A Systematic Characterization of Aquaporin-9 Expression in Human Normal and Pathological Tissues

Abstract

AQP9 is known to facilitate hepatocyte glycerol uptake. Murine AQP9 protein expression has been verified in liver, skin, epididymis, epidermis and neuronal cells using knockout mice. Further expression sites have been reported in humans. We aimed to verify AQP9 expression in a large set of human normal organs, different cancer types, rheumatoid arthritis synovial biopsies as well as in cell lines and primary cells. Combining standardized immunohistochemistry with high-throughput mRNA sequencing, we found that AQP9 expression in normal tissues was limited, with high membranous expression only in hepatocytes. In cancer tissues, AQP9 expression was mainly found in hepatocellular carcinomas, suggesting no general contribution of AQP9 to carcinogenesis. AQP9 expression in a subset of rheumatoid arthritis synovial tissue samples was affected by Humira, thereby supporting a suggested role of TNFα in AQP9 regulation in this disease. Among cell lines and primary cells, LP-1 myeloma cells expressed high levels of AQP9, whereas low expression was observed in a few other lymphoid cell lines. AQP9 mRNA and protein expression was absent in HepG2 hepatocellular carcinoma cells. Overall, AQP9 expression in human tissues appears to be more selective than in mice.

Introduction

Aquaporin-9 (AQP9), a water, glycerol and urea channel, has been independently identified in mice (Tsukaguchi et al. 1998) and humans (Ishibashi et al. 1998). In these original reports, AQP9 mRNA expression was found in a few tissues, most notably liver and leukocytes in humans (Ishibashi et al. 1998; Tsukaguchi et al. 1999), as well as liver and immature sperm in mice (Tsukaguchi et al. 1998). Since then, AQP9 immunoreactivity has been found in many tissues, including rat and human epididymis (Elkjaer et al. 2000; Pastor-Soler et al. 2001), various cell types of rodent and primate brain (Elkjaer et al. 2000; Badaut et al. 2001; Badaut et al. 2004; Amiry-Moghaddam et al. 2005; Arcienega et al. 2010), mouse spinal cord (Oshio et al. 2004), human chorioamnion and placenta (Damiano et al. 2001; Wang et al. 2004), mouse and human inner ear (Huang et al. 2002; Degerman et al. 2011), mouse and human small intestine (Okada et al. 2003), rat and human prostate (Wang et al. 2008; Hwang et al. 2012), human skeletal muscle (Inoue et al. 2009), urothelium (Rubenwolf et al. 2009), rat and porcine oviduct (Skowronski et al. 2009), human adipose tissue (Rodriguez et al. 2011), human retina (Tran et al. 2013; Yang et al. 2013), and mouse and human skin (Sugiyama et al. 2014).

An analysis of protein expression with immunohistochemistry provides resolution at the cellular and subcellular levels; however, there is a considerable chance for false-positive identification of expression sites due to antibody cross-reactivity. Thus, recommendations for the use of strict controls have been made (Saper and Sawchenko 2003; Saper 2005). One definitive control for antibody specificity is the use of a target knockout, which is currently primarily possible in mice. AQP9 immunolocalization in knockout mice resulted in confirmed expression in liver, skin, and epididymis, but not in testis, spleen, muscle, brain, spinal cord, ovaries, or intestine. This lead to discussions about AQP9 expression in the brain (Arcienega et al. 2010), that were resolved by arguments of putative species differences as well as a confirmation of its expression in a very limited population of neurons (Mylonakou et al. 2009).

It is reasonable to assume that some described differences between humans and mice are due to species differences. Furthermore, it remains possible that, similar to the recorded AQP9 expression in Mus musculus, some previously described locations of AQP9 expression in Homo sapiens are false-positive identifications, based on an incorrect interpretation of antibody reactivity. One goal of the current study was therefore to systematically explore the distribution of AQP9 expression in humans. For this purpose, we used a previously well-characterized antibody (Elkjaer et al. 2000; Rojek et al. 2007; Jelen et al. 2013) for studying AQP9 expression within the Human Protein Atlas project tissue repository (Uhlen et al. 2010; Uhlen et al. 2015), including a large spectrum of normal and cancer tissues as well as cell lines and primary cells (Elkjaer et al. 2000; Rojek et al. 2007; Jelen et al. 2013). A further goal of the study was to re-evaluate a proposed tumor necrosis factor-alpha (TNFα)-dependent up-regulation of AQP9 in synovial tissue of rheumatoid arthritis patients (Nagahara et al. 2010) before/after treatment with the TNFα blocker Humira. Validation of the antibody and the immunohistochemical procedure were performed according to strict guidelines, and the results were compared with mRNA expression data based on shotgun RNA sequencing (RNA-seq).

Materials & Methods

Sample Preparation

Normal and cancer tissue samples used for protein and mRNA expression analyses were obtained from the Department of Pathology, Uppsala University Hospital, Uppsala, Sweden, as part of the sample collection governed by the Uppsala Biobank (http://www.uppsalabiobank.uu.se/en/). The tissue samples used were anonymized in accordance with the approval and advisory report from the Uppsala Ethical Review Board (Reference # 2002-577, 2005-338 and 2007-159 (protein) and # 2011-473 (RNA)). Normal tissue samples used for protein and mRNA expression analyses were not obtained from the same individuals and are not matched. In addition, duplicate samples of synovial tissue from six rheumatoid arthritis patients before and after treatment with Humira were included. Cell lines used in the analyses were derived from DSMZ, ATCC or academic research groups (kindly provided by cell line founders). Information on sex and age of the donor, tissue origin and supplier can be found at http://www.proteinatlas.org/about/cellines.

Protein Profiling

Generation of tissue microarrays (TMAs), immunohistochemical staining and slide scanning were performed as previously described (Kampf et al. 2012). In brief, formalin-fixed and paraffin-embedded (FFPE) tissue samples were collected from the pathology archives based on hematoxylin-eosin stained tissue sections showing representative normal histology for each tissue type. Representative cores (1 mm in diameter) were sampled from the FFPE blocks and assembled into two TMAs, containing normal tissue samples from a total of 144 individuals. Moreover, 216 cancer duplicate samples corresponding to 20 different cancer types were assembled into four TMAs. Duplicate samples from 46 cell lines, 10 leukemia blood cell samples, and 2 samples of normal peripheral blood mononuclear cells (PBMC) were used to create a cell microarray (CMA). Cell lines were cultured according to instructions provided by the suppliers. Following fixation in formalin, cells were dispersed in agarose prior to paraffin embedding. The processing of cells enables the generation of cell blocks mimicking tissues with respect to fixation and dehydration, and thus facilitates comparisons between the cell and tissue arrays. This procedure has been described and evaluated previously (Andersson et al. 2006). TMA and CMA blocks were cut into 4-µm sections using waterfall microtomes (Microm HM 355S, Thermo Fisher Scientific; Freemont, CA), dried at room temperature overnight, and baked at 50°C for 12–24 hr prior to immunohistochemical staining. Rheumatoid arthritis samples were analyzed on full sections using the same staining procedure as for the TMAs. Validation experiments using consecutive sections with anti-AQP9 RA2674/685 and antibodies towards CD68, CD20 and CD3E were performed on full sections of normal human liver and appendix. For antigen retrieval, a pressure boiler (Decloaking chamber, Biocare Medical; Walnut Creek, CA) was used. The slides were boiled for 4 min at 125°C in PT Module Buffer, pH 9 (Thermo Fisher Scientific) before cooling to 90°C. The total processing time was approximately 45 min. Slides were deparaffinized in xylene, hydrated in graded alcohols, and then blocked for endogenous peroxidase activity in 0.3% hydrogen peroxide. The primary antibody anti-AQP9 RA2674/685 (Elkjaer et al. 2000) was diluted 1:325 in UltraAb Diluent (Lab Vision; Freemont, CA), and, as a secondary reagent, a dextran polymer visualization system was used (UltraVision LP HRP polymer, Lab Vision). For both the primary antibody and the secondary reagent, slides were incubated for 30 min at room temperature. Slides where then developed for 10 min using Diaminobenzidine (Lab Vision) as a chromogen. All incubations were followed by rinsing in Wash buffer (Lab Vision) for 5 min. Slides were counterstained in Mayer’s hematoxylin (Histolab) and mounted under coverslips in Pertex (Histolab). Incubation with PBS instead of primary antibody served as negative control. Anti-AQP9 antibodies sc-28623 and sc-74409 (Santa Cruz Biotechnology), as well as two internally generated antibodies were also tested on the immunohistochemistry platform but were evaluated as not suitable due to nonspecific staining and inconsistency with RNA-seq data (Supplementary Fig. 1). Immune markers were used as follows: anti-CD68 (AMAb90874, 1:10,000; Atlas Antibodies, Stockholm, Sweden), anti-CD20 (M0755, 1:1000; DakoCytomation, Glostrup, Denmark) and anti-CD3E (AMAb90879, 1:1000; Atlas Antibodies). Automated immunohistochemistry was performed using Autostainer 480 instruments (Lab Vision), followed by slide scanning using Scanscope XT (Aperio; Vista, CA). The corresponding high-resolution images of immunohistochemically stained tissue samples were evaluated and annotated using a web-based annotation system (unpublished). In brief, the manual score of immunohistochemistry-based protein expression was determined as the fraction of positive cells defined in different tissue and cell samples divided into four different groups: 0% to 1%, 2% to 25%, 26% to 75% or >75%. The intensity of the immunoreactivity was subdivided into negative, weak positivity, moderate positivity or strong positivity. The fraction of positive cells and the intensity of the immunoreactivity were then combined into a total score: 0 = not detected (negative, or weak positivity in <25% of the cells), 1 = low expression (weak positivity in 25% to 100% of the cells, or moderate in <25%), 2 = medium expression (moderate positivity in 25% to 100% of the cells, or strong in <25%), and 3 = high expression (strong positivity in >25% of the cells). Immunolabeling results in cells and cell lines were also assessed using an automated image analysis software, TMAx (Definiens; Munich Germany), which automatically measures staining intensity and counts the fractions of stained cells, generating a score between 0 and 20,000. This automated scoring has been evaluated previously (Stromberg et al. 2007).

Antibody Validation

Suitability of anti-AQP9 RA2674/685 for human AQP9 detection was further evaluated by immunoblotting (Supplementary Fig. 2). Total protein lysates were prepared using ProteoExtractVR Complete Mammalian Proteome Extraction Kit (CalbiochemVR, Merck KGaA; Darmstadt, Germany) from selected cell lines and tissues. Protein preparations (15 μg of RT-4, U-251MG, liver, and tonsil as well as 25 μg of HSA- and IgG-depleted plasma) and a marker (PageRuler Plus Prestained Protein Ladder; Thermo Fisher Scientific) were loaded on precast 4% to 20% Criterion SDS–PAGE gradient gels (Bio-Rad Laboratories, Montreal QC, Canada). The gels were run under reducing conditions, followed by transfer to PVDF membranes (Bio-Rad) using Trans-Blot turbo (Bio-Rad) according to the manufacturer’s recommendations. Membranes were activated in methanol before blocking (5% dry milk, 0.5% Tween-20, 1×TBS; 1 mM Tris–HCl, 0.15 M NaCl) for 1 hr at room temperature with constant shaking. The membranes were then incubated with primary antibody diluted 1:500 for 1 hr, followed by washing (1 mM Tris–HCl, 0.15 M NaCl, 0.05% Tween-20) and then incubated for 45 min with the secondary HRP-conjugated antibody (swine anti-rabbit 1:3000, DakoCytomation). A CCD-camera (Bio-Rad) was used for detection of signal from the substrate (Immobilon Western Chemiluminescence HRP Substrate; Merck KGaA).

Transcript Profiling (RNA-seq)

Tissues samples were embedded in Optimal Cutting Temperature (O.C.T.) compound and stored at -80°C. A hematoxylin and eosin-stained frozen section (4 µm) was prepared from each sample using a cryostat and the CryoJane® Tape-Transfer System (Instrumedics; St. Louis, MO). Each slide from 32 different normal tissues was examined by a pathologist to ensure sampling of histologically normal tissue. Three sections (10 µm) were cut from each frozen tissue block and collected into a tube for subsequent RNA extraction. The tissue was homogenized mechanically using a 3-mm steel grinding ball (VWR; Radnor, PA). Total RNA was extracted from both cell lines and tissue samples using the RNeasy Mini Kit (Qiagen; Hilden, Germany) according to the manufacturer’s instructions. The extracted RNA samples were analyzed using either an Experion automated electrophoresis system (Bio-Rad Laboratories) with a standard-sensitivity RNA chip or an Agilent 2100 Bioanalyzer system (Agilent Biotechnologies; Palo Alto, CA) with a RNA 6000 Nano Labchip Kit. Only samples of high-quality RNA (RNA Integrity Number ≥7.5) were used in the following mRNA sample preparation for sequencing.

Data Analysis

Sequencing of a total number of 122 samples from different individuals corresponding to a total of 32 different tissues and organs, as well as 85 cell line samples was performed as previously described using an Illumina Hiseq2000 (Illumina; San Diego, CA) and the standard Illumina RNA-seq protocol (Fagerberg et al. 2014). In brief, processed reads were mapped to the human genome (GRCh37) using Tophat v2.0.8b (Trapnell et al. 2009). To obtain quantification scores for all human genes and transcripts across all samples, FPKM (fragments per kilobase of exon model per million mapped reads) values were calculated using Cufflinks v2.1.1 (Trapnell et al. 2010). The gene models from Ensembl build 78 were used in Cufflinks. The average FPKM value of all individual samples for each tissue was used to estimate the gene expression level.

Results

AQP9 protein expression was analyzed using immunohistochemistry on a large set of normal tissues, cancer tissues and cell lines assembled into tissue and cell microarrays, as well as on full sections of rheumatoid arthritis samples before and after treatment with Humira. The proteomics analysis was complemented with transcriptomic profiling using deep sequencing (RNA-seq) on 32 out of the 44 normal organs, as well as 44 out of the 46 analyzed cell lines (Uhlen et al. 2015). The normalized mRNA abundance was calculated as FPKM-values, with an FPKM = 1 roughly corresponding to one mRNA molecule per average cell in the sample (Hebenstreit et al. 2011).

Expression of AQP9 in Normal Tissues

A manual assessment of the immunohistochemical staining pattern was performed in 44 different normal tissues in a total of 144 tissue cores, corresponding to 83 different cell types (Table 1), out of which 82 cell types were analyzable. The analysis revealed that a high expression of AQP9 was observed exclusively in hepatocyte membranes (Fig. 1A). In order to further investigate potential positivity in Kupffer cells, staining was performed on consecutive sections of normal human liver with antibodies towards AQP9 and CD68 (Supplementary Fig. 3A). The additional staining confirmed membranous expression in hepatocytes, whereas no expression was observed in Kupffer cells. A medium level of expression of AQP9 was shown in a smaller subset of lymphoid cells in appendix tissue (Fig. 1B) and in hematopoietic cells of the bone marrow (Fig. 1C). These locations represent the tissues with 1^st, 2^nd and 3^rd highest relative AQP9 mRNA expression (FPKM = 224.3, 66.2 and 31.4, respectively), indicating a high consistency between immunohistochemical staining pattern and mRNA expression. Furthermore, a medium level of AQP9 expression was observed in a subset of cells in red pulp of the spleen (Fig. 1D), the organ with the 4^th highest mRNA expression (FPKM = 14.5). Moderate positivity in a fraction of the cells was also found in non-germinal center cells of the tonsil (Fig. 1E) and renal glomeruli (Fig. 1F), where essentially no mRNA expression was observed. No or very low AQP9 immunoreactivity was observed in all other examined tissues that expressed low to very low AQP9 mRNA levels (Table 1). The protein expression in a subset of lymphoid cells in the appendix, the tissue with 2^nd highest AQP9 mRNA expression, was further analyzed with staining on consecutive sections of normal human appendix using antibodies towards AQP9, CD68 (for macrophages), CD20 (for B-cells), and CD3E (for T-cells) (Supplementary Fig. 3B). The staining of AQP9 in appendix tissue appeared to overlap with the expression of a subset of cells positive for CD68 in lymphoid cells outside the germinal centra, hence representing a subset of macrophages. A similar, weaker positivity was observed also in macrophages of the sections stained with CD20 and CD3E, interpreted as nonspecific background staining. No overlap was observed between AQP9 and the specific positivity representative of B-cells or T-cells.

Table 1.

Expression of AQP9 Protein and AQP9 mRNA in Normal Tissues.

Tissue	Cell Type	Protein Expression	Concluded Protein Expression	RNA Expression (FPKM)
Liver	hepatocytes	3	Y	224.3
	bile duct cells	0	N
	Kupffer cells	0	N
Gallbladder	glandular cells	0	N	5.2
Pancreas	exocrine glandular cells	0	N	0.4
	islets of Langerhans	0	N
Oral mucosa	squamous epithelial cells	0	N	N/A
Salivary gland	glandular cells	0	N	0.1
Esophagus	squamous epithelial cells	0	N	1.1
Stomach	glandular cells	0	N	0.5
Duodenum	glandular cells	0	N	0.3
Small intestine	glandular cells	0	N	0.5
Appendix	glandular cells	0	N	66.2
	lymphoid tissue	1	Y
Colon	endothelial cells	0	N	0.4
	glandular cells	0	N
	peripheral nerve/ganglion	0	N
Rectum	glandular cells	0	N	0.4
Kidney	cells in glomeruli	1	N	0.1
	cells in tubules	1	N
Urinary bladder	urothelial cells	0	N	8.5
Testis	cells in seminiferous ducts	0	N	0.2
	Leydig cells	0	N
Epididymis	glandular cells	0	N	N/A
Prostate	glandular cells	0	N	0.4
Seminal vesicle	glandular cells	0	N	N/A
Breast	adipocytes	0	N	N/A
	glandular cells	0	N
	myoepithelial cells	0	N
Vagina	squamous epithelial cells	0	N	N/A
Cervix, uterine	squamous epithelial cells	0	N	N/A
	glandular cells	0	N
Endometrium	glandular cells	0	N	1.1
	cells in endometrial stroma	0	N
Fallopian tube	glandular cells	0	N	7.1
Ovary	follicle cells	0	N	0.2
	ovarian stroma cells	0	N
Placenta	trophoblastic cells	0	N	3.4
	decidual cells	0	N
Skin	fibroblasts	0	N	14.4
	keratinocytes	0	N
	Langerhans cells	0	N
	melanocytes	0	N
	epidermal cells	0	N
Bone marrow	hematopoietic cells	1	Y	31.4
Lymph node	germinal center cells	0	N	0.5
	non-germinal center cells	0	N
Tonsil	germinal center cells	1	N	0.7
	non-germinal center cells	1	N
	squamous epithelial cells	0	N
Spleen	cells in white pulp	0	N	14.5
	cells in red pulp	1	Y
Cerebral cortex	neuronal cells	0	N	1.2
	endothelial cells	0	N
	glial cells	0	N
	neuropil	0	N
Hippocampus	neuronal cells	0	N	N/A
	glial cells	0	N
Lateral ventricle	neuronal cells	0	N	N/A
	glial cells	0	N
Cerebellum	Purkinje cells	0	N	N/A
	cells in granular layer	0	N
	cells in molecular layer	0	N
Thyroid gland	glandular cells	0	N	0.5
Parathyroid gland	glandular cells	0	N	N/A
Adrenal gland	glandular cells	0	N	0.8
Nasopharynx	respiratory epithelial cells	0	N	N/A
Bronchus	respiratory epithelial cells	0	N	N/A
Lung	pneumocytes	0	N	7.2
	macrophages	0	N
Heart muscle	myocytes	0	N	0.2
Skeletal muscle	myocytes	0	N	0
Smooth muscle	smooth muscle cells	0	N	5.1
Soft tissue	adipocytes	0	N	13.3
	fibroblasts	0	N
	peripheral nerve	0	N

FPKM (fragments per kilobase of exon model per million mapped reads) values were calculated using Cufflinks v2.1.1 (Trapnell et al. 2010).

Figure 1.

Protein expression of AQP9 in normal tissues. (A) High membranous expression was observed in hepatocytes. (B) Medium level cytoplasmic expression was found in a subset of lymphoid cells in appendix (a, b). In bone marrow (C) and cells in red pulp of spleen (D), a subset of cells showed medium level cytoplasmic expression (c-f). (E) Moderate AQP9 immunoreactivity was also observed in cytoplasm of a subset of non-germinal center cells in tonsil (g, h). (F) In kidney, endothelial cells of renal glomeruli (i) displayed medium level expression, whereas renal tubules (j) showed weak expression. Scale, 100 µm.

Expression of AQP9 in Cancer Tissues

AQP9 protein expression in cancer tissues was explored in 216 patient samples representing 20 common cancer types, out of which 204 samples were analyzable (Table 2). In liver cancer, medium to high membranous expression of AQP9 was observed in five out of eight analyzed hepatocellular carcinomas (Fig. 2A), whereas the remaining three hepatocellular carcinomas displayed low expression. No expression was observed in the four cholangiocarcinomas (Fig. 2B). Moreover, a medium-level expression was found in one out of 12 renal cancers (Fig. 2C), as well as one out of 10 urothelial cancers (Fig. 2D). No or only very low AQP9 immunoreactivity was observed in the remaining samples (n=190).

Table 2.

Expression of AQP9 Protein in Cancer Tissues.

Tissue	Intensity (No. of Samples)
Breast cancer	3 (0)	2 (0)	1 (1)	0 (11)
Carcinoid	3 (0)	2 (0)	1 (0)	0 (4)
Cervical cancer	3 (0)	2 (0)	1 (0)	0 (12)
Colorectal cancer	3 (0)	2 (0)	1 (0)	0 (9)
Endometrial cancer	3 (0)	2 (0)	1 (1)	0 (10)
Glioma	3 (0)	2 (0)	1 (0)	0 (12)
Head and neck cancer	3 (0)	2 (0)	1 (1)	0 (3)
Hepatocellular carcinoma	3 (4)	2 (1)	1 (3)	0 (0)
Cholangiocarcinoma	3 (0)	2 (0)	1 (0)	0 (4)
Lung cancer	3 (0)	2 (0)	1 (2)	0 (10)
Lymphoma	3 (0)	2 (0)	1 (0)	0 (12)
Melanoma	3 (0)	2 (0)	1 (1)	0 (11)
Ovarian cancer	3 (0)	2 (0)	1 (0)	0 (12)
Pancreatic cancer	3 (0)	2 (0)	1 (0)	0 (11)
Prostate cancer	3 (0)	2 (0)	1 (1)	0 (9)
Renal cancer	3 (0)	2 (1)	1 (0)	0 (11)
Skin cancer	3 (0)	2 (0)	1 (2)	0 (10)
Stomach cancer	3 (0)	2 (0)	1 (0)	0 (11)
Testis cancer	3 (0)	2 (0)	1 (0)	0 (11)
Thyroid cancer	3 (0)	2 (0)	1 (0)	0 (4)
Urothelial cancer	3 (0)	2 (0)	1 (0)	0 (9)

Figure 2.

Protein expression of AQP9 in cancer tissues. High expression was observed in the membranes of tumor cells in hepatocellular carcinomas (A), whereas no expression was found in cholangiocarcinomas (B). Moreover, a high to medium level of membranous expression was noted in single cases of renal cancer (C) and urothelial cancer (D). Scale, 100 µm.

Expression of AQP9 in Rheumatoid Arthritis Patients

Immunohistochemical analysis of the staining pattern in synovial lining and mesenchymal cells of samples from six rheumatoid arthritis patients was evaluated before and after treatment with Humira (Table 3). Two patients showed low cytoplasmic expression of AQP9. In one of these patients, the expression of AQP9 in synovial lining was downregulated after treatment with Humira, whereas its expression in mesenchymal cell appeared unaltered. In the other patient, the low expression in synovial lining cells remained so after treatment, whereas expression appeared absent in mesenchymal cells (Fig. 3). No expression of AQP9 was detected in the remaining four samples.

Table 3.

Expression of AQP9 protein in rheumatoid arthritis tissues before and after treatment with Humira.

Patient	Treatment	Protein Expression Synovial Lining	Protein Expression Mesenchymal Cells
1	Before	0	0
	After	0	0
2	Before	0	0
	After	0	0
3	Before	0	0
	After	0	0
4	Before	1	1
	After	0	1
5	Before	1	1
	After	1	0
6	Before	0	0
	After	0	0

Figure 3.

Protein expression of AQP9 in synovial rheumatoid arthritis patients. Low cytoplasmic expression was observed in synovial lining (a) and mesenchymal cells (b) of two patients before Humira treatment (A and C). In one patient (B), the expression in synovial lining (c) was lost after Humira treatment, whereas the cytoplasmic expression was still observed in mesenchymal cells (d). In the other patient (D), the expression was unaltered in synovial lining (e), while no expression was found in mesenchymal cells (f). Scale, 100 µm.

Expression of AQP9 in Cell Lines

Protein expression in 46 cell lines and 12 samples of primary cells was analyzed based on manual assessment using the same scoring criteria and cut-offs as for the tissues (Table 4). The analysis revealed a medium to high level of membranous AQP9 expression in a majority of LP-1 cells (Fig. 4A), but low or medium expression in a smaller fraction of the cells in several lymphoid cell lines, such as U266/70 (Fig. 4B), and also in the HL-60 myeloid cell line. Most of the other cell lines and all primary cells were negative. The manual assessment was complemented with automated image analysis. These automated results were very similar to the results generated manually; however, shades of brown were interpreted as background and hence were scored as “not detected” by the experienced manual annotator (CL) but as “low protein expression” by the image analysis software.

Table 4.

Expression of AQP9 Protein and AQP9 mRNA in Cell Lines.

Cell line	Type	Protein Expression (Manual Score)	Protein Expression (Automated Score)	Concluded Protein Expression	RNA Expression (FPKM)
Myeloid cell lines
HEL	Erythroleukemia cell line	1	809	N	0
HL60	Acute promyelocytic leukemia cell line	1	2092	N	0.1
HMC-1, K562, NB-4, THP-1, U-937		0	456-1299	N	0-0.1
Lymphoid cell lines
LP-1	Multiple myeloma cell line	3	2621	N	N/A
Daudi	Human Burkitt lymphoma cell line	1	707	N	0
HDLM-2	Hodgkin lymphoma cell line	1	2509	N	0
REH	Pre-B cell leukemia cell line	1	1465	N	0
RPMI-8226	Multiple myeloma cell line	1	1387	N	0
U-266/70	Multiple myeloma cell line	1	977	N	0
U-266/84	Multiple myeloma cell line	1	1094	N	0
Karpas-707, MOLT-4, U-698		0	151-338	N	0
Brain cell lines
U-87 MG	Glioblastoma, astrocytoma cell line	0	508	N	16.4
SH-SY5Y, U-138 MG, U-251 MG		0	239-526	N	0
Lung cell lines
A549, SCLC-21H		0	429, 268	N	0
Abdominal cell lines
CACO-2, CAPAN-2, Hep G2		0	471-482	N	0
Female reproductive system cell lines
AN3-CA, EFO-21, HeLa, SiHa		0	416-753	N	0
Metastatic breast adenocarcinoma cell lines
MCF7, SK-BR-3		0	390, 442	N	0
Urinary, male reproductive system cell lines
NTERA-2, PC-3, RT4		0	124-352	N	0
Skin cell lines
A-431, HaCaT, SK-MEL-30, WM-115		0	187-469	N	0-0.1
Sarcoma cell lines
RH-30, U-2 OS, U-2197		0	365-725	N	0-0.2
Miscellaneous cell lines
BEWO, HEK 293, HTh 83, TIME		0	157-1319	N	0
Primary patient leukocytes/leukemia cells
Leukemia, AML	Acute myeloid leukemia	0	470	N	N/A
Leukemia, B-ALL	Acute B Lymphoblastic leukemia	0	1239	N	N/A
Leukemia, T-ALL	T-cell acute lymphoblastic leukemia	0	1274	N	N/A
Leukemia, CML	Chronic myeloid leukemia	0	180	N	N/A
PBMC	Peripheral blood mononuclear cells from healthy donors	0	442	N	N/A

An extended version of this table is available at http://www.proteinatlas.org/ENSG00000103569-AQP9/cell. FPKM (fragments per kilobase of exon model per million mapped reads) values were calculated using Cufflinks v2.1.1 (Trapnell et al. 2010). N, no; N/A, not available.

Figure 4.

Protein expression of AQP9 in cell lines. (A) High expression was observed in membranes of LP-1 myeloma cells. Medium level expression in a subset of cells was found in U266/70 cells (B), whereas no or only very low expression was noted in U-87 MG cells (C) and HepG2 cells (D). Scale, 100 µm.

Transcriptomic profiling of 44 of the cell lines revealed selective AQP9 mRNA expression in the U-87 MG glioblastoma cell line only (FPKM=16.4), but no expression of AQP9 observed in U-87 MG at the protein level (Fig.4C). Moreover, no expression either at the mRNA or protein level was observed in the hepatocellular carcinoma cell line HepG2 (Fig. 4D).

Discussion

The goal of the current study was to systematically explore AQP9 protein expression in human tissues and cell lines using well-characterized immunological reagents (Jelen et al. 2013). Whereas mRNA and protein may not always be present in a cell simultaneously, mRNA expression can nevertheless be considered a useful criterion for discriminating true protein expression from nonspecific immunolocalization signals. Results from parallel mRNA shotgun sequencing were thus compared to observed sites of AQP9 protein expression.

Out of the 44 investigated normal human tissues, we were able to confidently confirm AQP9 expression only in human liver. Further sites of low-intensity immunolocalization signals were interpreted as negative by an experienced observer based on the following criteria: 1) AQP9 is a known plasma membrane protein, based on verified function in hepatocyte glycerol uptake (Jelen et al. 2011; Calamita et al. 2012; Jelen et al. 2012). Immunolocalization at the cell periphery was thus considered as a strong indication of true signal. 2) Intracellular signals were nevertheless considered as potentially true-positive because AQP9 localization in mitochondria has been described (Amiry-Moghaddam et al. 2005). Furthermore, AQP9 has to be translated in the ER and needs to proceed through the Golgi network before reaching the plasma membrane. Expression sites were thus considered plausible if AQP9 mRNA expression was at least 5% of the liver AQP9 mRNA expression level.

Consequently, low to moderate immunolocalization signals that were detected in a subset of leukocytes or hematopoietic cells in the appendix, bone marrow and spleen were considered as bona fide expression sites of AQP9 protein (Jelen et al. 2011; Calamita et al. 2012; Jelen et al. 2012). Staining on consecutive sections of appendix with additional lymphoid markers suggested cytoplasmic expression in a subset of macrophages; however, it could not be confirmed if this positivity reflected true protein expression or represented nonspecific antibody binding. Intracellular immunolocalization signals in tissues and cell lines with less than 5% of liver AQP9 mRNA expression were considered as likely false-positive. This was especially considered likely if weak immunolocalization signals were seen in a high number of cells, e.g. as observed in kidney. Since the utilized procedure largely eliminates endogenous peroxidases and biotin as sources of false-positive signals, such signals likely represent nonspecific immunoreactivity.

This general picture of highly restricted AQP9 expression differs significantly from a literature review that suggests further expression sites within the human body, including expression sites in epididymis (Pastor-Soler et al. 2001), placenta (Damiano et al. 2001; Wang et al. 2004), small intestine (Okada et al. 2003), prostate (Wang et al. 2008; Hwang et al. 2012), human skeletal muscle (Inoue et al. 2009), urothelium (Rubenwolf et al. 2009), adipose tissue (Rodriguez et al. 2011) and skin (Sugiyama et al. 2014), as well as additional sites that have not been studied here.

In principal, these differences may be due to false-negative findings in our study; for example, due to a lack of sensitivity, the presence of splice variants or post-translationally modified AQP9 protein variants expressed in some tissues that are not detected by anti-AQP9 RA2674/685, or the type of antiserum used in this study. Furthermore, differences may result from genetic variability between tested subjects or false-positive findings in the cited studies; for example, due to antibody cross-reactivity with nonspecific epitopes. Other well-known parameters that affect the outcome of immunohistochemistry are the use of appropriate epitope retrieval and the use of manual versus automated immunohistochemistry platforms (O’Hurley et al. 2014). It can be anticipated that the automated system combined with the TMA technology used in the Human Protein Atlas project may minimize false-positive or false-negative findings based on reproducibility issues.

Considering sensitivity, RNA-seq is known to afford excellent quantification of mRNA expression levels over a large, dynamic range (Wang et al. 2009). However, mRNA translation efficiency and protein stability are expected to vary among tissue types and cell types and will depend on cellular status. Furthermore, mRNA expression levels are derived from the mean expression in all cell types of a tissue biopsy, whereas protein immunoreactivity can be resolved at the cellular level. To base calculations of immunolocalization sensitivity on mRNA expression values is thus speculative but would suggest probable detection of immunoreactivity resulting from approximately 30% or more of cellular AQP9 expression as compared to mean liver expression levels, whereas a per-cell signal from mean liver AQP9 expression of approximately 15% or lower may be below the detection limit. Further consideration of these numbers suggests that a low level expression of AQP9 in human skin, urothelium and adipose tissue is plausible but unfortunately below the detection limit of our procedure. In addition, AQP9 expression at any relevant level appears unlikely in the small intestine, skeletal muscle and testis, based on mRNA shotgun sequencing results. The RNA-seq results are in full agreement with independently determined AQP9 expression values described in the GTEx Portal (gtexportal.org) as well as BioGPS (biogps.org). We did not observe AQP9 immunoreactivity in human epididymis; however, epididymis was not among the tissues included in the RNA-seq tissue panel generated as part of the Human Protein Atlas and, hence, cross-validation of this result is not possible.

With regard to genetic variability between tested subjects, it should be noted that normal tissues from three different individuals in the present investigation showed no clear differences. Concerning alternative AQP9 splicing, three putative protein-forming splice variants have been identified to date (Ensembl accession number: ENSG00000103569). None of these alter the carboxyl-terminus of the protein, and all AQP9 protein isoforms derived from known splice variants should thus be equally well detected by the antiserum used in this study. Tissue-specific, alternative posttranslational modifications of the antigenic region remain a hypothesis without evidence.

Technical weaknesses of immunolocalization studies that may cause false-positive findings are well known. Recognized problems may include the use of commercial antisera raised against undisclosed peptides (Saper and Sawchenko 2003; Saper 2005). Examples of such antisera that frequently have been used in the characterization of AQP9 expression are AQP91-A from Alpha Diagnostics as well as several sera from Santa Cruz Biotechnology (Okada et al. 2003; Wang et al. 2004; Damiano et al. 2006; Warth et al. 2007; Wang et al. 2008; Rubenwolf et al. 2009; Arcienega et al. 2010; Rodriguez et al. 2011; Hwang et al. 2012; Sugiyama et al. 2014). We have previously characterized AQP91-A and two antisera from Santa Cruz Biotechnology alongside the antiserum used in this study for use in immunolocalization studies of human tissue (Jelen et al. 2013). One antiserum from Santa Cruz Biotechnology as well as AQP91-A did not detect ectopically expressed human AQP9 in paraffin-embedded cells. In the present investigation, two other antisera from Santa Cruz Biotechnology were tested and evaluated as not suitable for immunohistochemistry. Differences in tissue preparation as well as batch differences may explain that the same sera could be found suitable in other studies.

AQP9 mRNA expression in leukocytes was observed along with the first identification of this transcript (Ishibashi et al. 1998; Tsukaguchi et al. 1999), and utilizing anti-AQP9 RA2674/685, we have previously found that, in peripheral blood preparations, AQP9 expression was localized in granulocytes (Jelen et al. 2013). In the present investigation, no expression was observed in PBMCs; however, a subset of hematopoietic cells in bone marrow along with subsets of cells in appendix and spleen tissues displayed distinct positivity, possibly representing granulocytes. Hepatocytes were the only normal cell type showing a clear membranous expression consistent with the expected localization for AQP9 protein, whereas other positive cell types showed a more diffuse cytoplasmic staining.

In addition to the systematic evaluation of AQP9 expression in normal human tissues, we utilized anti-AQP9 RA2674/685 to investigate AQP9 expression in 20 types of cancer tissues. Aquaporins are known to contribute to tumor metastasis (Hu and Verkman 2006), vascularization (Saadoun et al. 2005), and chemo-resistance (Gao et al. 2012). Except for expression in hepatocellular carcinoma, we only found a sporadic, low-level expression of AQP9 in all other investigated biopsies. The overall low expression level of AQP9 in malignant human tissue biopsies argues against a general or important role of AQP9 in cancer biology. Our observations differ from the frequent identification of AQP9 in malignant ovarian tumors (Yang et al. 2011a; Yang et al. 2011b) but agree with very low AQP9 mRNA expression in these cancer types.

We did, however, consistently detect high membranous AQP9 protein expression in hepatocellular carcinoma. This was supported by high mRNA expression in hepatocellular carcinomas only, based on transcriptomic analysis of 67 samples from 18 different types of cancer (unpublished data). In this dataset, hepatocellular carcinomas showed a mean FPKM of 294.5, whereas the mean FPKM in cholangiocarcinomas and other cancer types were 29.5 and 5.0, respectively. The agreement between the protein and mRNA sequencing results makes false-positive immunolocalization an unlikely explanation. Although mRNA samples and tissue biopsies were not obtained from the same patients, the results conflict with previous reports of consistent, drastic AQP9 reduction in hepatocellular carcinoma (Jablonski et al. 2007; Padma et al. 2009). Possible explanations for discrepancies may include the small number of patient biopsies sampled in this as well as in previous studies—possibly selected from a genetically homogenous population—or inconsistencies in tissue fixation that may have led to loss of AQP9 immunolabeling previously. Another explanation as discussed above may involve less likely scenarios of splicing/posttranslational modifications that specifically affected the antiserum utilized in the previous study. In addition, we did not find any indication of AQP9 expression in glioma, which again agrees with the low AQP9 mRNA expression levels (unpublished data) and differs from previous observations of others (Warth et al. 2007; Tan et al. 2008). We have extensively studied this tissue before with the same antiserum as used here, and found AQP9 expression in some patients yet in a small cell population (Jelen et al. 2013).

Several recent studies have detected AQP9 mRNA and protein expression changes in inflammatory diseases (Mesko et al. 2010; Mesko et al. 2013). In one study, TNFα-dependent AQP9 upregulation in synovial lining cells as well as mesenchymal cells of osteoid and rheumatoid arthritis patients was observed (Nagahara et al. 2010). These findings are supported by a mechanistic study, identifying an AQP9-dependent role of neutrophils in a murine model of psoriasis (Moniaga et al. 2015). We have thus studied AQP9 expression and a potential modulation thereof by the TNFα inhibitor Humira. In principle, our results support the findings by Nagahara et al. (2010) with regard to TNFα-dependent regulation of AQP9 expression. However, AQP9 expression in rheumatoid arthritis seems to be moderate and limited to a subset of patients. We emphasize that this aspect of our study was exploratory, and does not afford general conclusions due to the small number of biopsies analyzed.

Furthermore, AQP9 expression was studied in 46 cell lines. AQP9 mRNA expression was low or absent in most cases and AQP9 protein expression was identified in very few cases accordingly. LP-1 malignant myeloma cells that have been characterized as showing an early plasma B-cell phenotype (Pegoraro et al. 1989) represented the only notable exception and expressed high levels of AQP9; however, mRNA expression has not been investigated in this cell line. On the other hand, we did not detect expression of AQP9 protein and AQP9 mRNA in HepG2 hepatocellular carcinoma cells. These cells have occasionally been used to investigate AQP9 expression and function before (Lee et al. 2005; Rodriguez et al. 2011; Yokoyama et al. 2011; Gu et al. 2015; Lebeck et al. 2015; Qiu et al. 2015). The discrepancy may be explained by specific cell culture conditions that could have affected AQP9 expression.

Based on a systematic evaluation of AQP expression in a large spectrum of normal and cancer tissues, we find that AQP9 is expressed in a narrow range, with hepatocytes as the only major expression site in normal human tissues. AQP9 has previously been considered as a potential drug target in type 2 diabetes (Rojek et al. 2007), and the observed tissue-specific expression may be beneficial for drug development from a safety perspective.