|Home | About | Journals | Submit | Contact Us | Français|
Assessing the molecular control of development and cell fate in individual cells in the intact mammary epithelium has not been possible to date. By exploiting an intraductal retrovirus (RCAS)-mediated gene delivery method to introduce a marker gene, we found that ductal epithelial cells are turned over with a half time of approximately one month in adult virgin mice. However, following RCAS-mediated introduction of a constitutively activated STAT5a (caSTAT5a), caSTAT5a-activated ductal epithelial cells expand and replace other cells in the epithelium, eventually forming a mammary gland resembling that in a late pregnant mouse, suggesting that STAT5a activation alone is sufficient to mediate pregnancy-induced mammary cell expansion, alveolar cell fate commitment, and lactogenesis. Furthermore, such caSTAT5a-induced alveolar differentiation does not require ovarian functions, although caSTAT5a-induced cell proliferation is partly reduced in ovariectomized mice. In conclusion, in this first report of studying the developmental role of a gene in a few cells in a normally developed virgin mammary ductal tree, STAT5a activation causes alveolar fate commitment and lactogenesis, and with the help of ovarian hormones, drives alveolar expansion.
The mammary ductal tree in a mature virgin mouse is composed of an inner layer of epithelial cells and an outer layer of myoepithelial cells. Under the influence of ovarian hormones, a low level of proliferation – less than 5% – is maintained to replenish cells that die off. However, the exact cell turnover rate of the ductal epithelial cells in a mammary gland is not known. Commonly available genetic approaches including knockout and transgenic models usually are not appropriate for answering this question, since individual cells usually cannot be traced. Consequently, it is also not known what regulates the rate of epithelial cell turnover in the mammary gland.
During pregnancy, this relatively quiescent mammary epithelium undergoes rapid expansion and differentiation, forming densely populated lobuloalveoli. Many transcription factors are known to be required for mammary cell proliferation and alveolar differentiation. One of the best studied of these is STAT5a, a member of the signal transducer and activator of transcription (STAT) family of transcription factors. In the mammary gland, STAT5a is activated primarily by prolactin binding to its cognate receptor, while STAT5b is activated primarily by growth hormone (GH). In knockout mouse studies, STAT5a was found to be required for alveologenesis and alveolar progenitor cell survival. Ablation of Elf5, both a transcriptional target of STAT5a and a transcriptional activator of the STAT5a gene, resulted in loss of lobuloalveolar structures. In support of the essential role of STAT5a in alveologenesis, a constitutively active STAT5a-Gyrase B fusion protein has been reported to induce genes encoding casein and whey-acidic protein (WAP) in a cultured cell line (KIM-2). Furthermore, transgenic expression of wild-type STAT5a or a constitutively active variant of STAT5a was found to promote differentiation during pregnancy.
We have recently reported the adaptation of a retrovirus gene delivery method for genetic manipulation of individual ductal epithelial cells in an intact mammary gland. In this system, an expression vector (RCAS) modified from avian retrovirus subgroup A is used to stably integrate a gene of interest (GOI) into the chromosomes of cells in a mouse line (MMTV-tva) that are made susceptible to RCAS infection by transgenic expression of the gene encoding TVA, the RCAS virus receptor. Furthermore, in mammalian cells, proviral RCAS does not produce any detectable amounts of viral gene products other than the GOI. Since the virus is delivered intraductally, ex vivo infection and transplantation are not necessary, eliminating the complications associated with ex vivo manipulations and loss of mammary structural integrity. Using this approach, we examined the rate of mammary ductal epithelial cell turnover, and studied the impact of STAT5a activation on the turnover rate and the cell fate in the absence of a pregnancy or any ovarian hormone.
Intraductal injection of 107 IUs of RCAS-GFP into 12-week-old virgin MMTV-tva mice led to an infection rate of 0.25 ± 0.08% mammary cells per gland (Toneff et al, unpublished). In order to estimate the cell turnover rate of mammary ductal epithelial cells tagged by RCAS, here we chose to use a virus that makes an endogenous protein to avoid any toxicity or immune response that may associate with the expression of GFP. We created a reporter RCAS virus expressing murine β-actin tagged by a single HA epitope. (The HA tag encoded by RCAS vectors does not elicit any detectable immune response in the mammary gland.) We injected this virus (approximately 107 IUs per gland; 3 glands per mouse) intraductally into 15 MMTV-tva mice at 12 weeks of age, and euthanized 5 infected mice each at days 4, 14, and 31. Using immunohistochemical staining for the HA epitope, we surveyed the number of infected cells in 10 sections across the depth of the entire mammary gland. As expected, infected cells were predominantly luminal (Figure 1A). (Only 0.4 ± 0.2% of positively stained cells expressed the keratin 5 basal cell marker at day 4 post infection [Toneff, et al, unpublished]). While the number of positively stained cells changed very little from day 4 to 14, it decreased by approximately 50% by day 31 (Figure 1B). Altogether, these data suggest that mammary epithelial cells in adult virgins turn over relatively slowly, with a half life of approximately 30 days. The loss of these virus-tagged cells is most likely due to apoptosis, since 2.1 ± 0.1% of cells of the infected ducts were TUNEL positive (n = 3 mice) (Figure 1C).
To examine whether STAT5a activation is sufficient to cause mammary cell expansion, we first cloned a FLAG-tagged constitutively activated version of murine STAT5a into the RCAS vector (Supplemental figure 1A). Forty eight hours after transfecting this vector (RCAS-caSTAT5a) into the DF1 chicken fibroblast producer line, we collected the cell lysates and validated by western blotting that caSTAT5a(FLAG) was produced as an intact protein (Supplemental figure 1B), and by immunofluorescence staining that caSTAT5a(FLAG) was localized to the nucleus (Supplemental figure 1C). Furthermore, by injecting the resulting viral particles intraductally into MMTV-tva transgenic mice and staining the infected glands four days later, we confirmed that caSTAT5a(FLAG) could be detected in the nucleus in a small subset of mammary luminal epithelial cells (Supplemental figure 1D).
Using this virus, we infected the contralateral glands of the same 15 mice that we used for infection with RCAS-β-actin(HA), and again quantified the number of infected cells at days 4, 14, and 31, in a manner similar to that described. caSTAT5a(FLAG)-positive cells expanded by 3.2-fold by day 14 (from 2.5 ± 0.4 cells per CM2 of a section to 8.2 ± 3.3-cells per CM2), and by 10.9-fold by day 31 (to 27.8 ± 3.3 cells per CM2) (Figure 1B). These data suggest that STAT5a activation causes the otherwise relatively quiescent ductal epithelial cells in virgin mice to undergo rapid expansion. Accordingly, in a given ductal section that showed evidence of infection (by finding at least one positively stained cell), the average number of infected cells increased by 3.1-fold by day 14 (from 4.3 ± 1.3% to 13.3 ± 1.9%), and by 9.9-fold by day 31 (to 42.3 ± 4.2%) (Supplemental figure 2). In contrast, the average number of RCAS-β-actin(HA)-infected cells per infected ductal section remained relatively constant by day 14 (from 7.8 ± 1.7% on day 4 to 8.0 ± 1.1% on day 14), and decreased by 44% by day 31 (to 4.4 ± 1.4%) (Supplemental figure 2). Moreover, compared with RCAS-β-actin(HA)-infected cells, RCAS-caSTAT5a(FLAG) infected cells had a much lower level of apoptosis (0.2 ± 0.2%, p=0.8e−03, Figure 1C) at one month post infection. These data indicate that STAT5a activation provides a survival signal for mammary epithelial cells and slows their normal turnover. In addition, caSTAT5a also greatly stimulated the proliferation of this otherwise relatively quiescent population of cells, based on immunostaining for Ki67 (Figure 3C).
Having found that STAT5a activation in virgin mammary epithelium causes rapid cell expansion, we next tested whether STAT5a activation affected the differentiation of these ductal epithelial cells. One month following intraductal injection of 12-week-old MMTV-tva mice (n=5) with RCAS-caSTAT5a(FLAG), clusters of alveoli-like structures with lipid droplets within cells and in the luminal space were readily detected in all H&E-stained sections of all injected glands (3–4 clusters per section) (Figure 2A). As expected, no alveolar-like structures were found in the contralateral glands injected with RCAS-β-actin(HA) (Figure 2A). To confirm that these alveoli-like foci were indeed induced by STAT5a activation, we stained these sections for the FLAG epitope tag. All of these alveoli-like foci were positive for the FLAG tag (Figure 2A). In contrast, in sections of mammary glands infected by RCAS-β-actin(HA), the HA tag was found only in single cells scattered in the ductal epithelium (Figure 2A).
To further confirm the differentiation status of these caSTAT5a-induced alveoli-like structures, we stained them for markers associated with milk-producing alveolar cells. Adipophilin is specifically localized in cytoplasmic and secreted lipid droplets in differentiating and lactating mammary glands. Indeed, all alveolar-like structures and their associated luminal spaces were positive for adipophilin (ADPH) as well as β-casein and WAP (Figure 2B), the predominant milk proteins. As expected, glands injected with RCAS-β-actin(HA) did not exhibit any positive staining for these markers (Figure 2B). These data indicate that activated STAT5a alone can induce functional differentiation in mammary ductal epithelial cells and initiate alveologenesis in the absence of pregnancy hormones. In addition, Elf5, which is a known transcriptional target of STAT5a and has been reported to induce alveolar differentiation, was expressed to a greater extent in these alveoli-like structures than in the control ducts – 34.2 ± 3.2% of cells in caSTAT5a-postive structures showed high staining intensity, while only 4.2 ± 2.4% of cells in β-actin-positive ducts did (n=4 mice, p=2.4e−07) (Figure 2B and Supplemental figure 3).
Reproductive hormones released by ovaries – in particular, estrogen and progesterone – are essential for mammary epithelium expansion during several phases of mammary gland development, including alveolar bud formation and expansion. In order to test whether the above observed effects of caSTAT5a are dependent on ovarian hormones, we intraductally injected ten 12-week-old virgin MMTV-tva mice with RCAS-caSTAT5a(FLAG), performed ovariectomy or sham surgery four days later, and analyzed the infected mammary glands after another four weeks. Ovariectomy prevented the formation of any alveolar clusters (Figure 3A); however, lipid droplets and milk proteins were still readily visible in the cytoplasm and lumen of the ducts that were found to be positive for the FLAG tag, by H&E and by immunostaining for adipophilin and β-casein (Figure 3A & B), although WAP expression was modestly reduced (data not shown). As in intact mice, elevated Elf5 was also found in the infected ducts in these ovariectomized mice (data not shown). These data suggest that caSTAT5a-induced alveolar differentiation does not require ovarian functions, but that caSTAT5a-induced cell expansion is at least partially dependent on the continued presence of reproductive hormones.
To detect any residual proliferative effect of caSTAT5a in the absence of ovarian hormones, we stained for Ki67 in sections from the above RCAS-caSTAT5a(FLAG)-infected ovariectomized mice (n=4). caSTAT5a(FLAG)-positive alveoli contained 9.6 ± 2.5% Ki67-positive cells, significantly higher than the basal level of proliferation (4.1 ± 0.3%) (Figure 3C), indicating that STAT5 activation is capable of inducing mammary cell proliferation in the absence of ovarian hormones. However, this level of proliferation is significantly less than that observed for caSTAT5a-induced proliferation in intact mice (24.5 ± 2.1%); thus, ovarian hormones still play a significant part in alveolar expansion stimulated by activated STAT5a. Of note, lower proliferation in caSTAT5a-positive cells in ovariectomized mice is not an artifact of reduced caSTAT5a expression: We have reported that the RCAS LTR is not affected by ovariectomy, and we have confirmed that caSTAT5 was produced at similar levels in intact vs. ovariectomized infected mice based on staining intensity of immunohistochemistry and immunofluorescence (Supplemental figure 4).
Having observed high levels of both proliferation and differentiation in these virgin glands at one month following activation of STAT5a, we next asked whether these scattered foci of alveolar clusters would regress, as expected for differentiated cells, or would resist the normal cell turnover and continue to expand with time. To this end, we infected an additional cohort of MMTV-tva mice (12 weeks old; n=20) with RCAS-caSTAT5a(FLAG). One year post infection, the RCAS-caSTAT5a(FLAG)-infected glands appeared white, and harbored a milk-like liquid (data not shown). By whole mount staining, the majority of the infected glands were composed entirely of dense lobuloalveolar structures, resembling a late pregnant gland (Figure 4A). Lipid droplets within cytoplasm and in the lumen were observed throughout the gland (Figure 4A), and these droplets produced ADPH, WAP and β-casein based on immunohistochemical staining (Figure 4B). As expected, nearly all epithelial cell as well as some of the basal cells in these alveoli stained positive for caSTAT5a(FLAG) (Figure 4A and data not shown), and Elf5 was more highly produced in those lobuloalveoli than in uninfected control mammary ducts (Figure 4B). Interestingly, based on immunohistochemical staining with an antibody against total STAT5, the total STAT5 in the caSTAT5a-cells did not appear to have increased over the basal level of STAT5 detected in non-infected control cells, but all detectable STAT5 was in the nucleus, suggesting that even the endogenous STAT5 was activated in these differentiated alveolar cells (Supplemental figure 5). In aggregate, activation of STAT5a alone in mammary epithelial cells in virgin glands can induce cell expansion, a commitment to alveolar fate, and persistent lactogenesis.
Our in vivo studies in intact and developed mammary glands demonstrate that mammary epithelial cells exhibit a relatively long half life, and that STAT5a activation suppresses their normal turnover and causes rapid cell expansion, alveolar differentiation, and persistent lactogenesis. Unexpectedly, the entire mammary gland can be replaced by the progeny of approximately 0.3% of the cells that were initially transduced with retrovirus expressing caSTAT5a. Importantly, caSTAT5a-induced differentiation and lactogenesis do not require ovarian functions, although the caSTAT5a-induced cell proliferation partly depends on ovarian hormones.
Traditional studies of cell turnover in the mammary gland have only provided the net gain or loss of all cells or a subset of cells in whole tissue. There are a paucity of tools for cell fate tracking of individual cells in intact mammary glands, although mosaic analysis with double markers (MADM) and other genetic recombination methods have been used to mark specific cells in other epithelial tissues. Our approach provides the opportunity to introduce genetic markers or potential cell fate regulators into individual ductal epithelial cells, so that the cell fate of individual cells in the context of a normal surrounding environment can be examined. With this approach, we demonstrated a slow turnover of the ductal epithelial cell population. Of note, the MMTV LTR, which directs the tva expression and thus the spectrum of the cells that are susceptible to infection in our mice, is most strongly active in differentiated ductal epithelial cells. While these more differentiated infected cells turn over relatively quickly; the less differentiated cell population that is not targeted by this virus may remain for a much longer time.
With this cell-tracking approach, we discovered a master role of STAT5a activation in driving both proliferation and alveolar differentiation of ductal epithelial cells. These results also reveal the extraordinary potential of ductal epithelial cells in proliferation and bipotential differentiation into both alveolar cells and myoepithelial cells. Our results are consistent with previous reports concluding an essential role for STAT5a in regulating mammary alveolar differentiation, and a new report that transplantation of mammary cells infected by a lentivirus expressing an activated version of STAT5 can induce alveoli-like expansion in cleared fat pads in a virgin host.
Approximately 0.3% of the mammary epithelial cells are infected in this viral system. This pool of cells diminished over time (based on the result using RCAS-β-actin), suggesting little progenitor potential among them and consistent with the characteristics of the MMTV promoter which defines the expression of tva and thus the cells susceptible to infection. However, this small pool of cells, after gaining caSTAT5a, expanded and replaced the entire mammary gland after one year. The positive staining of caSTAT5 was even observed in the basal layer. These data suggest that STAT5a activation can induce a bipotential progenitor potential in this RCAS-infected subset of relatively quiescent and differentiated mammary ductal epithelial cells; although the infected cells were likely heterogenous and only a subset of them might have the potential to undergo bipotential expansion and differentiation. Nevertheless, our results are consistent with previous reports of a role of STAT5 in regulating alveolar progenitors and stem cells, and provide a potential molecular mechanism for the generation of the so called “parity-induced mammary cells”, which have been found to have stem and progenitor properties.
Alveolar cells are reported to be produced and quickly turned over during the estrus cycle in virgin mice, and lactogenic cells undergo involution rapidly especially after the cessation of suckling. Therefore, it is a surprise that these caStat5-induced highly differentiated and lactogenic cells did not disappear over time, but rather expanded and persisted in virgin animals. This observation may be due to reported antiapoptotic roles of STAT5a, which transcriptionally activates Bcl2 and BclXL. Persistent STAT5a signaling may also interfere with STAT3 activation, which is known to occur at the end of lactation to cause mammary cell apoptosis through transcriptional activation of the genes encoding suppressor of cytokine signaling 3 (Socs3) and CCAAT/enhancer binding protein delta (C/EBPδ). Indeed, we did not detect activated STAT3 in these infected glands by immunohistochemical staining (Supplemental figure 6), and nor did we detect an increase in the level of apoptosis over a low baseline level in the non-infected glands based on a TUNEL assay (data not shown). However, we do not know whether involution may activate STAT3 in these caSTAT5a-expressing cells and cause them to undergo apoptosis if the infected mice were allowed to progress through a reproductive cycle.
Like prolactin-Jak2-STAT5 signaling, estrogen and progesterone signaling have also been linked to both mammary cell proliferation and differentiation. These pathways also appear to interact with each other. For example, acute treatment of ovariectomized mice with estrogen and progesterone has been shown to induce STAT5a expression; conversely, STAT5a can induce ER transcription. However, it was not previously known whether STAT5-regulated mammary cell proliferation or differentiation actually requires these ovarian hormones. Our results provide strong evidence that in the absence of any ovarian function, STAT5a activation is sufficient to cause ductal epithelial cells to undergo alveolar differentiation and lactogenesis as well as a modest increase in cell proliferation. However, ovarian hormones can significantly increase the proliferative response to STAT5a activation.
Transgenic overexpression of wild-type STAT5a or an activated version of STAT5a (which is a fusion of part of STAT5a with the transcriptional activation domain of STAT6 and the kinase domain of Jak2), or even a dominant-negative version of STAT5a has been reported to occasionally cause tumors with a long latency (8–12 months) after repeated reproductive cycles. Occasional tumors also arose after the first or second round of pregnancy and lactation in mice that have been transplanted with mammary cells ex vivo infected with a lentivirus expressing STAT5a(S711F). Only one tumor was observed in twenty RCAS-caSTAT5a-infected mice followed for over one year, and at necropsy two glands were found to have a focal hyperplasia based on whole mount and H&E staining. The rare progression of caSTAT5a+ cells in our study may be due to a smaller pool of STAT5a-activated cells, a different subset of cells with STAT5a activation, a different expression level of activated STAT5a, a different version of activated STAT5a, or the tumor-inhibitory effects of the surrounding normal and intact virgin ductal epithelium.
In conclusion, using a retrovirus-mediated in vivo gene expression approach in developmentally normal mammary glands, we discovered that the ductal epithelial cells in the adult virgins turned over slowly, that STAT5a activation leads to rapid cell expansion with concurrent alveolar differentiation and persistent lactogenesis, and that these effects are largely independent of ovarian functions.
The MMTV-tva transgenic mice (on the FVB background) have been reported. Animal experimentation was approved by the Institutional Animal Care and Use Committee of Baylor College of Medicine, and conducted under the guidelines described in the NIH Guide for the Care and Use of Laboratory Animals.
The plasmid pMX-STAT5a1*6-FLAG plasmid carrying murine STAT5a with two activating mutations (H299R and S711F) was kindly provided by Dr. Toshio Kitamura, University of Tokyo, Tokyo, Japan. This caSTAT5 gene along with its FLAG tag at its C-terminus was released by restriction digestion with EcoRI and NotI, blunt-ended by Klenow fragment of DNA Polymerase II and cloned into the PmeI site of RCAS-Y, a modified version of RCASBP(A). The insert sequence of this plasmid, RCAS-caSTAT5a(FLAG), was confirmed by DNA sequencing. RCAS-β-actin(HA) has been described elsewhere.
DF-1 chicken fibroblasts were transfected with RCAS vectors using the Superinfect transfection reagent (Qiagen, Valencia, CA), and maintained in DMEM supplemented with 10% FBS in 37°C incubator with 5% CO2. The virus supernatant was concentrated 100-fold by centrifugation at 125,000× g for 90 minutes. The virus pellet were resuspended in medium and prepared as aliquot. The titer of virus was determined by limiting dilution in DF1 cells. Virus is delivered into mouse mammary gland by intraductal injection as described.
Mammary glands were removed and fixed in 10% formalin at 4°C overnight. Tissue were embedded in paraffin, sectioned to 3-μm, and stained with hematoxylin and eosin (H&E). For whole mount analysis, the inguinal mammary glands were fixed in acetic acid/ethanol for 2–4 hours at room temperature and stained with carmine alum.
Immunostaining was performed on de-paraffined tissues sections after antigen retrieval (simmering in 0.1M citric acid, pH6.0, at 145°C for 10 minutes in a pressure cooker). For immunoperoxidase staining, the Vectastain Elite ABC system (Vector Laboratories, Burlingame, CA) was used following manufacturer's instructions. The antibodies used were anti-HA (HA.11, 1:500, Covance), anti-FLAG (M2, 1:500, Sigma), and anti-Ki67 (2011-11, 1:500, Novacastra). For immunofluorescence staining, the following primary antibodies were used: anti-adipophilin (RDI-PROGp40; 1:100; Fitzgerald Industries), anti-Ki67 (2011-11; 1:500; Novacastra), anti-FLAG (M2; 1:500; Sigma-Aldrich), anti-WAP (H-131, 1:200, Santa Cruz Biotech), and anti-STAT5a (L-20, 1:100, Santa Cruz Biotech). The sections were incubated with primary antibodies one hour at room temperature, followed by 30-minute incubation with Alexa Fluor 568 or Alexa Fluor 488-conjugated anti-rabbit antibodies (1:500, Molecular Probes) or FITC-conjugated anti-mouse antibodies (1:500, BD Biosciences). The slides were mounted with Vectashield (Vector Laboratories).
Apoptotic cells were determined by the DeadEnd Fluorometric TUNEL System (Promega). DAPI counterstain was used to visualize nuclei. TUNEL-positive cells were scored in at least 5 fields per section.
A two-tailed Student's t test was applied for all comparisons, and p values of 0.05 or less were defined statistically significant. Continuous variable results were always reported as Mean ± SEM for each group. Pearson's Chi-squared test was applied for analyzing the quantification of Elf5 staining.
(A) Diagram of the RCAS vector with caSTAT5a(FLAG) inserted behind the second splice acceptor. (B) Western blotting of cell lysates of DF1 cells transfected (lanes 1 & 2) or infected (lane 3) with RCAS-caSTAT5a(FLAG), or of DF1 cells transfected with RCAS-GFP (lane 4). The lysate from non-transfected DF1 cells was included as a control (lane 5). An anti-FLAG antibody was used to detect the FLAG-tagged caSTAT5a (arrowhead). (C) Immunofluorescence staining for FLAG in RCAS-caSTAT5a(FLAG)-transfected DF1 cells. (D) Immunohistochemical staining of paraffin sections of mammary glands from adult MMTV-tva mice infected with RCAS-caSTAT5a(FLAG). The infected mice were euthanized after 4 days. A non-infected gland was also stained for comparison.
The anti-HA and anti-FLAG stained sections as described in the legend of Figure 1 were evaluated for expansion of infected cells within a duct. The percentages of infected cells per infected ductal section were determined by counting at least 3–5 sections per mouse, and were plotted in logarithmic scale.
The sections as described in the legend of Figure 2 were stained for Elf5 by immunohistochemistry. The percentages of Elf5-negative, -low or -high cells were determined by counting at least 5 infected ducts per section (N=4 mice per group), and analyzed by Pearson's Chi-squared test. The P value is indicated.
The sections as described in the legend of Figure 3 were stained for FLAG by immunohistochemistry and immunofluorescence.
The sections as described in the legend of Figure 4 were stained for total STAT5 by immunofluorescence. Arrowheads indicate cytoplasmic staining of STAT5, and arrows indicate nuclei staining of STAT5.
The sections as described in the legend of Figure 4 were stained for phospho-STAT3 by immunohistochemistry. Normal involuting mammary glands were included as a positive control.
The authors thank Drs. Darryl Hadsell, Monique Rijnkels, Michael Lewis, and Lothar Hennighaussen for stimulating discussions of the work presented in this manuscript; the Pathology Core Facility at the Breast Center for tissue processing; the Transgenic Mouse Facility for animal husbandry; and Dr. Gary Chamness for helpful comments on the manuscript. This work was supported in part by a U.S. Department of Defense CDMRP grant BC073703 (to Y.L.) and NIH grants CA113869 and CA124820 (to Y.L.). J.D is supported by the SPORE career development award (P50-CA058183).
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.