An immunohistochemical identification key for cell types in adult mouse prostatic and urethral tissue sections

Though many methods can be used to identify cell types contained in complex tissues, most require cell disaggregation and destroy information about where cells reside in relation to their microenvironment. Here, we describe a polytomous key for cell type identification in intact sections of adult mouse prostate and prostatic urethra. The key is organized as a decision tree and initiates with one round of immunostaining for nerve, epithelial, fibromuscular/hematolymphoid, or vascular associated cells. Cell identities are recursively eliminated by subsequent staining events until the remaining pool of potential cell types can be distinguished by direct comparison to other cells. We validated our identification key using wild type adult mouse prostate and urethra tissue sections and it currently resolves sixteen distinct cell populations which include three nerve fiber types as well as four epithelial, five fibromuscular/hematolymphoid, one nerve-associated, and three vascular-associated cell types. We demonstrate two uses of this novel identification methodology. We first used the identification key to characterize prostate stromal cell type changes in response to constitutive phosphatidylinositide-3-kinase activation in prostate epithelium. We then used the key to map cell lineages in a new reporter mouse strain driven by Wnt10aem1(cre/ERT2)Amc. The identification key facilitates rigorous and reproducible cell identification in prostate tissue sections and can be expanded to resolve additional cell types as new antibodies and other resources become available.

Though many methods can be used to identify cell types contained in complex tissues, most require cell disaggregation and destroy information about where cells reside in relation to their microenvironment. Here, we describe a polytomous key for cell type identification in intact sections of adult mouse prostate and prostatic urethra. The key is organized as a decision tree and initiates with one round of immunostaining for nerve, epithelial, fibromuscular/ hematolymphoid, or vascular associated cells. Cell identities are recursively eliminated by subsequent staining events until the remaining pool of potential cell types can be distinguished by direct comparison to other cells. We validated our identification key using wild type adult mouse prostate and urethra tissue sections and it currently resolves sixteen distinct cell populations which include three nerve fiber types as well as four epithelial, five fibromuscular/hematolymphoid, one nerve-associated, and three vascular-associated cell types. We demonstrate two uses of this novel identification methodology. We first used the identification key to characterize prostate stromal cell type changes in response to constitutive phosphatidylinositide-3-kinase activation in prostate epithelium. We then used the key to map cell lineages in a new reporter mouse strain driven by Wnt10a em1(cre/ERT2)Amc . The identification key facilitates rigorous and reproducible cell identification in prostate tissue sections and can be expanded to resolve additional cell types as new antibodies and other resources become available. PLOS

Introduction
The GenitoUrinary Development Molecular Anatomy Project (GUDMAP, www.gudmap.org) is a multi-laboratory consortium dedicated to providing the scientific and medical communities with hypothesis-generating data and tools to facilitate research. A recent initiative is to build a repository of annotated genitourinary tract immunohistochemical images from cre expressing reporter mouse strains. The image repository will facilitate mouse strain selection by investigators, critical evaluation of research results by manuscript and grant reviewers, and generally enhance the rigor and reproducibility of cre/lox research studies. The most significant challenge in developing this repository is to accurately assign cre lineage-labels to known genitourinary cell types. We considered multiple approaches for identifying lineage labeled cells including standard immunostaining, cell sorting, and RNA sequencing. A single round of immunostaining is a possible approach for some applications but is insufficient for comprehensive cell identification in complex tissue sections. For example, while a single round of immunostaining can be deployed to distinguish one cell type from a limited pool of closely related cells in culture (e.g. myofibroblasts from fibroblasts), the sheer diversity of cells in an intact tissue section (e.g. myofibroblasts, fibroblasts, fibrocytes, myocytes, pericytes) substantially challenges single round immunostaining for cell identification in situ [1,2]. Cell sorting and single cell RNASeq address the challenge of differentiating closely related cell types in complex tissues, but destroy tissue organization, cell interactions, and information about a cell's spatial location.
We sought a comprehensive method for identifying cell types in tissue sections and were inspired by the polytomous and dichotomous identification keys used in taxonomy and phylogenetics [3]. Stepwise observations are used to systematically rule out potential cell identities until a final determination can be achieved. An identification key is diagnostic in that it can be used to distinguish a specific cell type from a broader class of cells and is differential in that it can be used to distinguish one cell from another. Immunostaining is well suited for decision making in cell identification keys because it reduces data dimensionality to a dichotomous variable: cells are either stained or unstained. We tested over 70 antibodies to identify antibody combinations (multiplexes) with the greatest power to resolve subsets of prostatic nerve fibers, epithelial cells, fibromuscular and hematolymphoid cells, and perivascular cells. We then constructed a polytomous key which organizes a series of multiplex immunostains into an optimal sequence for comprehensive cell type identification. Potential cell identities are recursively eliminated by each round of staining until cells are definitively distinguished by direct comparison with other cells. Here, we describe our mouse prostate and urethral cell identification key and provide images of identified cell types and a list of validated antibodies for multiplex immunostaining in paraffinembedded mouse prostate tissue sections. We also demonstrate two uses of our cell identification key: objectively describing stromal cell distribution changes in a new genetically-induced mouse model of prostate cancer and identifying lineage labeled cells in a new cre-expressing mouse reporter strain. We anticipate this key will serve as a foundational framework for cell identification and will be broadened in the future to include additional cell types, tissues, and species.
Genotyping was conducted as described by Jackson Laboratories. Mice were housed in Udel 1 Polysulfone microisolator cages; the room was on 12-h light and dark cycles; room temperature was typically 20.5 ± 5˚C; humidity was 30-70%. Mice were fed a 5015 Diet (PMI Nutrition International, Brentwood MO) from conception through weaning (PND 21) and an 8604 Teklad Rodent Diet thereafter (Harlan Laboratories, Madison WI). Feed and water were available ad libitum, and cages contained corn cob bedding. To activate cre in Wnt10a creERT2 ; tdtomato fl/fl mice and their genotypic controls, male mice were given a single intraperitoneal injection of tamoxifen (100 mg/kg, Sigma #T56482, St.Louis, MO) dissolved in sterile corn oil at postnatal day 3 and prostates and urethras were evaluated two months later. Prostates and urethras were evaluated in four month old Pbsn4 cre ;PIK3ca Ã mice. All other tissues were evaluated at postnatal day 50. All mice were euthanized by CO 2 asphyxiation.

Immunohistochemistry
Tissue sections (5-15 μm thickness) were deparaffinized with xylene and rehydrated with graded ethanol. Sections for histological analysis were routinely stained with hematoxylin/ eosin or fluorescent immunostaining was conducted as described previously with two modifications to the protocol [8]. A decloaking chamber (Model DC2002, Biocare Medical, Pacheco, CA) was used with either 10 mM sodium citrate (pH 6.0) or 10 mM Tris-EDTA (pH 9.5) for antigen unmasking and non-specific sites were blocked for 1 hr in TBSTw containing 1% Blocking Reagent (Roche Diagnostics, Indianapolis, IN), 5% normal donkey sera, and 1% bovine serum albumin fraction 5 (RDBTw). Antibodies are listed in S1 Table. Some tissue sections were imaged using an Eclipse E600 compound microscope (Nikon Instruments Inc., Melville, NY) fitted with a 20x dry objective (Plan Fluor NA = 0.75; Nikon, Melville, NY) and equipped with NIS elements imaging software (Nikon Instruments Inc.) Fluorescence was detected using DAPI (2-(4-amidinophenyl)-1H -indole-6-carboxamidine), FITC, Texas Red (Chroma Technology Corp, Bellows Fall, VT), and CY5 filter cubes (Nikon, Melville, NY). Other sections were imaged using an SP8 Confocal Microscope (Leica, Wetzlar, Germany) fitted with a 20x oil immersion objective (HC PL Apo CS2 NA = 0.75; Leica, Wetzlar, Germany). Samples were excited and detected using the recommended settings for each secondary antibody fluorophore. Images were captured at 1024x1024 resolution using LASX 8 software (Leica, Wetzlar, Germany). For Pbsn4 cre ;PIK3ca Ã mice, one sagittal section of lower urinary tract was stained and imaged from two litter independent mice. For all other studies one sagittal section of lower urinary tract was stained and imaged from each of at least three mice deriving from at least three separate litters. Representative images of dorsal prostate and ventral prostatic urethra were acquired from each image.

Application of the identification key to mouse prostate and prostatic urethra
The male mouse genitourinary tract consists of multiple organs and the type and distribution of cells within organs differs across body axes. While a complete identification key for all parts of the genitourinary tract is beyond the scope of this study, we focus here on the pelvic portion of the mouse genitourinary tract and its urethral and prostatic constituents (Fig 1A and 1B). To enable cell visualization at single cell resolution, we concentrated on two discrete regions. We examined the dorsal prostate external to the rhabdosphincter (Fig 1C) because, in mice susceptible to prostate cancer, this region develops tumors which exhibit molecular similarities to human prostate cancer [9]. We also examined the ventral prostatic urethra internal to the rhabdosphincter ( Fig 1D). This is the region where many prostatic ducts drain into the urethra and where proliferative and fibrotic pathologies have been observed in mouse models of benign prostatic hyperplasia and urinary obstruction [10].

Nerves
The prostate and prostatic urethra contain neural fibers which transmit efferent signals to control smooth muscle contraction and prostatic secretory function, transmit afferent signals to respond to environmental stimuli, and provide trophic support to nearby non-neuronal cells [11][12][13]. To begin discriminating the different nerve subtypes within the prostate we applied our identification key methodology (Fig 2A). An initial stain using antibodies against myelin basic protein (MBP) and class III beta-tubulin (TUBB3) is performed. From this initial stain, putative Schwann cells (MBP+;TUBB3-) and generalized nerve fibers (MBP-;TUBB3+) are identified (Table 1). MBP+;TUBB3-cells are relatively scarce in dorsal prostate (Fig 2B), but abundant in prostatic urethra (S2B Fig). To further classify nerve fibers, subsequent immunostains with antibodies against tyrosine hydroxylase (TH), calcitonin gene related peptide (CGRP), and solute carrier family 18 (vesicular monoamine), member 3 (SLC18A3, also known as vesicular acetylcholine transporter, VaCHT) are used to create three divisions of TUBB3+ fibers. TH is considered a marker for adrenergic fibers in various tissues (Table 1). TH+;TUBB3+; fibers are abundant in dorsal prostate stroma, often encircling prostate ducts ( Fig 2C). Of the three nerve divisions contained within our existing key, TH+ fibers are the most abundant in dorsal prostate and prostatic urethra. CGRP is a marker of sensory c-fibers in various tissues (Table 1) and CGRP+;TUBB3+ fibers are present throughout the stroma of the dorsal prostate ( Fig 2S), but are particularly concentrated in the prostatic urethra (S2D Fig). We also observe CGRP+;TUBB3-cells in prostate epithelium (S2D Fig) and conclude these are prostate neuroendocrine cells based on their location and frequency relative to other prostate epithelial cells. CGRP has not been described previously as a marker of mouse prostatic neuroendocrine cells, but is present in bile duct and lung neuroendocrine cells [14,15]. SLC18A3 is a marker of cholinergic fibers (Table 1)   The image is a representative 5 μm LUT section immunostained with antibodies against cadherin 1 (CDH1, also known as e-cadherin, red), actin alpha 2 (ACTA2, also known as smooth muscle actin, green) and DAPI (blue). Sequential image tiles were assembled to reveal the entire lower urinary tract. Two regions of interest were captured for validation of cell types in subsequent figures: (C) the dorsal prostate external to the rhabdosphincter and (D) the prostatic urethra, located near the bladder neck and internal to the rhabdosphincter. Abbreviations are: AP, anterior prostate; BL, bladder; DP, dorsal prostate; RS, rhabdosphincter; SV, seminal vesicle; UR, pelvic urethra; VP, ventral prostate; DAPI, 2-(4-amidinophenyl)-1H -indole-6-carboxamidine. https://doi.org/10.1371/journal.pone.0188413.g001

Fibromuscular and hematolymphoid cells
Prostate stroma mediates glandular development [51], prostate cancer growth [52][53][54][55] and fibrosis, the latter of which has been recently linked to urinary dysfunction [56][57][58]. Prostate stroma also harbors smooth muscle which can contract inappropriately and contribute to urinary symptoms in aging men [59]. Furthermore, prostate inflammation is one of the most significant predictors of lower urinary tract dysfunction in men and has been associated with chronic pelvic pain [60][61][62][63]. Pinpointing cell types responsible for behaviors that contribute to clinically significant prostate diseases will guide mechanistic studies of prostate homeostatic regulation and disease processes.
Prostatic fibromuscular and hematolymphoid cells are initially characterized by a single round of immunostaining with an antibody against protein tyrosine phosphatase, receptor type, C (PTPRC), which is present in mast cells, T cells, B cells, fibrocytes, and macrophages (Table 1), and with an antibody against vimentin (VIM), which is present in some hematolymphoid and many non hematolymphoid stromal cells (Fig 3A). Based on results of this first immunostain, additional immunostains are applied to identify specific stromal cell subtypes. We observe occasional PTPRC+ cells that are also actin alpha 2 (ACTA2) positive within the periductal smooth muscle layer, and PTPRC+ cells also rarely reside within prostate epithelium ( Fig 3B). Other potential inflammatory cells, characterized by a ACTA2-;VIM+;PTPRC + staining pattern (Table 1), are scattered throughout interductal stroma (Fig 3B). Fibrocytes have been described as cells with simultaneous expression of PTPRC, ACTA2, and vimentin (VIM) ( Table 1). We observe PTPRC+ cells external to the periductal smooth muscle layer of the dorsal prostate and clustered near blood vessels (Fig 3B). A small subset of these are ACTA2+;VIM+;PTPRC+.
The smooth muscle sheath surrounding mouse prostate ducts consists almost exclusively of smooth muscle myocytes. Previous studies have found that androgen receptor (AR) expressed in the prostatic stroma mediates a significant proportion of morphological and pathological processes. During ductal morphogenesis, prostatic stromal AR is essential for epithelial cell growth. Prostatic stromal AR abundance is also important in prostate cancer progression. Prostate stromal cells begin losing AR expression during cancer progression and low AR expression in prostatic stroma is commonly found in patients who have developed resistance to androgen ablation therapy for prostate cancer [26,29]. To determine the relative abundance of AR in ACTA2+ myocytes of the adult mouse prostate performed IHC divide ACTA2+ myocytes into two subpopulations. The majority of periductal smooth muscle myocytes are ACTA2+;AR+; while ACTA2+;AR-myocytes are rare (Fig 3C, S3C Fig). This observation is consistent with previous findings in rats [28]. Whether these smooth muscle cells are functionally distinct remains to be explored.

Vascular and perivascular cells
The role of angiogenesis in prostate cancer is well recognized [66] and while the role in BPH is still emerging, it is no less important. Microvessel density surrounding epithelial and stromal  BPH nodules is enhanced relative to adjacent normal tissue [67] and is elevated in prostates of men with symptomatic BPH [68,69], particularly those who have failed surgical treatment [70]. Vascular smooth muscles cells regulate prostate and urethral hemodynamics, synthesize blood vessel wall components, and influence endothelial proliferation [71][72][73]. The role of other perivascular cells (pericytes) in adult mouse prostate is unknown, but they likely participate in angiogenesis [71,74,75] and may also contribute to pathological collagen production during organ fibrosis [76][77][78][79].
Prostate vascular and perivascular cell identification requires a multiplex stain with antibodies against ACTA2, PDGFRB, and PECAM (Fig 5A). Endothelial cells have been characterized as ACTA2-;PDGFRB-;PECAM+ (Table 1) and are present in microvessels within prostate periductal smooth muscle and interductal stroma (Fig 5B and 5C, S5B and S5C Fig). Prostatic vascular smooth muscle myocytes were described previously as ACTA2+;PDGFRB-;PECAM-( Table 1) and these cells are associated with a majority of vessels in the prostate and prostatic urethra, but are organized in a discontinuous pattern around vessels (Fig 5B and 5C and S5B  Fig). Pericytes have been previously identified as ACTA2-;PDGFRB+;PECAM- (Table 1). In the dorsal prostate, pericytes are interwoven with ACTA2-;PDGFRB-;PECAM+ cells and are bounded by vascular smooth muscle myocytes and endothelial cells (Fig 5B). We do not observe pericytes in the prostatic urethra (S5B Fig). control mice. Images are representative of three mice. Abbreviations: SYP, synaptophysin; KRT5, keratin 5; KRT8/18, keratin 8/18; RFP, red fluorescent protein; DAPI, 2-(4-amidinophenyl)-1H -indole-6-carboxamidine; Scale bar is 25 μm. Using the identification key to objectively describe fibromuscular cell distribution changes during formation of genetically-induced prostate cancer We next sought to demonstrate a utility of the cell identification key. Stromal thickening and appearance of a reactive stroma accompany many solid tumors, including prostate cancer [55]. However, the precise composition of tumor-reactive stroma and an objective characterization of how it changes in situ over time have not been determined. We used the identification key to determine how prostate fibromuscular cell populations change in a novel mouse model of genetically induced prostate cancer. The mouse strain models activation of PIK3/ AKT signaling which accompanies many solid tumors [80], including prostate adenocarcinoma [81].  6B). We observed fewer periductal ACTA2+ cells in Pbsn4 cre ; PIK3ca Ã mice compared to controls, though the relative abundance of ACTA2+;AR+ to ACTA2+;AR-cells is similar to controls (Fig 6C). Staining for fibroblasts and myofibroblasts indicate that a majority of Pbsn4 cre ;PIK3ca Ã mouse interductal stromal cells are putative fibroblasts (VIM+;ACTA2-;S100A4+) and do not express ACTA2 (Fig 6D).
Using the identification key to objectively describe lineage-labeled cells in a novel cre reporter mouse strain A principal reason for building the prostate cell identification key is to establish an objective mechanism for characterizing cre expressing mouse strains. While several mouse strains have been created for genetically manipulating prostate luminal or basal epithelial cells, there is no central repository of information about the stage of development when the cre transgene is first expressed, whether the cre is expressed across all cells or evenly expressed across prostate lobes, and whether cre expression is the same in the distal and proximal parts of prostatic ducts. Though many studies have used cre expressing mouse strains to manipulate gene expression in prostate stroma, there is no centralized database describing which specific stromal cells express the cre and the cre distribution pattern. To address both of these needs, we are characterizing several cre expressing mouse strains, including some generated by GUD-MAP, and will use the identification key for cre validation.
To demonstrate proof of concept, we used the identification key to validate reporter gene distribution for Wnt10a creErt2 . We previously observed that Wnt10a mRNA is selectively expressed in a subset of basal epithelial cells within the female and male mouse urogenital sinus, including prostatic ductal bud epithelium in males [82]. This Wnt10a creErt2 mouse strain was generated to enable genetic manipulation of this epithelial subset. To identify the cell types deriving from this epithelial cell population, Cre was activated in postnatal day 3 male mice (Wnt10a creErt2 ;tdtomato fl/fl ) by administering a single i.p. dose of tamoxifen and tdtomato reporter expression was mapped to prostate epithelial cell types when the mice reached 50 days of age. Tdtomato positive cells are infrequent and are mapped to KRT8/18-;KRT5+;SYP-(basal) and KRT8/18+;KRT5-;SYP-(luminal) epithelial cells of the dorsal prostate and prostatic urethra (Fig 7). Tdtomato is not observed in any KRT8/18+-KRT5-;SYP+ (neuroendocrine) cells of the prostate or prostatic urethra.

Discussion
Here we described a polytomous key for immunohistochemical identification of cell types in adult mouse prostate and prostatic urethra. The key currently identifies sixteen unique cell types and some of its many uses include objectively defining lineage labeled cells, proliferating cells, or changes in cellular distribution in response to aging, inflammation, benign hyperplasia, cancer, or environmental exposures.
We created the mouse prostate cell identification key with the expectation it would expand to accommodate additional cell types as knowledge about prostate cellular composition is advanced by RNASeq, flow cytometry, and other methods. For example, this study utilized PTPRC as a biomarker of cells deriving from a hematolymphoid lineage and we multiplexed PTPRC with ACTA2 and VIM antibodies to identify putative fibrocytes, which derive from hematolymphoid progenitors. Though our initial version of this key does not further resolve hematolymphoid cell types, additional branches can be incorporated into future versions to account for the more than 14 different monoclonal antibodies used to differentiate hematolymphoid cell types [83].
A future opportunity afforded by this key is to objectively characterize changes in prostatic cellular composition during prostate glandular development, response to androgen deprivation, or during regeneration of the castrated prostate. The key is ideally paired with genetic lineage tracing methods that incorporate an indelible label into a cell and its daughter cells. For example, we determined that Wnt10a expressing cells in the postnatal day 5 prostate give rise to a limited subset of prostatic basal and luminal epithelial cells in the adult mouse but do not appear to vigorously expand and thus likely have limited progenitor activity.
We also used our key to characterize prostate stromal cell distribution changes in mice expressing a constitutively active form of PI3K in prostate epithelial cells. We observed formation of a new stromal microenvironment that closely resembles the reactive stroma observed near human cancer tumors. In many human cancers, including prostate, a major cellular component of the reactive stroma are myofibroblasts [84]. To differentiate myofibroblasts from smooth muscle cells in prostate stroma, earlier studies used complementary but not sequential IHC. These studies found that human prostate cancer is characterized by increased percentages of VIM+ cells, with no change in ACTA2 positivity [85]. Our results expand on previous findings by using a wider panel of antibodies to examine the reactive stroma with single cell resolution and found that in Pbsn4cre;PIK3ca Ã mice, putative fibroblasts (S100A4+;VIM+; ACTA2-) dominate the reactive stroma. Although a majority of the cells within the prostate stroma of Pbsn4cre;PIK3ca Ã mice are ACTA2-and likely not myofibroblasts, the potential for the fibroblasts present in the stroma to undergo phenoconversion to a myofibroblast-like phenotype and begin producing ACTA2 as disease progresses is highly plausible. Also, because the stromal response to pathologies such as cancer differs among mouse strains [86][87][88], investigators can use our key to compare the stromal cellular makeup among strains and at differing spatial locations with respect to sources of inflammation or tumor boundaries. Similar comparisons can be made between or among species. Although initially validated for mouse prostate, many of the antibodies used in our key are advertised to work in multiple species and organ systems and validating them across species is a future goal.