The enhanced green fluorescent protein (GFP) reporter has been widely adopted for tracking cell lineage. Here, we compare three transgenic mouse strains in which GFP is considered “ubiquitously expressed,” with the GFP transgene under control of the chicken β-actin (CBA) or human ubiquitin C (UBC) promoter. We compared the expression of GFP using flow cytometry, direct tissue fluorescence, and immunostaining with multiple commercially available anti-GFP antibodies. Mice of CBA-GFP strain 1Osb have strong but variegated expression of GFP in adult liver, kidney, small intestine, and blood. Mice of CBA-GFP strain Y01 have the highest proportion of GFP-positive peripheral blood cells yet limited GFP expression in liver, intestine, and kidney. UBC-GFP mice express GFP only weakly in solid organs and variably in blood. Direct fluorescent detection of GFP in formalin-fixed, paraffin-embedded tissue sections was the simplest approach, but it was useful only in high-expressing strains and potentially subject to artifact because of tissue autofluorescence. Immunofluorescence using either primary goat or primary rabbit antibodies was much more sensitive and allowed better discrimination of authentic signal from autofluorescence. Immunohistochemical staining was less sensitive than direct fluorescence or immunofluorescence and was subject to false-positive signal in the small intestine. In conclusion, there is considerable variability of expression within and between GFP transgenic strains. None of the tested strains gave truly ubiquitous GFP expression. A detailed analysis of GFP expression in one's tissues of interest must guide the choice of reporter mouse strain when GFP is used as a marker of cell lineage or donor origin.
Disclosure of potential conflicts of interest is found at the end of this article.
Enhanced green fluorescent protein (GFP) has become the marker of choice for many types of cell transplantation and lineage marking experiments. It is readily detected in live, frozen, or fixed cells using established methods and is detected with high sensitivity and specificity in suspended cells by flow cytometry. The sensitivity and relative simplicity of GFP detection make it appealing. However, there are limitations of GFP as a marker (reviewed in ). For example, GFP may lose its direct fluorescence during tissue fixation or subsequent processing. For this reason, immunostaining with commercially antibodies is often used for GFP detection. In addition to these technical concerns, there can be considerable biological variability of GFP expression, even among similar cell types in a single animal . Prior to conducting transplantation or lineage-marking experiments, it is essential to conduct a thorough analysis of the expression pattern of GFP in appropriate control animals.
Several lines of “ubiquitously expressing” GFP transgenic mice are available, yet there can be tremendous variability at multiple levels of expression due to differences in absolute abundance of GFP and/or in the percentage of cells expressing GFP. High-level expression in the blood does not necessarily indicate high-level expression in solid organs. Even mice that appear to glow bright green under appropriate fluorescent light  may not be uniformly green at the level of individual cells.
Strains of GFP mice often have individuals with high, intermediate, or low expression and may be bred selectively for high or low expression . However, high levels of GFP expression may have subtle toxic effects, so that over time, unselected transgenic lines may “drift” toward a relatively lower level of expression.
Judicious selection of GFP transgenic strain and careful optimization of the detection method should improve experimental consistency. In this report, we compare three different transgenic lines of GFP-expressing mice by different detection methods.
Materials and Methods
All mice used in the study were sacrificed at 6–8 weeks of age. The chicken β-actin-enhanced GFP strain C57BL/6-TgN(ACTB-EGFP)1Osb, originally developed by Dr. Masaru Okabe (Osaka University, Osaka, Japan) , was obtained from the Jackson Laboratory (Bar Harbor, ME, http://www.jax.org; stock no. 003291). This transgene consists of a chicken β-actin promoter, cytomegalovirus (CMV) enhancer, β-actin intron, and bovine globulin polyadenylation signal. The strain C57BL/6-TgN(ACTB-EGFP)OsbY01 (also developed by Dr. Okabe and obtained with permission from Dr. Okabe and Dr. William Fleming, The Ohio State University, Columbus, OH) represents the same transgenic construct as the initial strain, but it is integrated into a different chromosomal site . The third GFP mouse strain analyzed, C57BL/6 Tg(human ubiquitin C [UBC]-GFP)30Schaa/J (stock no. 004353; Jackson Laboratory), consists of GFP under the control of the human ubiquitin C promoter, an SV40 polyadenylation signal, and a bovine growth hormone polyadenylation signal . Normal controls were age and sex-matched nontransgenic C57BL/6 mice from the Jackson Laboratory.
Blood was collected from the retroorbital plexus of anesthetized mice into heparinized capillary tubes. Following erythrocyte lysis in ammonium chloride (PharmLyse; BD Biosciences, San Diego, http://www.bdbiosciences.com), nucleated cells were washed and resuspended in phosphate-buffered saline (PBS) (pH 7.2) supplemented with 2% bovine serum albumin (BSA) (Sigma-Aldrich, St. Louis, http://www.sigmaaldrich.com). The proportion of cells expressing GFP was determined by flow cytometry using appropriate nontransgenic controls for comparison.
Tissue Collection and Processing
Mice were treated humanely under a protocol approved by the Yale University Institutional Animal Care and Use Committee. Mice were sacrificed under deep anesthesia with ketamine and xylazine. Organs were quickly dissected into sections not more than 5 mm thick and then fixed in formalin for 4 hours at room temperature. Formalin was replaced by 70% ethanol until tissues were embedded in paraffin. Five-micrometer sections were prepared by the Research Histology Service at Yale University.
Slides were heated to 60°C for 2 minutes, deparaffinized in Citrisolv (Fisher Scientific International, Fairlawn, NJ, http://www.fisherscientific.com), and rehydrated through graded ethyl alcohols to PBS. Slides were rinsed in distilled water and air-dried. Vectashield with 4,6-diamidino-2-phenylindole (DAPI) (Vector Laboratories, Burlingame, CA, http://www.vectorlabs.com) and glass coverslips were applied. Slides were viewed under fluorescence microscopy using appropriate filters. Fluorescent emission spectra were acquired using a Leica SP2 confocal microscope (Heerbrugg, Switzerland, http://www.leica.com) equipped with three lasers (argon, 543 nm HeNe, and 633 nm HeNe) and three spectral detectors. Each excitation wavelength was used sequentially, and excitation wavelengths of 458, 488, 514, and 543 nm were selected with an acousto-optical tunable filter with subnanometer resolution. Putative GFP-positive and -negative areas were gated and imaged simultaneously using identical settings, and the mean emission intensity values for each area were measured in 2-nm increments (a λ series) throughout the relevant spectrum. Mean intensity values for the putative GFP-positive areas were normalized to a peak intensity value of 100, and the corresponding background signal from GFP-negative tissue was normalized identically. Positive control tissue included fixed skeletal muscle from 1Osb GFP transgenic mice. Spectra of soluble GFP at pH 7 were obtained from Molecular Probes Inc. (Eugene, OR, http://www.probes.invitrogen.com).
Antigen retrieval was performed using Signet pH-All-2 (product no. 1922; Signet Laboratories, Dedham, MA, http://www.signetlabs.com) for 30 minutes in steam, and solution was then cooled to room temperature and washed in PBS. Endogenous peroxides were blocked using 3% H2O2 in PBS at room temperature for 10 minutes. Endogenous biotin was blocked using the Avidin/Biotin Blocking Kit (Vector Laboratories). Nonspecific binding of immunoglobulins was blocked by incubating slides in a 10% solution of normal serum from the species in which the secondary antibody was prepared (goat or donkey).
The primary antibodies, isotype controls, and secondary antibodies used are indicated in Table 1. Primary antibodies and isotype controls were diluted in PBS supplemented with 1% BSA and applied to tissue sections at a final concentration of 10 μg/ml. Following overnight incubation at 4°C, slides were washed three times in PBS, and secondary antibody was applied at a final concentration of 10 μg/ml in PBS with 1% BSA for 1 hour at 37°C.
Table Table 1.. Antibodies
Biotinylated secondary antibodies were applied at a final concentration of 10 μg/ml in PBS with 1% BSA for 1 hour at 37°C. Antibody complexes were detected using a streptavidin-horseradish peroxidase complex (Vectastain ABC Elite system; Vector Laboratories) and diaminobenzidine (DAB) substrate (Vector Laboratories). Slides were counterstained with hematoxylin (Vector Laboratories), cleared in ethanol, and permanently mounted (Permount; Fisher Scientific).
Slides were deparaffinized and antigen retrieval performed as described above, except that peroxidase and biotin blocking steps were omitted. Primary antibody was rabbit or goat polyclonal anti-GFP (Table 1). Appropriate rabbit or goat IgG isotype negative controls were run in parallel. Nontransgenic tissues served as additional negative controls. Fluorescent-conjugated secondary antibodies were goat anti-rabbit IgG AlexaFluor 568 (A11036; Molecular Probes) or donkey anti-goat IgG AlexaFluor 568 (A11057; Molecular Probes) applied at 10 μg/ml in PBS with 1% BSA for 1 hour at 37°C. Slides were washed in PBS, rinsed in water, and then air-dried. Fluorescence was visualized using Vectashield with DAPI (Vector Laboratories).
Photomicrographs were taken using an Olympus BX51 fluorescence microscope (Tokyo, http://www.olympus-global.com) using bright-field illumination or appropriate fluorescence excitation and emission filters. Fluorescent images were captured with a digital camera, using IPLab software (BD Biosciences). Bright-field images were captured with QCapture Pro software (QImaging, Burnaby, BC, Canada, http://www.qimaging.com).
The percentage of GFP-positive blood cells was compared by the t test. A p value <.05 was considered significant.
Comparison of Anti-GFP Antibodies
To clarify which antibody works best in paraffin sections, we first compared three commercially available anti-GFP antibodies (Table 1) on multiple tissues from mice of the 1Osb strain or nontransgenic controls, using horseradish peroxidase-DAB immunohistochemistry (Fig. 1; supplemental online Fig. 1). In each case, equivalent concentrations of isotype control antibody were run in parallel to test for nonspecific staining.
As shown in Figure 1A, the rabbit polyclonal antibody to GFP stains a subset of hepatocytes in a mouse of the 1Osb strain, with little or no nonspecific staining in isotype (Fig. 1B) or nontransgenic (Fig. 1C) negative controls. The goat polyclonal anti-GFP antibody stains in a similar pattern (Fig. 1D), although with reduced intensity. The mouse monoclonal antibody also stains GFP-expressing hepatocytes in a similar pattern (Fig. 1G), but there was substantial nonspecific background staining seen in endothelial and Kupffer cells in the liver (Fig. 1H, 1I).
Immunohistochemical staining of GFP in the small intestine of the same mouse using rabbit, goat, or mouse primary antibody, compared with isotype and nontransgenic controls, is shown in supplemental online Figure 1. In the small intestine, a high rate of false-positive staining of cells in both the epithelium and lamina propria was seen with isotype (supplemental online Fig. 1B, 1E, 1H) and nontransgenic controls (supplemental online Fig. 1C, 1F, 1I) for rabbit, goat, or mouse primary antibody.
Comparison of GFP Detection Methods
Direct detection of GFP fluorescence in formalin-fixed, paraffin-embedded sections has been regarded by some investigators as insensitive and unreliable because of rapid loss of soluble, cytoplasmic GFP from the tissue during fixation and processing. We found direct detection of GFP to be simple and sensitive in 1Osb mice, where the absolute level of GFP expression is high. We also noted that the intrinsic fluorescence of GFP is not eliminated by antigen retrieval in steam (Fig. 2B, 2E), a testament to the remarkable heat stability of GFP fluorescence in formalin-fixed paraffin sections. We immunostained for GFP with a secondary antibody conjugated to AlexaFluor 568 (Fig. 2C) to distinguish intrinsic fluorescence from immunofluorescence. In the liver of 1Osb mice, where GFP expression is high, the same variegated pattern was observed by direct GFP fluorescence, immunofluorescence, and immunohistochemistry (compare Fig. 2B vs. 2C and Fig. 2E vs. 2F). To confirm that the green signal seen by direct fluorescence was GFP rather than autofluorescence, we performed a spectral analysis (Fig. 3). Background red autofluorescence in the liver is shown in Figure 3A. The pattern is uniform and does not resemble the pattern seen by direct GFP fluorescence (Fig. 3B), indicating that the green fluorescence was not autofluorescence. To confirm that the bright green cells represented GFP rather than autofluorescence, clusters of 1Osb hepatocytes were visually categorized as GFP-bright or GFP-negative (Fig. 3C). These areas are indicated by red and yellow boundaries, respectively, in Figure 3C. Emission spectra from these areas were compared with each other and with soluble GFP protein or skeletal muscle from a 1Osb GFP transgenic mouse (Fig. 3D) to confirm that the clusters of bright green hepatocytes seen by direct fluorescence express bona fide GFP.
GFP Expression in Different Transgenic Lines
GFP Expression in Peripheral Blood.
GFP expression in peripheral blood was assessed in each strain of GFP mice by flow cytometry. Representative histograms for each strain are shown in the top row of Figure 4. Nontransgenic negative controls for each experiment are shown in the second row. Mice of the 1Osb (n = 5) or UBC-GFP (n = 3) strain expressed GFP in only 70%–75% of blood cells, whereas mice of the Y01 strain (n = 5) expressed GFP in 93% of nucleated peripheral blood cells, a significantly greater percentage than either 1Osb mice (p < .0001 vs. Y01) or UBC-GFP mice (p < .05 vs. Y01).
GFP Expression in Organs.
Using immunohistochemistry, GFP could not be reliably distinguished from background in Y01 or UBC-GFP mice (data not shown). We found immunofluorescent detection using the rabbit or goat anti-GFP antibodies to be the most sensitive and reliable method for liver, small intestine, and kidney. Antigen retrieval with the Signet pH-All-2 reagent was superior to standard citrate buffer or no antigen retrieval (data not shown). Therefore, we used immunofluorescence to compare GFP expression in paraffin sections of liver, small intestine, and kidney in the three transgenic strains and compared them with nontransgenic controls. Representative GFP immunofluorescence photomicrographs using the goat anti-GFP are shown in Figure 5. Similar results were obtained with the rabbit anti-GFP (not shown). In mice of the 1Osb strain, GFP expression was very strong in some hepatocytes yet absent in neighboring hepatocytes (Fig. 5A). Erythrocyte autofluorescence was seen in all liver samples (Fig. 5A–5C) and is readily distinguished from GFP-positive hepatocytes by the characteristic shape and intravascular location of erythrocytes. In the small intestine, GFP was highly, but not uniformly, expressed in differentiated villus enterocytes (Fig. 5D). Intestinal smooth muscle and Paneth cells in the crypts appeared to express GFP as well, but this interpretation was limited by a high level of background staining in smooth muscle and Paneth cells in the negative controls (Fig. 5E, 5F). In the kidney, GFP was highly expressed in glomeruli (Fig. 5G). Renal tubular cells (Fig. 5G) exhibited a variegated pattern of GFP expression, reminiscent of that seen in hepatocytes. Erythrocyte autofluorescence was also noted in the kidney negative controls (Fig. 5H, 5I).
Comparison of the pattern of GFP expression in the three transgenic lines using immunofluorescence revealed additional differences (Fig. 6). In the 1Osb liver, GFP expression was very strong in some hepatocytes but weak or absent from their neighbors (Fig. 6A). In contrast, Y01 liver had strong GFP expression in bile ducts and Kupffer cells but not hepatocytes (Fig. 6B). The UBC liver expressed GFP so weakly (Fig. 6C) that it was essentially indistinguishable from nontransgenic controls (Fig. 6D).
In the small intestine of 1Osb mice, GFP was expressed strongly in mature enterocytes at the villus tip and in Paneth cells at the base of the crypt but less strongly in lamina propria cells and smooth muscle (Fig. 6E). In Y01 small intestine, GFP expression was strong in villus enterocytes, lamina propria, Paneth cells, and smooth muscle (Fig. 6F). In UBC intestine, GFP was strong in mature villus enterocytes and smooth muscle but not Paneth cells or lamina propria (Fig. 6G). The nontransgenic intestine was effectively GFP-negative, although the small area of apparent red staining illustrates the potential for false-positive detection when a portion of the tissue section lifts or curls from the surface of the slide (Fig. 6H).
In the kidney, a subpopulation of renal tubular cells in both 1Osb mice and Y01 mice expressed GFP strongly (Fig. 6I, 6J). In contrast, GFP was abundant in Y01 glomeruli but not 1Osb glomeruli. UBC kidney (Fig. 6K) was essentially indistinguishable from nontransgenic control (Fig. 6L).
In addition to the interstrain comparison described above, we also examined the variability of GFP expression in organs among mice of the same transgenic strain. We found little variation among individual Y01 or UBC-GFP mice (data not shown) but marked variation among mice of the 1Osb strain, confirming an earlier report . Three representative 1Osb livers are shown in supplemental online Figure 2. The proportion of hepatocytes expressing GFP varied from approximately 80% to less than 50%. Similar variability was noted in the small intestine and kidney (not shown).
GFP has understandably become a popular marker for biological lineage tracing and transplantation studies. To better interpret the results of lineage tracing experiments using GFP, we have undertaken a detailed analysis of widely available antibodies and GFP mice. In 1Osb or Y01 mice, GFP was readily detected in formalin-fixed tissue sections from liver, intestine, or kidney. We deliberately restricted our analysis to formalin-fixed paraffin sections, because they are markedly superior to frozen sections in preservation of tissue morphology. We found that immunofluorescent detection of GFP was more sensitive than DAB immunohistochemistry or direct fluorescence. In UBC-GFP mice, GFP was seen in small intestine but was so weak in liver and kidney that it could not be reliably detected by any method we tested. It remains possible that GFP could be detected in frozen tissue sections from UBC-GFP mice.
We found that the goat polyclonal antibody (Abcam) and the rabbit polyclonal antibody (Molecular Probes) performed well in liver and kidney. However, in the small intestine, all three primary anti-GFP antibodies tested caused nonspecific staining of villus enterocytes when detected with DAB, underscoring the critical importance of appropriate negative controls when using this method. Immunofluorescence reduced, but did not eliminate, this problem in the intestine.
Among mice that strongly express GFP in epithelial cells, such as 1Osb mice, the variegation of expression in those cells can limit the sensitivity of detection of transplanted cells by a factor equivalent to the proportion of donor cells that do not express GFP. For epithelial cells of the liver, intestine, and kidney, this proportion may exceed 50% (Figs. 4, Figure 5.–6). Enrichment of the donor population by fluorescence-activated cell sorting selection for GFP-expressing cells prior to transplantation might reduce this problem.
The transgenic construct is identical between the Y01 mouse and the 1Osb mouse, so the site of transgene integration likely accounts for at least part of the difference in expression between these lines. A similar variegated pattern was seen in multiple organs of CMV-GFP mice . The mechanism of silencing of GFP in tissues of the 1Osb mouse is unknown. In hematopoietic stem/progenitor cells of 1Osb mice, GFP expression is gradually lost with differentiation . However, the murine stem cell vector retrovirus conferred long-term, stable expression of GFP, but not DsRed, in transfected mouse hematopoietic stem cells , indicating that hematopoietic differentiation alone does not account for GFP silencing. Silencing of GFP in lentiviral-transfected rat hepatoblasts has been reported when the GFP was driven from the CMV promoter/enhancer but not from the albumin promoter . Therefore, one may speculate that cytosine methylation of CMV regulatory elements plays a role in GFP silencing in the 1Osb mouse.
The phenomenon of in vivo somatic deletion of the toxic albumin-uroplasminogen activator transgene from mouse hepatocytes, followed by expansion and repopulation of the liver by revertant clones, provides a dramatic example of the proliferative and regenerative potential of hepatocytes . However, if GFP toxicity leads to somatic deletion of the transgene, then one would expect to find large clusters of GFP-negative hepatocytes, rather than the apparent stochastic arrangement we observed. Others have proposed that the practical problem of GFP silencing can be overcome by increasing detection sensitivity using a tyramide amplification protocol in frozen tissue sections . Whether this approach will work with paraffin sections has not been reported.
We did not specifically examine the effect of cellular injury on GFP expression in these mice, but we and others  have learned that many types of liver injury dramatically increase autofluorescence, potentially leading to false-positive GFP expression when assessed by direct fluorescence. This problem may account for the unusually high degree of apparent hepatocyte engraftment from transplanted GFP donor bone marrow after CCl4 injury  and provides further impetus for using immunofluorescence and/or confocal microscopy with variable excitation and emission filters to distinguish GFP from autofluorescence.
The reported ability of bone marrow-derived stromal and hematopoietic stem cells to differentiate across traditional developmental boundaries has generated a great deal of interest and controversy [14, –16]. Widely different results of bone marrow “plasticity” experiments are likely due not only to differences in experimental design but also to our limited understanding of the molecular events underlying the required alterations in the cell's gene expression pattern, described as nuclear reprogramming. Nuclear reprogramming might be induced by local cues in the microenvironment or could occur through cell-cell fusion, as has been shown in the fumarylacetoacetate hydrolase knockout mouse after transplantation of wild-type bone marrow [17, 18]. It is not clear that GFP expressed in bone marrow stem cells or mature blood cells would necessarily continue to be expressed in epithelial or other marrow-derived cells after nuclear reprogramming, whether or not the reprogramming occurs by cell fusion. In the case of the Y01 GFP mouse strain, high expression of GFP in blood cells was not matched by high expression in hepatocytes. Thus, it would not be surprising that GFP or any other transgenic reporter would underestimate the frequency of nuclear reprogramming events. Genotypic markers such as the Y chromosome are not subject to such vagaries of transgene expression, indicating that genotypic markers rather than reporters, such as GFP, may be the preferred markers of donor cell origin in cell transplantation experiments.
In this report, we have investigated the assumption that “ubiquitous” GFP mice express GFP at high levels in every cell and pointed out that expression of GFP in blood cells does not predict expression of GFP in tissues, even when a “housekeeping” promoter such as β-actin is used. GFP remains a valuable reporter for many biological applications, but a thorough analysis of its expression in control tissues should be undertaken prior to executing transplantation or lineage-tracking experiments. The optimal method of GFP detection depends on the tissue of interest and must be accompanied by appropriate controls.
Disclosure of Potential Conflicts of Interest
The authors indicate no potential conflicts of interest.
This work was supported by NIH Grants DK61846, HL073742, and DK073404 and by the Yale Center of Excellence in Molecular Hematology.