Technology optimization

Seventeen genes were chosen to represent well established pathology biomarkers or novel tissue-specific or cancer biomarkers discovered within the Human Protein Atlas project [5]. These included bromodomain containing 1 (BRD1), chromogranin-A (CHGA), histone-lysine N-methyltransferase (EZH2), family with sequence similarity 174 member B (FAM174B), glutamate decarboxylase 1 (GAD1), janus kinase 3 (JAK3), junction plakoglobin (JUP), keratin 17 (KRT17), v-yes-1 Yamaguchi sarcoma viral related oncogene homolog (LYN), mix1 homeobox-like 1 (MIXL1), antigen Ki-67 (MKI67), phosphodiesterase 6A (PDE6A), platelet endothelial cell adhesion molecule 1 (PECAM1), protein tyrosine phosphatase type C (PTPRC), special AT-rich sequence-binding protein 2 (SATB2), villin1 (VIL1) and zinc finger protein 473 (ZNF473). Desired PCR products and RNA probes were obtained for all genes (Figure S1).

We next determined whether RNA probe length affects sensitivity or specificity of ISH signals using two different approaches. In the first approach, RNA probes for CHGA and KRT17 were fragmented in duplicate using alkaline- and metal ion-catalyzed methods to investigate if shorter probe length would give a higher signal due to better tissue penetration. Both protocols were adapted for use with DIG-UTP-labeled RNA probes. No significant difference in signal could be seen between fragmented and unfragmented probes when evaluating staining in normal colon, kidney, liver, spleen and tonsil (data not shown). The second approach was to investigate whether longer probes would have an impact on ISH signal. Three different probes targeting PDGFRB (platelet-derived growth factor receptor beta) with lengths of 500 bp, 1000 bp and 1500 bp were generated. No significant difference in signal between the probe lengths could be seen when evaluating staining in kidney glomeruli, however the background was increased for 1000 bp and 1500 bp probes when comparing sense and antisense RNA probes (data not shown) [7].

Adoption of the GenePaint system for human tissue arrays

To facilitate a large-scale approach, optimization of proteinase K concentration was performed. Three genes with various expression levels, PDGFRB, KRT17 and beta actin (ActB), were chosen and evaluated on a tissue microarray with normal colon, kidney, liver, spleen and tonsil. Concentrations tested were 2.5, 5, 10, 20, 40, 60, 80 and 100 µg/ml. For ActB a concentration of 40 µg/ml was sufficient to provide clear and distinct signals without affecting tissue morphology. For KRT17 and PDGFRB, a proteinase K concentration of 60 µg/ml was optimal when comparing all the different tissues on the array and was therefore chosen for further experiments. To visualize genes expressed at low levels, double cycles of tyramide-based signal amplification was introduced. Furthermore, the importance of fresh tissue sections has been ascertained since weak or no signal was seen for tissue sections older than 3 weeks (data not shown).

Validation of ISH staining on tissue arrays

Consecutive tissue sections were stained using sense probe ISH, antisense probe ISH and immunohistochemistry to compare the gene expression on the mRNA and protein level. For a subset of the analyzed genes (CHGA, KRT17, LYN, MKI67, PDE6A, PECAM1, PTPRC and VIL1), immunohistochemistry-based protein expression patterns are well known and established. These targets were used to compare the expression of protein and the corresponding transcripts for validation of the scalable ISH procedure. The intermediate filament protein KRT17 was, as expected, highly expressed in a wide selection of differentiated epithelial cells, as demonstrated by staining of the tonsil (Figure 1, A–C and Figure S3) and in bronchus (Figure S3). CHGA, a marker for neuroendocrine differentiation, showed strong staining in neuroendocrine cells of the gut (Figure 1, D–F), in pancreatic islets and endocrine organs such as the parathyroid (Figure S3). The well characterized proliferation marker Ki-67 is expressed in G1, S, G2 and M stages of the cell cycle and often used to assess proliferating cells within tumors. Proliferating cells in the base of normal colonic crypts (Figure 1, G–I), in anal vulva and in tonsil (Figure S3) are shown. The ISH signal for Ki-67 was localized to specific regions in the nucleus in agreement with data from the mouse [8]. PECAM1 was expressed in the endothelial cells in most organs, exemplified by the richly vascularized placenta (Figure 1, J–L), endometrium (pre menopause) and lymph node (Figure S3). VIL1 is a protein expressed in the brush border of epithelia and as such highly expressed in kidney tubules and glandular cells of the gastrointestinal tract, such as the colon (Figure 1, M–O), appendix and duodenum (Figure S3). Furthermore, for CHGA, PECAM1 and VIL1, two independent RNA probe pairs (Table S1) were used for cross validation of probe specificity. The tissue and cellular distribution pattern of expression for both mRNA and protein were similar for these five genes, confirming the sensitivity and specificity of automated in situ hybridization on FFPE TMAs. For PDE6A, IHC showed staining in alveoli, skeletal muscle, bronchus and cells in the renal glomeruli. mRNA and protein correlation was only seen in two cases of skeletal muscle. Established antibodies for LYN and PTPRC showed distinct membranous and cytoplasmic staining in lymphoid tissues and glandular cells of the gastrointestinal tract, and selective cytoplasmic staining in lymphoid tissue, respectively. No correlation was seen between protein and mRNA expression (Figure S5). For PTPRC, the lack of ISH staining was cross validated with two independent RNA probe pairs (Table S1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Validation of scalable in situ hybridization by parallel immunohistochemistry of pathology biomarkers on tissue arrays.

ISH signals are seen as blue/purple staining with nuclei counterstained in methyl green, whereas IHC signals are in brown with hematoxylin counterstain. A–C: Keratin 17 (KRT17) in tonsil, D–F: Chromogranin A (CHGA) in colon, G–I: Ki-67 (MKI67) in colon, J–L: Platelet endothelial cell adhesion molecule 1 (PECAM1) in placenta, M–O: Villin-1 (VIL1) in colon. Tissue arrays were hybridized with sense control probes (A, D, G, J, M), antisense probes (B, E, H, K, N), or immunostained with antibodies (C, F, I, L, O) targeting the respective transcripts or protein products. All images were derived from slides scanned with a 40× objective.

https://doi.org/10.1371/journal.pone.0032927.g001

Validation of novel antibody biomarkers by ISH on tissue arrays

The antibody-derived expression in the Human Protein Atlas project suggested either a tissue-specific expression or clinical usefulness in cancer diagnostics or therapy for the novel putative biomarkers BRD1, EZH2, FAM174B, GAD1, JAK3, JUP, MIXL1, SATB2 and ZNF473. Therefore, analysis of their in situ mRNA and protein expression in normal tissues could provide validation of antibody specificity. BRD1 is known by immunohistochemistry to be expressed in Purkinje and neuronal cells in CNS, cells in seminiferus ducts in testis, glandular cells in prostate, adrenal gland and appendix, and macrophages in lung, which was confirmed by ISH (Figure 2, A–C and Figure S4). EZH2 is a potential cancer biomarker as overexpression of the nuclear protein is seen in a variety of aggressive cancers, including breast cancer, prostate cancer and glioblastoma multiforme [9]. Gene expression was found on both transcript (cytoplasmic staining) and protein (nuclear staining) in glandular cells in appendix (Figure 2, D–F), duodenum and fallopian tube (Figure S4), cells in seminiferus ducts in testis and myocytes in heart muscles. JUP was suggested by IHC to be a tissue-specific biomarker expressed in adnexal and epidermal cells of the skin, which was confirmed by ISH (Figure 2, G–I) and in squamous epithelial cells of the tonsil (Figure S4). SATB2 is a novel biomarker for colorectal cancer [10], where gene expression profiling has demonstrated association of transcriptional downregulation with metastasis and poor prognosis [11]. Antibodies to SATB2 showed nuclear staining in glandular cells of the lower gastrointestinal tract, including colonic crypt cells (Figure 2, L). Concomitant array ISH confirmed SATB2 transcript expression in cytoplasm restricted to the epithelium of the appendix, colon (Figure 2, J–K) and rectum (Figure S4), but not other organs. For validation of these four novel biomarkers, two independent RNA probe pairs were used for each gene (Table S1). GAD1 showed similar ISH and IHC staining in Purkinje cells and cells in the molecular layer of the cerebellum, but no correlation was seen in other tissues. The specificity of the novel antibodies to FAM174B, JAK3, MIXL1 and ZNF473 could not be verified by ISH (Figure S5). For JAK3, the lack of ISH staining was cross validated with two independent RNA probe pairs but for FAM174B, MIXL1 and ZNF473, only one probe pair could be designed due to either too short protein coding regions or too high homology with other genes (Table S1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 2. Validation of potential tissue-specific and cancer antibody biomarkers by in situ hybridization on tissue arrays.

ISH signals are seen as blue/purple staining with nuclei counterstained in methyl green, whereas IHC signals are in brown with hematoxylin counterstain. A–C: Bromodomain containing 1 (BRD1) in adrenal gland, D–F: Histone-lysine N-methyltransferase (EZH2) in appendix (nuclear and cytoplasmic staining for protein and mRNA, respectively), G–I: Junction plakoglobin (JUP) in skin, J–L: Special AT-rich sequence-binding protein 2 (SATB2) in colon (nuclear and cytoplasmic staining for protein and mRNA, respectively). Tissue arrays were hybridized with sense control probes (A, D, G, J), antisense probes (B, E, H, K), or immunostained with antibodies (C, F, I, L) targeting the respective transcripts or protein products. All images were derived from slides scanned with a 40× objective.

https://doi.org/10.1371/journal.pone.0032927.g002

Parallel analyses of SATB2 protein and transcript expression on FFPE colorectal cancer arrays

Protein and mRNA expression of SATB2 was assessed in colorectal cancers using an array encompassing 60 primary tumors in duplicate. Annotation was performed based on a three-grade scale. One pair did not contain tumor tissue, and of the remaining 59 samples, 44 showed complete concordance between IHC and ISH expression in glandular cells (Figure 3). Twelve showed semi concordance as staining was observed but with different intensities, whereas 3 samples displayed no correlation. Agreement between protein and mRNA expression pattern was observed in a total of 56 samples (Cohen's kappa test, κ = 0.68).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 3. Validation of SATB2 antibody by in situ hybridization in a selection of arrayed colorectal cancers.

ISH signals are seen as blue/purple staining with nuclei counterstained in methyl green, whereas IHC signals are in brown with hematoxylin counterstain. Tissue arrays were hybridized with sense control probe (A, D, G, J), antisense probe (B, E, H, K) or immunostained with antibody (C, F, I, L). All images were derived from slides scanned with a 40× objective.

https://doi.org/10.1371/journal.pone.0032927.g003

Ethics statement

The use of HPA tissue arrays in this study is covered by the HPA ethical permit (EPN Uppsala 2002/577, 2005/338) and ethical permit to the investigators (EPN Uppsala 2007/116). The identities of arrayed tissue samples are not known to the investigators, nor will they be release into the public domain.

Probe generation

The probe generation procedure is outlined in Figure S2. We developed an in-house software that selects transcript regions suitable for hybridization probe design in the genes of interest by minimizing homology with transcripts of other genes in the human RefSeq transcriptome, ensuring that the probe sequences are present in RefSeq transcripts of the gene of interest, and including regions to span exon boundaries. Two independent PCR primer sets for amplification of 500–600 nt products were generated for each gene using Primer3 [12]. A T7 promoter sequence (5′-GCGTAATACGACTCACTATAGGG-3′) was incorporated in the 5′-end of the forward or reverse primer, respectively, to generate the sense or antisense riboprobe template. All primer pairs used in probe generation are summarized in Table S1.

Stratagene's MegaMan human transcriptome library, a collection of cDNA created using mRNA from 32 human tissues and 34 human cancer cell lines, was used for PCR (www.genomics.agilent.com). Each reaction contained 2 µl 10×PCR buffer (166 mM (NH₄)₂SO₄, 670 mM Tris pH 8.8, 100 mM 2-mercaptoethanol, 67 mM MgCl₂), 2 µl dNTPs (10 mM, GE Healthcare, Uppsala, Sweden), 1.2 µl DMSO (Sigma-Aldrich, St. Louis, MO, USA), 0.4 µl forward primer (50 µM, Sigma-Aldrich), 0.4 µl reverse primer (50 µM, Sigma-Aldrich), 0.2 µl Taq polymerase (5 U/µl, Invitrogen, Carlsbad, CA, USA), 1 µl MegaMan human transcriptome library (160 ng/µl, (Agilent Technologies, Santa Clara, CA, USA)) and MilliQ water to a volume of 20 µl. The touchdown PCR conditions included annealing temperatures ranging from 64°C to 57°C; 96°C for 2 min; 3 cycles of 96°C for 10 s, 64°C for 10 s and 70°C for 30 s; 3 cycles of 96°C for 10 s, 61°C for 10 s and 70°C for 30 s; 3 cycles of 96°C for 10 s, 58°C for 10 s and 70°C for 30 s; 41 cycles of 96°C for 10 s, 57°C for 10 s and 70°C for 30 s, and a final extension for 5 min at 70°C. The PCR products were run on a 1.25% agarose gel for product size confirmation. Sense and antisense riboprobes labeled with digoxigenin were generated by in vitro transcription of PCR products with DIG RNA labeling mix (Roche, Rotkreuz, Switzerland) in a total reaction volume of 5 µl. One µg of each RNA probe was run on a 6% TBE-Urea gel (Invitrogen) for size confirmation. Stocks were made by diluting RNA probes with MilliQ water to 200 ng/µl and stored in −80°C. Probe working solutions of 20 ng/µl were made from stock solutions and stored in −20°C.

RNA probes were fragmented using two different approaches. For alkaline-catalyzed fragmentation, giving fragments ranging from 50–150 bp, RNA probes were digested with 0.2 M sodium carbonate buffer (pH 10.2) at 60°C. The incubation time is calculated by t = L₀−L_f/kL₀L_f, where t is incubation time in minutes, L₀ and L_f are initial and final lengths of the probe in kb, and k is the rate constant for hydrolysis (approximately 0.11 kb⁻¹min⁻¹). The reactions were stopped by adding sodium acetate (pH 4.7) and samples were ethanol precipitated [13]. One µg of each RNA probe was run on a 6% TBE-Urea gel for size confirmation. For metal ion-catalyzed fragmentation, giving fragments between 60–200 bp, RNA probes were incubated with 100 mM zinc chloride in 100 mM Tris-HCl (pH 7) at 70°C for 15 min. The reactions were stopped with 0.5 M EDTA (pH 8) [14]. One µg of each RNA probe was run on a 6% TBE-Urea gel for size confirmation.

Tissue preparation

The tissue arrays used were the standard arrays utilized for immunohistochemistry in the Human Protein Atlas project (www.proteinatlas.org). These arrays encompass triplicate 1-mm cores of 48 different types of non-malignant human tissue [15] and duplicate 1-mm cores of primary colorectal cancer [10]. Tissue microarrays with FFPE tumor and normal tissues were sectioned to 6 µm and mounted on Superfrost Plus microscope slides.

In situ hybridization

Tissue sections were dewaxed in xylene, followed by hydration in a graded ethanol series (100%, 95% and 80%, 3 min in each). The hybridization was automated using Tecan GenePaint (Tecan Ag, Männedorf, Switzerland) essentially as described [2], [4], [16]. All procedures were performed at room temperature unless otherwise stated. Briefly, endogenous peroxidase activity was blocked in 0.7% H₂O₂ in methanol, followed by deproteinization in 0.2 M HCl and digestion by 60 µg/ml proteinase K (Roche). The tissue sections were pre-hybridized in hybridization buffer (Ambion, Foster City, CA, USA) without probe for 30 min and then incubated with 200 ng/ml digoxigenin-labeled riboprobe for 4 hours at 64°C. After hybridization, tissue sections were washed in 5× saline-sodium citrate (SSC), 50% formamide and 0.1× SSC. To detect bound probe, an anti-digoxigenin antibody (150 U/ml, Roche) conjugated to horseradish peroxidase was added. After five washes in blocking buffer, a tyramide biotin amplification step was performed to increase the in situ detection sensitivity (Perkin Elmer, Waltham, MA, USA). The deposited biotin was detected by alkaline phosphatase-conjugated neutravidin (2 mg/ml, ThermoScientific, Waltham, MA, USA), which cleaves the chromogenic substrate nitroblue tetrazolium (NBT) 5-bromo-4-chloro-3-indolyl phosphate (BCIP) to produce a blue/purple precipitate at the site of hybridization. After 30 min development time, the chromogenic reaction was stopped by incubating the tissue sections in a buffer containing EDTA, followed by fixation in 4% PFA. Nuclei were counterstained with 2% methyl green and the slides were dehydrated in graded alcohols and mounted in a resin-based medium.

Immunohistochemistry

Briefly, slides were deparaffinized in xylene, hydrated in graded alcohols and blocked for endogenous peroxidase in 0.3% hydrogen peroxide diluted in 95% ethanol. For antigen retrieval, a decloaking chamber (Biocare Medical, Walnut Creek, CA; USA) was used. Slides were immersed and boiled in citrate buffer, pH6 (Lab Vision, Värmdö, Sweden) for 4 min at 125°C and then allowed to cool to 90°C. Automated IHC was performed essentially as described [17], using an Autostainer 480 instrument (Lab Vision). Tissue sections were incubated with primary antibodies (Table S2) and a dextran polymer visualization system (UltraVision LP HRP polymer, Lab Vision) for 30 min each at room temperature and slides were developed for 10 min using diaminobenzidine (Lab Vision) as chromogen. All incubations were followed by a rinse in wash buffer (Lab Vision). Slides were counterstained in Mayer's hematoxylin (Histolab, Gothenburg, Sweden) and cover slipped using Pertex (Histolab) as mounting medium. Incubation with 1× PBS instead of primary antibody served as negative control.

Annotation of ISH and IHC tissues

The Aperio ScanScope CS Slide Scanner system (Aperio Technologies, Vista, CA, USA) was used to digitize whole-slide images of ISH and IHC arrays at 40-fold magnification. The intensity of ISH and IHC staining was manually evaluated and scored using an annotation system similar to that described in [18] with a three grade scale; where 0 denotes absence of signal, 1 presence of weak to moderate signal and 2 presence of strong signal. Cohen's Kappa test was used to determine the agreement between ISH and IHC.