of the two groups, again with similar age and gender mixes between the groups. Full details of the patients/samples are provided in Table 1 and examples of the tissues used are shown in Fig. 1.

Table 1. Details of the patients/FFPE tissue blocks included for (A) the normal renal tissue analysis and (B) for the RCC tissue analysis
Block age groupSample numberYear of blockPatient ageGenderGradepTProteins identifieda
  1. a

    Proteins identified indicate the number of proteins present in each individual sample out of the total present in each dataset by the criteria of two significant and one unique peptide. Note that for the RCC samples, technical replicates of samples 3, 6, and 8 contributed to the dataset.

(A) Normal kidney
T11200162M--2131
 2200163M--2159
 3200169F--2154
 4200156M--2206
T25200658F--2188
 6200669M--2046
 7200659M--2083
 8200659F--2227
T39200855F--1793
 10200959M--2101
 11200981F--2180
 12200962M--2266
T413201067F--2033
 14201058F--2041
 15201160M--2141
 16201144M--2186
(B) Conventional (clear cell) RCC
T11200159M31a1778
 2200148M321722
 3200179M31a1945
 4200168M311584
T45201068M31a1482
 6201146M31b1335
 7201152M31a1723
 8201178M321795
Figure 1.

Normal kidney and RCC tissue samples used in the analysis. Haemotoxylin and Eosin (H&E) stained sections of representative normal kidney cortex (upper panel) and conventional (clear cell) RCC (lower panel) are shown. Scale bar = 100 μm.

To assess technical variability and provide some indication of between- and within-study reproducibility, two further independent technical replicates were carried out for each of two normal kidney samples (four and 16) with lysates being prepared and analyzed by LC-MS/MS at the same time as the RCC samples. Similarly, technical replicates of three RCC samples (three, six, and eight) were also included.

2.2 Protein extraction

Sections of 10 μm were cut onto plain glass slides and dried on a heating block at 60°C for 30 min. Sections were dewaxed by incubating in xylene (2 × 5 min), ethanol (5 min), 90% v/v ethanol (5 min), and 70% ethanol (5 min) then an area of tissue equivalent to approximately 5 cm2 was scraped into 400 μL SDS lysis buffer (62.5 mM Tris-HCl pH 6.8, 4% w/v SDS, 10% v/v glycerol, 100 mM DTT). Where needed, macrodissection with a scalpel by comparison with parallel Haemotoxylin and Eosin (H&E) stained sections was undertaken to select appropriate areas. Lysates were mixed by vortexing, incubated at 105°C for 45 min then on ice for 5 min, passed through a needle to shear DNA (if required), microfuged for 10 min at 4°C and stored at −80°C.

2.3 Tryptic digestion

Protein was heated (95°C, 2 min) and a volume equivalent to 250 μg was digested by filter-aided sample preparation [7, 15] using a 30-kDa cutoff Amicon Ultra-0.5 mL centrifugal filter (Millipore). After the application of sample diluted in UA buffer (50 mM ammonium bicarbonate containing 8 M urea) the following additions were then made sequentially to the filter, centrifuging each time, and discarding the flow through: 400 μL UA buffer ×3, 100 μL UA buffer containing 40 mM iodoacetamide (this was incubated at room temperature for 10 min prior to centrifugation), 400 μL UA buffer, 400 μL 50 mM ammonium bicarbonate ×4. After the addition of 45 μL 50 mM ammonium bicarbonate to the filter, Sequencing Grade Modified Trypsin (Promega 18 800 U/mg; 20 μg resuspended in 80 μL 50 mM ammonium bicarbonate) was added to give an enzyme to protein ratio of 1:50 and samples were incubated overnight at 37°C. Peptides were microfuged into the collection tube, 100 μL of water was added to the filter, which was centrifuged and the flow-through combined with the first eluate. Peptides were concentrated using a Savant SPD131DDA SpeedVac (Thermo Scientific) and their concentration determined using a NanoDrop 8000 (Thermo Scientific).

2.4 LC-MS/MS and data analysis

Peptides (2 μg) were separated by online reversed-phase capillary LC and analyzed by electrospray MS/MS. Each sample was analyzed in triplicate. Samples were injected onto a 20-cm reversed-phase fused-silica capillary emitter column made in-house (inner diameter 75 μm, packed with 3.5 μm Kromasil C18 media) using an UltiMate 3000 RSLCnano nanoflow system (Dionex). The LC setup was connected to a linear quadrupole ion trap-orbitrap (LTQ-Orbitrap) Velos mass spectrometer (Thermo Scientific) equipped with a nanoelectrospray ion source (Proxeon). The total acquisition time was 240 min, the major part of the gradient (from 10 to 220 min) being 4–25% ACN in 0.1% formic acid at a flow rate of 400 nL/min. Survey MS scans (scan range of 300–1600 amu) were acquired in the orbitrap with the resolution set to 60 000. Up to 20 most intense ions per scan were fragmented and analyzed in the linear trap. The acquired data from triplicate MS runs for each sample were combined and searched against an International Protein Index (IPI 3.83) human protein sequence database using the MaxQuant computational proteomics platform version 1.2.0.18 [16]. Proteins were identified using the Andromeda peptide search engine [17] integrated into the MaxQuant environment. A decoy version of the IPI human database was used to estimate peptide and protein false discovery rate. The maximum protein and peptide false discovery rates were set to 0.01. Carbamidomethylation of cysteine was set as a fixed modification, with protein N-terminal acetylation and oxidation of methionine as variable modifications, enzyme: trypsin/P, maximum number of missed cleavages: 2. The MaxQuant analysis was carried out in the following three stages: first the initial 16 normal kidney samples, second the normal kidney technical replicates from both parts of the study, and finally the RCC samples. The processed MS data generated by MaxQuant is presented in Supporting Information data Tables 1–3.

2.5 Statistical analysis

The homogeneity of numbers of protein identifications was assessed using the nonparametric Wilcoxon signed-rank and Kruskal–Wallis rank sum tests. The consistency of protein identifications between replicates was visualized using Venn diagrams; a protein identification in a sample was taken as a non-zero summed peptide ion intensity in the MaxQuant proteinGroups file. The Pearson's product moment correlation coefficient was calculated to compare the agreement of biological and technical replicates. The consistency of protein quantitation across different ages of tissue was assessed using a permutation test for each protein in turn. Empirical cutoffs for differential expression analysis were determined based on a Tukey's running median smoother method applied to the 95% quantiles of the normal tissue protein abundance data for five points equally spaced on the logged summed ion intensity scale. All p-values refer to two-sided statistical tests and all analysis were undertaken in the R environment for statistical computing (R Development Core Team, Vienna).

3 Results

In initial work-up studies to assess the success of using FFPE tissue samples in protein profiling studies by label-free MS, proteins were extracted from freshly stored FFPE normal kidney tissue using Rapigest [14] and SDS-based lysis buffers, digested with trypsin and analyzed by LC-MS/MS. The results were compared with matched fresh frozen tissue processed in parallel. In all cases, approximately 2000 proteins were identified with at least two significant and one unique peptide (range, 1889–2153) with no difference being seen between frozen and FFPE tissues or between the different protocols. Based on these results, the simpler approach of using an SDS-based lysis buffer was selected for all downstream work. The study then went on to examine the effect of length of time of block storage on the results obtained.

In the first phase of the study, four normal renal kidney tissues that had been collected in each of four time periods were analyzed: 2001 (T1); 2006 (T2); 2008/2009 (T3), and 2010/2011 (T4). Overall 2663 proteins were identified in the total dataset with the criteria of at least two significant and one unique peptide. The number of proteins per sample ranged from 1793 to 2266 (mean 2121, SD 111; see Table 1). There was no difference in the number of proteins identified between tissue blocks of different ages (Kruskal–Wallis test, p = 0.596). Overall there was good overlap in identities between samples, for example, 1732/2529 (68%) proteins in the T1 samples and 1636/2502 (65%) proteins in the T4 samples were common to all four samples in each group and less than 5% of proteins within any individual sample were unique (see Fig. 2A and Supporting Information Fig. 1). Similarly, there was good overlap in protein identities between samples in the different FFPE block age groups (Fig. 2B). Comparison of quantification using MaxQuant-based label-free analysis using LFQ intensities for proteins that were common between different samples also showed highly statistically significant correlations, with r = 0.854–0.959 for pairwise comparisons between samples within each age group and r = 0.954–0.969 for comparisons between the different age groups (Fig. 3A).

Figure 2.

Overlap in proteins identified between FFPE tissue blocks in different age groups and between technical replicates in normal kidney cortex and RCC. Examples for overlap in protein identities are shown (A) between FFPE tissue samples in block age group T1 (∼10 years), (B) between block age groups (T1–T4 for normal kidney and T1 and T4 for RCC), and (C) between technical replicates. In each case, results are shown for normal kidney cortex (left panel) and RCC (right panel). The full set of figures for all comparisons is presented as Supporting Information Fig. 1.

Figure 3.

Correlation of protein abundance (LFQ intensity) between FFPE tissue blocks in different age groups and between technical replicates in normal kidney cortex and RCC tissue. Examples for correlation of protein abundance are shown (A) between block age groups T1 and T4 and (B) pairwise between technical replicates. In each case, results are shown for normal kidney cortex (left panel) and RCC (right panel). The full set of data is presented as Supporting Information Table 4. In all cases statistical significance was at a level of p < 10−15.

To assess technical variability both within and between experiments, two further lysates were prepared for samples 4 and 16 and analyzed independently of the first experiment. Similar numbers of proteins were identified compared with the initial dataset and there was good overlap in protein identities between the replicates (Fig. 2C). As expected, when assessing quantitation, these showed good correlation compared with each other (r = 0.925 and 0.939; Fig. 3B) and lower but still reasonable correlation with the initial analysis (r = 0.806/0.793 and 0.883/0.878, respectively).

In the second phase of the study examining conventional (clear cell) RCC tissues, four samples in two groups equivalent to the longest (T1) and shortest (T4) storage time of the normal tissues (banked in 2001 and 2010/2011) were examined. Overall, 2516 proteins were identified in the complete dataset, however, a smaller number of proteins were identified in each sample compared with the normal kidney tissue samples, with proteins per sample ranging from 1335 to 1945 (mean 1671, SD 194; see Table 1). This was not a general technical issue as replicates of normal kidney tissue samples reanalyzed at the same time as the RCC samples gave higher numbers of identities. Overlap in identities was also lower than that found in the normal kidney tissue samples with 1114/2370 (47%) and 1011/2173 (47%) of the proteins in T1 and T4, respectively, being common to all four samples (see Fig. 2A and Supporting Information Fig. 1). Good overlap was seen between T1 and T4, reflecting summation of data for four samples in each case (Fig. 2B). Importantly, the number of proteins identified was not statistically different between the different age groups (Wilcoxon rank sum test, p = 0.486) indicating that there was no deterioration of samples with time in storage. When assessing protein quantification, there was relatively good correlation between different samples (r = 0.804–0.961 for pairwise comparisons and 0.885 between T1 and T4; Fig. 3A) but again these were generally slightly lower and there was more variability than was found for the normal kidney tissue, but reflecting patient/tissue heterogeneity rather than age of the tissue block being analyzed. Technical replicates of three individual samples analyzed in parallel showed good reproducibility in terms of protein identities (Fig. 2C) and quantification (r = 0.940, 0.961, and 0.942 for RCC samples 3, 6, and 8, respectively; Fig. 3B).

To analyze total variability within each of the datasets for normal kidney and RCC tissues, the relationship between abundance (LFQ intensity) and pairwise same:same ratios was examined in each case for the T1 and T4 block age groups (Fig. 4). Analysis of the distribution of the normal kidney tissue comparison based on smoothed data gave a fold change of 2.5 at the 95% limit. Similar analysis of tumor tissue gave a fold change of 2.6. Given the lower correlation found in pairwise comparisons of RCCs, this suggests that there may be more extremes in the RCC data, but generally the majority of same:same fold changes have about the same spread as the normal kidney samples. When the different block age groups of normal kidney tissues were compared using a permutation test, 3.3% of proteins showed differences in level with p < 0.05 when all four groups were included; this reduced to 1.8% when considering just T1 and T4. For RCC, 7.1% changed, which is greater than that seen for normal renal tissue, but only slightly greater than the 5% expected by chance.

Figure 4.

Overall reproducibility of protein abundance in FFPE normal kidney cortex and RCC tissue. Scatter plots of abundance (log10 LFQ intensity) versus log2(ratio) for every possible pairwise comparison are plotted for the normal samples (upper panel) and the RCC samples (lower panel) in T1 and T4 block age groups.

Clinical Relevance

The ability to use the large collections of diagnostic formalin-fixed paraffin-embedded (FFPE) tissue samples routinely stored in pathology departments for proteomics-based biomarker discovery and validation has tremendous potential benefits, reducing limitations caused by lack of availability of appropriate samples for analysis. With recent methodological developments, such studies are now being performed but it is important that the effect of preanalytical variables is systematically evaluated to enable optimal experimental design and generation of unbiased results. The aim of this study, using label-free MS to profile normal kidney and renal cell carcinoma (RCC) tissues, was to characterize the effect of length of time of storage of tissue blocks in such archives on the quality of data produced. The overall technical and biological reproducibility of the approach was also investigated to assess the feasibility of using such an approach for the identification of biomarkers for RCC. The results support the use of archival FFPE tissue using blocks up to at least 10-year-old in biomarker discovery experiments.

4 Discussion

This systematic study examining a large number of proteins has shown that archival FFPE tissue stored for up to at least 10 years is suitable for proteomic analysis with no significant effect on the number of proteins identified or their quantitation. However, although on a global level the study gave very promising results, the possibility that particular proteins are selectively affected by archival storage time cannot be ruled out. Analysis of the variability between the normal kidney tissue samples showed good correlation both between different patients and between technical replicates prepared and analyzed together or in independent experiments separated by several months. This indicates that the label-free quantification strategy being employed is robust and that there is relatively high conservation of protein expression in normal kidney tissue between individuals, at least at this level of proteome coverage. A single sample (sample 9) behaved as an outlier both in terms of the number of proteins identified (1793 compared with the average of 2121) and its overall quantitative relationship to other samples in principal components analysis. The reason for this is unclear but it is possible that it may reflect preanalytical factors such as hypoxic interval during surgery, time to processing, or fixation time that would not be detected by H&E review but can influence downstream proteomic analysis. For example, studies looking at fixation time have shown that longer fixation times could be detrimental [18], although other studies showed no such deterioration [19].

Analysis of RCC samples also gave promising results. The difference in number of proteins identified between normal kidney and RCC tissues was not due to technical factors as replicates of the normal samples were extracted and profiled at the same time as the RCCs and these gave higher numbers. Rather it is likely to reflect the greater complexity of normal renal tissue in terms of the different cell types present, as well as the highly vascular nature of clear cell RCCs that results in a significantly larger proportion of the total protein present being derived from blood. In a series of matched samples, albumin comprised 17.0% (SD 9.5%) of RCC tissue by densitometric analysis of Coomassie-stained gels versus 11.3% (SD 2.9%) in normal kidney cortex (our unpublished data). The greater heterogeneity of RCC samples in terms of protein quantification is also not surprising. Despite the central role of loss of Von Hippel-Lindau (VHL) function in clear cell RCCs, there are many other less prevalent changes that also occur in the tumorigenic process, giving rise to significant variability in the tumor proteome. This is clearly illustrated by large-scale sequencing studies carried out in RCC, which have identified mutations in a number of genes in different subsets of tumors [20, 21]. This heterogeneity has obvious consequences for biomarker identification, where studies must use appropriately sized sample groups to ensure that they have sufficient power to allow biomarker identification and must incorporate downstream validation to corroborate initial findings.

Archival time has been examined in other studies and our study builds on these findings and additionally explores aspects of technical variability. Comparison of mouse liver samples stored for 1 wk versus 6 months showed no differences in numbers of proteins identified [22]. In a study of colon adenoma tissue samples stored for 1, 3, 5 and 10 years, a similar number of identified spectra and protein identities (approximately 400 protein groups) were found in triplicate runs of each of three samples per time point [18]; although no quantitative comparisons were undertaken. In a similar study of archived uterine mesenchymal tumors, three groups of three leiomyomas dating from 1990 to 2002 were profiled using capillary isotachophoresis and LC-MS/MS with spectral counting for quantification [23]. A good degree of protein overlap was seen although the oldest group returned fewer identities; however, a single alveolar soft part sarcoma dating from 1980 (i.e. 28 years prior to analysis) that was also included in the study gave a higher number of protein identities indicating that there was not a generalized ongoing loss in tissue quality over time. However, no recent samples were compared and so it was not possible to determine how comparable the results from the newest of these blocks would be to more newly fixed blocks. Our results complement these studies showing no detrimental effect of storage up to at least 10 years. This will allow samples to be selected from much larger banks that is critical for studies involving rare diseases or subtypes such as chromophobe RCC or renal oncocytoma.

The combined biological and technical variability presented here is also very encouraging, although it should be stressed that the samples were carefully selected to match for sex/age and, in the case of RCC, conventional (clear cell) tumors of grade 3 only were included to minimize biological variability. The cutoff values for 95% limits of the ratios of pairwise comparisons of protein intensities when looking at either normal or tumor tissue were approximately 2.5-fold, which can inform decisions about appropriate thresholds to set when considering criteria for differential expression in comparative analyses. In a similar analysis of human liver using capillary isotachophoresis and LC/MS-MS in which 4098 proteins were identified, more than 76% of proteins were shared between three individual samples and a Pearson correlation coefficient of more than 0.97 for spectral counts over a dynamic range of at least 103 was found between two individual samples [19]. A similar result was reported for protein quantitation between the three groups of leiomyoma samples stored in archives for 6, 11, and 18 years prior to analysis in the study described previously [23].

The number of protein identities presented here using a single reverse phase fractionation prior to MS/MS is good compared to other studies in the literature, including other RCC studies carried out to date, many of which profiled less than 1000 proteins. Analysis of the proteins identified in normal tissue using ingenuity pathway analysis (IPA; http://www.ingenuity.com/) indicated that 1549/2541 (61%) localized to the cytoplasm and organelles, 197/2541 (8%) to the extracellular space, 260/2541 (10%) to the plasma membrane, and 355/2541(14%) to the nucleus thus all cellular compartments were represented. This is not dissimilar to the results previously reported for fresh frozen and FFPE normal kidney tissue [14]. With the use of prefractionation, proteome coverage can be further increased as illustrated by the SAX prefractionation of normal and malignant laser capture microdissected FFPE colon samples, which resulted in profiling of more than 6000 proteins with label-free quantification [7].

Taken together, these data indicate that proteins can be successfully extracted from archival FFPE samples routinely collected and stored over long time periods and used to generate extensive datasets with label-free quantification making the use of archival specimens a realistic approach for biomarker identification. This technology will now be taken forward to identify biomarkers for different subsets of RCC.

Acknowledgments

We thank the patients and staff of the Urology and Oncology departments at St James's University Hospital. This work was funded by Cancer Research UK and the Medical Research Council (grant number G0802416).

The authors have declared no conflict of interest.

Ancillary