Quantitative comparison of DNA detection by GFP-lac repressor tagging, fluorescence in situ hybridization and immunostaining
© Kim et al; licensee BioMed Central Ltd. 2007
Received: 02 August 2007
Accepted: 20 December 2007
Published: 20 December 2007
GFP-fusion proteins and immunostaining are methods broadly applied to investigate the three-dimensional organization of cells and cell nuclei, the latter often studied in addition by fluorescence in situ hybridization (FISH). Direct comparisons of these detection methods are scarce, however.
We provide a quantitative comparison of all three approaches. We make use of a cell line that contains a transgene array of lac operator repeats which are detected by GFP-lac repressor fusion proteins. Thus we can detect the same structure in individual cells by GFP fluorescence, by antibodies against GFP and by FISH with a probe against the transgene array. Anti-GFP antibody detection was repeated after FISH. Our results show that while all four signals obtained from a transgene array generally showed qualitative and quantitative similarity, they also differed in details.
Each of the tested methods revealed particular strengths and weaknesses, which should be considered when interpreting respective experimental results. Despite the required denaturation step, FISH signals in structurally preserved cells show a surprising similarity to signals generated before denaturation.
The consistency of fluorescence detection signals with the in vivo distribution of the detected structure is an important technical issue in modern cell biology. Quality of generated signals may be influenced by applied fixation methods [1–3] as well as the approach used for detection. For the detection of specific DNA sequences in the cell nucleus, two methods are available : fluorescence in situ hybridization (FISH) can be applied to any sequence large enough to generate sufficient hybridization sites for DNA-probes, but only to fixed cells. In vivo labeling is possible with GFP fusions to DNA binding proteins such as the lac repressor which then binds to lac operator sequences within transgenes . By design this approach does not label endogenous eukaryotic sequences, except for some tandem repetitive sequences such as centromeres and telomeres . In investigations of the cellular localization of proteins, fusions to fluorescent proteins or immunostaining are commonly applied methods. Despite the widespread usage of these methods, however, simultaneous or sequential application to the same cells are scarce  and we are not aware of a detailed comparisons of these detection methods.
Here we provide a qualitative and quantitative comparison of signals from multi-labeling experiments where we applied the labeling methods mentioned above to the same structure. We used a mouse erythroleukemia (MEL) cell line that contained a large array with lac operator transgenes , which was in vivo labeled with GFP lac repressor fusion proteins expressed by the cells. Cells were fixed with buffered formaldehyde to maintain structural integrity [3, 8], permeabilized, immunostained and three-dimensional image stacks of GFP and immunostaining signals were recorded by fluorescence microscopy. Afterwards, cells were unmounted, subjected to FISH, relocated under the microscope and FISH-signals and again immunostaining-signals were recorded. Signals from all four recorded image stacks were then compared qualitatively and by quantitative digital image analysis. In additional experiments, we tested the similarity of two FISH signals generated by different probes against the same DNA sequence and for differences of GFP and FISH signals when no RNAse digestion was performed.
Verification of the image analysis approach with dual color FISH
Since a labeled substructure in one channel which is not present at exactly the same site in another channel will lead to a decrease of the CC value, we assumed this value to be very sensitive to structural differences of compared signals. We tested this assumption by computationally shifting one of the color channels 0.085 μm in x-direction, 0.065 μm in y-direction and 0.011 μm in z-direction, with subsequent tri-linear interpolation to obtain an image stack with voxels at the same position as the original stacks. Indeed, the median dropped from 0.81 to 0.71 for deconvolved image stacks while the more blurry, non-deconvolved image stacks showed only a drop from 0.94 to 0.91.
Comparison of multiple detection signals: in vivo GFP-lac repressor staining, antibody detection and FISH on the same transgene array
After recording of GFP and antibody signals, a postfixaton, subsequent FISH and again immunostaining was performed (see methods). Visual inspection revealed highly similar appearance of FISH and immunostaining signals, as well as similarity to signals recorded before FISH (Figure 2). Sometimes the FISH signal showed more substructure and higher contrast than other signals (Figure 2, top row). Also, we had the impression that both signals obtained after FISH occupied a larger volume than those obtained before FISH.
Therefore, we first measured volumes after segmenting signals by thresholding. The volume obtained for an individual signal by this approach may vary largely, depending on the subjectively chosen threshold. However, when signals are segmented by steady criteria and volume ratios of series of signals are compared to each other, the obtained ratios are reasonably stable. Because of the difficulties described we limited volume measurements to deconvolved signals with their improved signal to noise ratio. Volumes of GFP signals and simultaneously detected immunostaining were not distinguishable (mean 0.4 μm3 for both, 427 and 460 voxels, respectively). While FISH signals (0.5 μm3, 541 voxels) were only 27% larger than GFP signals and 18% larger than pre-FISH antibody signals, antibody signals after FISH (0.7 μm3, 794 voxels) showed a 73% volume increase compared to pre-FISH antibody signals. Accordingly, in 53 of 55 nuclei segmented antibody signals were larger after FISH.
Correlation coefficient values
Contribution of DNA-RNA hybridization to FISH signals
In this study, we quantitatively compared three detection methods widely used in studies of nuclear organization and beyond. Reassuringly, GFP, antibody detection and FISH produced signals similar in overall structure. However, we also could observe differences in details. Anti-GFP antibody signals at sites without GFP fluorescence argue for non-fluorescent GFP molecules, masking a part of the underlying structure. We have not formally proven that such an incomplete detection of GFP fusion proteins by GFP fluorescence also occurs in vivo, before fixation, but this appears as the most likely conclusion. Whether non-fluorescent GFP plays a role in fusion proteins with other partners is unclear at present. On the other hand, antibody detection can be hampered by incomplete penetration. In normal immunofluorescence assays it is difficult to estimate the magnitude of this problem. By using a fluorescent protein as target we could show that the preparation procedure applied here allows for nearly complete detection of fluorescent GFP in this cell type.
FISH signals showed a high similarity with the post-FISH antibody signals. We did not find that FISH signals would have less substructure than GFP signals, as it was described in a previous study , maybe as a consequence of harsher denaturation. On the contrary, some FISH signals had more substructure and higher contrast than the other three signals. It is possible that FISH is more sensitive, due to a larger number of fluorochromes per volume and a resulting higher signal to noise ratio, or due to an incomplete binding of the GFP-lac repressor to its target sites. Alternatively, this additional substructure may be caused by moving of the DNA during the denaturation step and the accompanying destruction of the ultrastructure which has been shown by electron microscopy . Post-FISH antibody signals were on average 73% larger than pre-FISH antibody signals, suggesting a certain spreading of the DNA together with DNA bound proteins, during FISH. Surprisingly, FISH signals themselves showed volumes only 27 or 18% larger than GFP- or pre-FISH antibody signals, maybe due to differences in detection efficiency. Changes during FISH, more specifically during the required denaturation step, were described in earlier studies [3, 8] and are reflected by the blurrier appearance of DAPI stained nuclei after FISH.
Immunostaining signals showed a high similarity to each other as well as to the simultaneously recorded GFP or FISH signal. Compared to other comparisons, however, the correlation between the two antibody signals from the same nucleus may have been positively influenced by the fact that it was these two signals that were used to align pre- and post-FISH image stacks. Using these signals for alignment of pre and post-FISH stacks seems to be the most reasonable approach, however, since here the same molecules are detected. Despite this favorable situation, in half of the deconvolved nuclei a CC value of only 0.76 or less was reached, further supporting the notion of changes in the underlying structure during denaturation.
In a previous study, we could show that the appearance of chromosomes or chromosomal regions detected by FISH varies substantially, depending on the fixation protocol applied : While formaldehyde fixed nuclei had relatively compact FISH signals, nuclei subjected to hypotonic swelling, methanol acetic acid fixation, dropping on slides and flattening by air drying (2D-FISH) displayed FISH signals with a dispersed structure, suggesting structural disruption . Our current study confirms that the in vivo organization of chromatin is well represented by FISH signals in formaldehyde fixed cells and thus strengthens the conclusion that this is not the case for the more spread-out signals in 2D-FISH preparations. Our current results show that the spatial distribution of GFP, immunostaining and FISH signals from the same structure in structurally preserved cell nuclei are largely overlapping at the light microscopy level, although they are not identical.
Each of the detection methods tested in this study carries its specific set of advantages and disadvantages. The signal-to-noise ratio of GFP signals was much lower than immunostaining or FISH signals. Also, GFP fusion proteins are apparently not always fluorescent, and this non-fluorescent fraction is not distributed equally. Therefore, GFP in vivo staining should not be regarded as being generally superior compared to other detection methods. In addition, despite the undisputed usefulness of GFP fusion proteins, their expression was linked to changes in the physiological state of living cells such as induction of apoptosis , dilated cardiomyopathy in transgenic mice , impairment of actin-myosin interactions [11, 12], inhibition of polyubiquitination , and cytokine induction . Interference with the physiological state of the cell can be excluded for staining techniques applied after fixation. However, antibody detection could be limited by the permeability of the sample, although the permeabilization we applied in the current study allowed comprehensive detection of fluorescent GFP. FISH, by design, requires denaturation of the DNA and thus a partial structural destruction of the sample. Giving this unavoidable disadvantage, we were actually surprised how similar post FISH signals still are on the light microscopic level in structurally preserved cell nuclei when compared to detection prior to denaturation.
Cells and dual color FISH
PALZ39E is a mouse erythroleukemia (MEL) cell line stably transgenic for a GFP lac repressor fusion protein and with multiple integrations of the 15 kbp plasmid pPALZ8.8 containing 64 repeats (2.5 kbp) of the lac operator binding site (lacO), β-globin regulatory sequences and a β-galactosidase reporter gene . The plasmid pPS8.8  also contains 64 copies of the lac operator but no β-globin or β-galactosidase related sequences. pPALZ8.8 and pPS8.8 were labeled by nick translation with Digoxigenin-dUTP or Biotin-dUTP, respectively. Fixation with 3.7% freshly made buffered formaldehyde (10 min), permeabilization treatment and FISH conditions were as described  (15 min 0.5% Triton, five freeze/thaw cycles in liquid nitrogen, 10 min 0.1 N HCl, no protease treatment, denaturation in 50% formamide/SSC at 75°C for 2 min). Detection was with Sheep-α-Dig-FITC (1:100, Roche Diagnostics, Mannheim, Germany) and Streptavidin-Cy5 (1:200, Rockland, Gilbertsville, PA).
Antibody detection of GFP
Cells were cultivated and fixed as above except that grided coverslips (Belco Biotechnology, Etched GRID coverslip 23 × 23 mm, stock No. 1916-92525, distributed by Electron Microscopy Sciences, Ft. Washington, PA, (#72264-23) and obtained through Science Services, Munich, Germany (#141204)) were used to allow relocation of cells. For immunostaining, cells were permeabilized for 10 minutes with 0.5 % Triton X-100 in PBS. Blocking was for 60 minutes or longer with 4% bovine serum albumine in PBS, incubation with primary antibody (RabbitαGFP, 1:500, Invitrogen, R970-01) was for 45 minutes or longer. After 3 washes (ea. 10 min) in PBS, a secondary, biotinylated GoatαRabbit (1:100 Biosource, Camarillo, CA) was applied which was finally detected with Cy5-conjugated streptavidin. Usage of a biotinylated antibody ensured that the signal would be detectable also later, after the denaturation required for FISH. Cy5 is spectrally sufficiently separated from GFP (or FITC) to exclude bleed-through from one fluorescence channel to the other. Nuclei were counterstained with DAPI, mounted with Vectashield (Vector, Burlingame, CA, USA) and fixed with a nail polish that did not disturb GFP fluorescence on microscopic slides for microscopic observation.
Microcopy was performed on a VisiScope Cell Explorer (Visitron Systems, Puchheim, Germany) based on a Zeiss Axiovert 200 mot microscope and a Spot RT-SE6 CCD Camera with Sony ICX285 chip, controlled by Metamorph Software. 3D-stacks were recorded with a 100 × N.A. 1.4 Zeiss PlanApo oil objective with a voxel size of 0.065 × 0.065 × 0.2 μm except for the multi-signal comparison without RNAse where voxels of 0.103 × 0.103 × 0.25 were obtained with a 63 × NA 1.4 PlanApo objective. The following filter sets were used: DAPI (360/40, 400LP, 470/40), GFP (470/40, 497LP, 522/40), and Cy5 (622/36, 647LP, 667/30). Deconvolution was performed with Huygens essential software (SVI, Hilversum, The Netherlands).
FISH after recording GFP and immunostaining signals
After the first round of 3D-microscopy, coverslips were unmounted from microscopic slides, washed several times in PBS and subjected to a post-fixation in 1% buffered formaldehyde for 10 minutes to fix the antibodies to their locations. To exclude hybridization of probe to RNA, digestion with DNAse free RNAse A (0,2 mg/ml in PBS for 24 hours at 37°C; Quiagen, Hilden, Germany) was performed except otherwise noted. Control experiments with Acridine Orange staining showed that cytoplasmic and nucleolar RNA was completely removed by this procedure (data not shown). Permeabilization treatment and FISH conditions were as described  (see also above). The plasmid pPS8.8  was used as probe, labeled with Digoxigenin-dUTP, except otherwise noted. Control experiments showed that fluorescence from GFP was completely lost after FISH (data not shown). All three steps of the antibody detection of GFP (see above) were repeated. In parallel, Digoxigenin detection was performed with Sheepα Dig-FITC (1:100, Roche Diagnostics). Microscopy and deconvolution were as described above. Comparison of the DNA counterstain appearances before and after FISH ensured correct identification of nuclei.
Volumes of fluorescent signals were determined with the plug-in Voxel Counter of the freely available open source software ImageJ  after threshold segmentation.
To quantify the structural similarity between GFP-, immunostaining- and FISH-signals, we developed a rigid registration approach. The whole procedure was applied to non-deconvolved as well as deconvolved data. Each 3D multi-channel image stack was corrected for chromatic aberration (measured with polychromatic beads) and cropped to the same image dimensions. For faster computational processing, calculations were performed only within a region of interest, defined by setting a low threshold to the signals of all four stacks and combining the segmented volumes. Since here it was more important not to miss any signal parts rather than to accurately reflect signal borders, as was attempted for volume measurements described above, these thresholds were newly determined for each image. A subsequently applied connected-components labeling algorithm identified the largest region in each GFP- or Cy5-channel, which was invariably the signal of interest, as confirmed by visual inspection.
where f and g denote the two signals and m f and m g the corresponding mean values.
We thank Andrew S. Belmont for helpful discussions, Max Schneider and Claudia Hepperger for help with lab work, Thomas Cremer for continuous support of our work and Christian Lanctôt for critical reading of the manuscript. This work was supported by the Deutsche Forschungsgemeinschaft.
- Belmont AS, Braunfeld MB, Sedat JW, Agard DA: Large-scale chromatin structural domains within mitotic and interphase chromosomes in vivo and in vitro. Chromosoma. 1989, 98 (2): 129-143. 10.1007/BF00291049.View ArticleGoogle Scholar
- Mongelard F, Vourc'h C, Robert-Nicoud M, Usson Y: Quantitative assessment of the alteration of chromatin during the course of FISH procedures. Fluorescent in situ hybridization. Cytometry. 1999, 36 (2): 96-101. 10.1002/(SICI)1097-0320(19990601)36:2<96::AID-CYTO2>3.0.CO;2-X.View ArticleGoogle Scholar
- Hepperger C, Otten S, von Hase J, Dietzel S: Preservation of large-scale chromatin structure in FISH experiments. Chromosoma. 2007, 116 (2): 117-133. 10.1007/s00412-006-0084-2.View ArticleGoogle Scholar
- Dirks RW, Tanke HJ: Advances in fluorescent tracking of nucleic acids in living cells. Biotechniques. 2006, 40 (4): 489-496.View ArticleGoogle Scholar
- Robinett CC, Straight A, Li G, Willhelm C, Sudlow G, Murray A, Belmont AS: In vivo localization of DNA sequences and visualization of large-scale chromatin organization using lac operator/repressor recognition. J Cell Biol. 1996, 135 (6): 1685-1700. 10.1083/jcb.135.6.1685.View ArticleGoogle Scholar
- Müller WG, Walker D, Hager GL, McNally JG: Large-scale chromatin decondensation and recondensation regulated by transcription from a natural promoter. J Cell Biol. 2001, 154 (1): 33-48. 10.1083/jcb.200011069.View ArticleGoogle Scholar
- Dietzel S, Zolghadr K, Hepperger C, Belmont AS: Differential large-scale chromatin compaction and intranuclear positioning of transcribed versus non-transcribed transgene arrays containing beta-globin regulatory sequences. J Cell Sci. 2004, 117 (Pt 19): 4603-4614. 10.1242/jcs.01330.View ArticleGoogle Scholar
- Solovei I, Cavallo A, Schermelleh L, Jaunin F, Scasselati C, Cmarko D, Cremer C, Fakan S, Cremer T: Spatial Preservation of Nuclear Chromatin Architecture during Three- Dimensional Fluorescence in Situ Hybridization (3D-FISH). Exp Cell Res. 2002, 276 (1): 10-23. 10.1006/excr.2002.5513.View ArticleGoogle Scholar
- Liu HS, Jan MS, Chou CK, Chen PH, Ke NJ: Is green fluorescent protein toxic to the living cells?. Biochem Biophys Res Commun. 1999, 260 (3): 712-717. 10.1006/bbrc.1999.0954.View ArticleGoogle Scholar
- Huang WY, Aramburu J, Douglas PS, Izumo S: Transgenic expression of green fluorescence protein can cause dilated cardiomyopathy. Nat Med. 2000, 6 (5): 482-483. 10.1038/74914.View ArticleGoogle Scholar
- Agbulut O, Coirault C, Niederländer N, Huet A, Vicart P, Hagège A, Puceat M, Menasché P: GFP expression in muscle cells impairs actin-myosin interactions: implications for cell therapy. Nat Methods. 2006, 3 (5): 331-10.1038/nmeth0506-331.View ArticleGoogle Scholar
- Agbulut O, Huet A, Niederlander N, Puceat M, Menasche P, Coirault C: Green fluorescent protein impairs actin-myosin interactions by binding to the actin-binding site of myosin. J Biol Chem. 2007, 282 (14): 10465-10471. 10.1074/jbc.M610418200.View ArticleGoogle Scholar
- Baens M, Noels H, Broeckx V, Hagens S, Fevery S, Billiau AD, Vankelecom H, Marynen P: The Dark Side of EGFP: Defective Polyubiquitination. PLoS ONE. 2006, 1: e54-10.1371/journal.pone.0000054.View ArticleGoogle Scholar
- Mak GW, Wong CH, Tsui SK: Green fluorescent protein induces the secretion of inflammatory cytokine interleukin-6 in muscle cells. Anal Biochem. 2007, 362 (2): 296-298. 10.1016/j.ab.2006.12.017.View ArticleGoogle Scholar
- ImageJ. [http://rsb.info.nih.gov/ij/]
This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.