Ethics clearance was obtained from the Health Research Ethics Committee of Stellenbosch University (ethics reference number: S17/10/203) and Biomedical Science Research Ethics Committee of the University of the Western Cape (ethics reference number: BM17/7/15). All patient specimens were anonymised before being archived for long-term storage and before they were accessed for the study. Patient consent was not required since it was a retrospective study using archival tissues.

This study conducted from August 2017 to July 2019, included retrospectively chosen human colorectal resection specimens acquired from the department of Anatomical Pathology at Tygerberg Hospital in Western Cape, South Africa. The specimens were preserved as FFPE blocks when the tissue was resected and archived between January 2016 and December 2017. Due to retrospective collection of the samples, the exact pre-analytical factors, such as the handling, fixation times and conditions, and storage conditions, were unknown and could not be accounted for. Table 1 shows the details of the three patient cases selected.

To ensure tissue quality and comparability, a pathologist reviewed the patient tissue sections after haematoxylin and eosin staining to select only specimens that had carcinomas with more than 90% viable tumour nuclei (Figure 1). Patient samples were also classified and diagnosed with low-grade or high-grade colorectal carcinoma after haematoxylin and eosin staining.

To overcome the effects of formaldehyde cross-linking, we opted to combine protein extraction techniques that employed the use of antigen retrieval, strong detergent concentration, as well as a synthetic polymer for protein precipitation (PEG 20 000). For protein purification before LC-MS/MS analysis, we used the Single-Pot Solid-Phase-enhanced Sample Preparation (SP3)^19,20 method, which ensures minimal sample loss during processing and was also found to be highly sensitive, therefore requiring less starting material (tissue).^12,19,20

The equivalence of 23 mm³ of manually micro-dissected FFPE tumour tissue was cut and processed for each patient case (Figure 2). Protein was extracted using a solution that consisted of 50 mM of ammonium bicarbonate (pH 8.0) and 2% sodium dodecyl sulphate (SDS) and either with or without the addition of 0.5% PEG. To further determine protein extraction buffer efficiency, the sample pellets were also assessed for residual proteins that were not extracted in the initial extraction. In total, 12 samples were analysed, including the WCPLs as well as the sample pellets (Figure 2).

The method used for sample processing, protein extraction and protein yield quantification was modified from the protocols used by Scicchitano²¹ and Wiśniewski²² and previously described in more detail by Rossouw.¹² Both WCPLs and sample pellets were subsequently processed by the hydrophilic interaction liquid chromatography (HILIC)/SP3 magnetic bead digestion method,¹⁹ before LC-MS/MS analysis (Figure 2).

The MagReSyn® (ReSyn Biosciences, Edenvale, Gauteng, South Africa) HILIC/SP3 method (using on-bead digestion) was used for protein purification and tryptic digestion (peptide generation) prior to LC-MS/MS analysis. The method was modified from the protocol used by Hughes¹⁹ and previously described in more detail by Rossouw.¹²

Mass spectrometry analysis of each sample’s peptides was performed using the Q-Exactive quadrupole-Orbitrap (Thermo Fisher Scientific, Waltham, Massachusetts, United States), which was coupled with a Dionex Ultimate 3000 nano-UPLC system as described before by Rossouw.¹² Using Xcalibur^TM (version 4.2) (Thermo Fisher, Waltham, Massachusetts, United States), spectral data was collected in a data-dependent manner and details are shown in Supplementary document – Table S1.

The raw spectral data were converted into ‘mascot generic format’ (Matrix Science, London, United Kingdom), which is a standard format used for tandem MS data that converts the raw data into a simpler format for subsequent database searches, using msConvert (ProteoWizard, Palo Alto, California, United States).²³ X!Tandem (version 2015.12.15.2)²⁴ (The Global Proteome Machine Organization, Winnipeg, Manitoba, Canada), MS Amanda (version 2.0.0.9706)²⁵ (Protein Chemistry Facility IMP/IMBA/GMI, Vienna, Austria), and MS-GF+ (version 2018.04.09)²⁶ (Pacific Northwest National Laboratory, Richland, Washington, United States) were used to identify peak lists from MS/MS spectra.¹² SearchGUI (version 3.3.3)²⁷ (Computational Omics and Systems Biology Group, Ghent University, Gent-Zwijnaarde, Belgium) was used to allow for simultaneous searches. A concatenated target-decoy²⁸ version of the Homo sapiens (73101, > 99.9%), Sus scrofa (1, < 0.1%) complement of the UniProtKB²⁹ reference proteome (UP000005640; 9606-Homo sapiens) (version downloaded on 29/10/2018) was used for protein identifications. SearchGUI generated the decoy sequences. The identification settings are shown in Table 2 and the certificate of analysis lists all algorithms settings used and validation thresholds (Supplementary document 3 – File S1). PeptideShaker (version 1.16.31)³⁰ (Computational Omics and Systems Biology Group, Ghent University, Gent-Zwijnaarde, Belgium) was used to infer peptide and protein identifications from spectrum identification data and validated at 1% false discovery rate estimated using the decoy hit distribution. D-score³¹ and phosphoRS score³² (threshold of 95.0 as implemented in the compomics utilities package³³) were used to score post-translational modification localisations. Adequate or acceptable reproducibility or reliability, as it pertains to the results (including Figure 3a and 3c), was defined as the observable extent (measured, for example, as the standard deviation) of stability within measured data points when measurements are repeated under similar experimental conditions.

Data were analysed and graphically visualised and displayed using Pandas, NumPy and Matplotlib Python packages (Python Software Foundation, Wilmington, Delaware, United States), as well as Microsoft Excel (Microsoft Corporation, Redmond, Washington, United States).

The merged lists of either peptide sequences or protein accession numbers (individual as well as protein groups) identified in each sample group and experimental condition were processed using Venny (version 2.1.0)³⁴ (BioinfoGP Service, Universidad Autónoma de Madrid, Madrid, Spain), which calculated and visually displayed the percentage overlap.

To determine the qualitative reproducibility of each experimental condition, the peptide identification overlap (Supplementary document – Figure S1) was computed using the peptide sequences identified for each sample from the data set (regardless of peptide abundance). From these results, the physicochemical properties of the peptides (unique as well as shared) for all conditions were assessed for each patient (Supplementary document – File S2).

Spectrum counting abundance indexes were estimated using the Normalised Spectrum Abundance Factor³⁵ as generated by the PeptideShaker software.^27,36 The Normalised Spectrum Abundance Factor values were normalised to facilitate comparisons and then used to calculate the Pearson’s correlation coefficient, for each pair of experimental conditions compared with regard to differential protein abundance, to determine the level of correlation between samples (Supplementary document – Figure S2).

ProPAS (version 1.1)³⁷ (State Key Laboratory of Proteomics, Beijing Institute of Radiation Medicine, Beijing, China) was used to calculate the physicochemical properties (hydropathicity [Kyte and Doolittle scale],³⁸ molecular weight and isoelectric point) of identified peptides for each sample analysed. Sample physicochemical characteristics were assessed and visualised using box and whisker plots.

Gene Ontology (GO) analysis was performed using protein annotations retrieved from Ensembl (www.ensembl.org) and GOSlim UniProtKB-GOA (www.ebi.ac.uk/GOA) with hypergeometric testing to determine GO term significance. The protein extraction buffers’ (with or without addition of PEG) protein selection bias, as well as residual proteins of the sample pellets, was assessed with regard to subcellular localisation, using GO analysis. Results from the GO annotation were visualised with a bar plot showing percentages of proteins belonging to each GO term and their location.

The percentages of missed cleavages for all samples were calculated and graphically visualised and displayed using Pandas, NumPy and Matplotlib Python packages (Python Software Foundation, Wilmington, Delaware, United States).

We processed the FFPE colonic resection tumour tissues of three patients (diagnosed as indicated in Table 1). For all three patients, non-fractionated LC-MS/MS analysis showed that overall, the WCPLs extracted with 2% SDS and PEG had lower numbers of identifications at both the peptide and protein levels (validated peptides = 6840 [± 588 standard deviation {s.d.}]) and validated proteins = 2302 (± 127 s.d.) (Figure 3a and Figure 3c). On the other hand, the WCPLs extracted without PEG showed higher numbers of identifications (validated peptides = 7058 [± 649 s.d.] and validated proteins = 2314 [± 230 s.d.]) with adequate reproducibility (Figure 3a and Figure 3c). The pellet samples extracted with 4% SDS showed higher overall variability at both peptide and protein levels. However, the numbers of peptide and protein identifications were high for pellets extracted without PEG (validated peptides = 5999 [± 2176 s.d.] and validated proteins = 1893 [± 555 s.d.]) and for pellets extracted with PEG (validated peptides = 4778 [± 1764 s.d.] and validated proteins = 1564 [± 456 s.d.]) (Figure 3a and Figure 3c).

For overlap calculated from merged lists of peptide sequences, 27.1% of identified peptides were shared or overlapped between all the experimental conditions (Figure 3b). Lower percentages of unique peptides were identified for the pellets (8.7% without PEG and 3.7% with PEG), compared to the WCPLs (10.3% without PEG and 11.8% with PEG). For overlap calculated from merged lists of protein accession numbers (individual as well as protein groups), 38% of identified proteins were shared or overlapped between all the experimental conditions. Lower percentages of unique proteins were identified for the pellets (5.1% without PEG and 3.5% with PEG), compared to the WCPLs (9.4% with and without PEG).

No substantial differences were observed for the physicochemical properties of the peptides for each patient (Supplementary document – File S2). All experimental conditions yielded comparable relative protein abundances, indicating that protein extraction with and without PEG did not introduce a substantial observable bias with regard to proteome composition.

The hydropathicity scales of all identified peptides generated from each experimental condition were similar (Figure 4a). The majority of proteins extracted (with and without PEG) and processed via the HILIC/SP3 method were hydrophilic, since the average hydropathicities of all samples were negative (in accordance with the Kyte and Doolittle scale³⁸ and as described by Farias³⁹). Some differences were observed between pellet samples and WCPLs (extracted with and without PEG). The pellet samples seemed slightly more hydrophobic or neutral (closer to 0) in nature compared to the WCPLs. However, neither the addition nor omission of PEG from the protein extraction buffer affected or showed a substantial hydropathicity preference or selection bias with regard to extracted peptides. The molecular weight ranges (majority > 1000 Dalton [Da] and < 2000 Da) (Figure 4b), as well as isoelectric point (pI) ranges (majority above pI 4 and below pI 7) (Figure 4c) of identified peptides were overall similar for all samples and experimental groups compared.

Overall, similar GO functional annotation profiles were obtained for all samples (Figure 5). The majority of proteins were preferentially extracted from the cytoplasm (< 90%), organelles (> 90%), intracellular region (> 90%) and extracellular region (> 60%).

All samples had a majority (> 80%) of fully cleaved peptides (0 missed cleavages), with approximately < 20% peptides with 1 missed cleavage, and approximately < 5% peptides with 2 missed cleavages (Figure 6). In addition, the HILIC/SP3 sample preparation method shows a similar range of missed cleavages in all samples and experimental conditions analysed.

The current study had access to tissue samples that were not limited with regard to sample volumes and concentrations required for MS analysis compared to, for example, limited samples such as fine needle biopsies. Therefore, it was neither feasible nor cost-beneficial for us to determine the effects of PEG at < 10 µg protein, since this was not compatible with the material we had available, and did not fall within the scope of the present study or studies stemming from it.¹²

Using FFPE human colorectal cancer resection tissue, we demonstrated that the addition of 0.5% PEG to protein extraction buffer resulted in overall lower peptide and protein identifications, compared to buffer without the addition of PEG. In addition, protein samples extracted without PEG showed higher reproducibility, and the addition of PEG to the protein extraction buffer generated lower percentages of unique peptides remaining in the sample pellets. By expanding on previous studies that only analysed FFPE animal tissues and human cells, we have demonstrated that high protein concentrations (> 10 µg) from FFPE human colon tissue also compromises the function of PEG. The data from this study, together with our recently published selection of protein purification protocols for different FFPE block ages,¹² should provide pathologists with an optimised methodological approach to exploit the use of archival FFPE tissue blocks.