Adherent HEK293 (from human embryonal kidney), HeLa (from human cervix carcinoma), TP366 and T98G (both from human glioblastoma) cells were grown in a 5% CO2 humidified atmosphere at 37°C. They were passaged in phenol red free DMEM (Dulbecco's modified Eagles) and supplemented with 10% fetal calf serum (Invitrogen Life Technologies, Carlsbad, CA, USA) and 1% glutamine (Biochrom, Germany). For in vivo imaging and measurements, cells were cultured sub-confluent in Falcon flasks and then transferred to 32 mm cover slips in 6-well-plates. After 48 hours, the cover slips with the adherent cells were washed in Hanks' Balanced Salts (PAN-Biotech, Aidenbach, Germany) and the full culture medium was replaced by phenol red free DMEM containing only 2% of fetal calf serum. Then cells were double-transfected with the mammalian expression vectors using FuGene HD (Roche Diagnostics, Mannheim, Germany) in different proportions, and with different DNA concentrations, depending on cell-line and construct, following the manufacturer's recommendations. We chose to work with transiently double transfected cells instead of stable cell lines in order not to depend too much on the specific genotype of one transfected cell.
For FFM-measurements, the cover slips were mounted on a measurement chamber developed in our laboratory 
allowing a working volume of 3 ml, 14 to 24 hours after transfection. The measurement chamber was placed on the stage of the FFM in an incubator compartment at 37°C, 5% CO2
Restriction enzymes were purchased from MBI Fermentas (Vilnius, Lithuania).
PCR amplifications were performed using PCR Master Mix (Promega, Madison, WI, USA). PCR purification and gel extraction kits were purchased from Macherey-Nagel (Düren, Germany). Plasmids were cloned in Escherichia coli XL10 (Stratagene, Amsterdam, Netherlands) and isolated using Maxi Prep kits (Macherey-Nagel, Düren, Germany), employed as proposed by the manufacturer.
The commercially available p-eGFP-N1 plasmids (Clontech, Saint-Germain-en-Laye, France) encode eGFP-monomers. The plasmids encoding the eGFP dimer, trimer and tetramer were a generous gift from Dr. M.M. Nalaskowski (Universitätsklinikum Hamburg-Eppendorf) and created following the method described in 
The human histone H2A gene was amplified by genomic PCR and inserted N-terminal of the enhanced cyan fluorescent protein (eCFP) into the promoter less plasmid peCFP-1 (Clontech). Upstream we inserted the HindIIIC fragment of simian virus 40 (SV40) in reverse direction, such that the fusion protein of 372 amino acid residues was expressed through the early SV40 promoter. In a second step, eCFP was replaced by mRFP1.
Cell lysis, gel electrophoresis and Western-blot
Dishes containing a confluent layer of transfected HeLa cells after 14 to 24 hours were washed 3 to 4 times with 100 nM STE-Buffer and the cells collected. The concentrated cell suspension was treated with a Dounce homogenizer, the resulting lysates were cleared of the largest debris by centrifugation and further purified via Vivaspin PES centrifugal filters and concentrators (Sartorius Stedim Biotech, Aubagne, France) with appropriate membrane size.
The soluble protein extracts were treated with non-reducing 5× native gel loading dye without boiling and assayed on a 10% native polyacrylamide gel with 0.1% SDS and a 1× Tris-Glycine-running-buffer pH 8.8. The in-gel fluorescence from the non-reduced eGFP samples was detected using a Typhoon 9410 Variable Mode Imager (Amersham Biosciences, Piscataway, NJ) using the 488-nm laser line for excitation and collecting the fluorescence at a wavelength of 520 nm.
For the Western blot, the same soluble cell extracts as for the native fluorescence gels were boiled with 3× Laemmli sample buffer for 5 minutes. Protein samples (15 µl sample eGFP-monomer, 11.25 µl eGFP-dimer, 11.25 µl eGFP-trimer and 7.5 µl eGFP-tetramer) together with molecular weight marker (15 µl; Broad Range, New England Biolabs) were subjected to sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) and electroblotted. The membrane was incubated with rabbit polyclonal Anti-GFP antibody (GTX26556, GeneTex, Inc.), diluted 1
2000, and - as loading control - the same membrane was incubated with mouse monoclonal Anti-β-Actin antibody (A 5441, Sigma), diluted 1
10000. As secondary antibodies peroxidase-conjugated AffiniPure Goat Anti-Rabbit and Goat Anti-Mouse, respectively, were used diluted 1
5000 (Jackson ImmunoResearch). All immune reactions were carried out in 10 mM Tris–HCl, pH 8.0, 150 mM NaCl, 0.05% Tween-20 (TBST) with 5% dried milk at RT with washing steps in TBST.
Fluorescence Fluctuation Microscopy (FFM)
We employed a laboratory-built setup, the Fluorescence Fluctuation Microscope (FFM) 
, which is a combination of Fluorescence Correlation Spectroscopy (FCS) and Confocal Laser Scanning Microscopy (CLSM). FFM combines an FCS module and a galvanometer mirror scanning unit attached to the side video port of an inverted IX-70 microscope (Olympus, Hamburg, Germany) with an UplanApo / IR 60× water immersion objective lens with a numerical aperture (NA) of 1.2 
Intracellular measurements were all performed at 37°C, in a 5% CO2 humidified atmosphere, in a incubator chamber (EMBL, Heidelberg, Germany) surrounding the whole microscope.
For fluorescence excitation, we used an argon-krypton laser from CVI Melles Griot (Bensheim, Germany), with the 488 nm line for eGFP and the 568 nm line for mRFP1. The emission from eGFP was detected from 515 to 545 nm and between 608 and 662 nm for mRFP1 with two avalanche photodiodes (APD) (SPCM-AQR-13, Perkin-Elmer, Wellesley, USA), after passing appropriate dichroic mirrors and filters for spectral separation and selection. FCS measurements were carried out at laser intensities from 5 to 9 kW·cm−2 for both laser lines. Laser-power was adjusted with the help of a polychromatic acousto optical modulator AOTF Nc (AA Opto Electronic, France), which allows precise control over the laser-power.
Using a home-made control software, we acquired confocal fluorescence images and randomly chose the FCS measurement points with either high or low chromatin density, but avoiding the nucleoli.
The signals coming from the APDs were fed into an ALV-5000/E correlator card (ALV Laser GmbH, Langen, Germany), where intensity fluctuations were recorded and their autocorrelation function simultaneously and almost in real-time, calculated.
Control temperature measurements were performed on a “The Cube & The Box” incubator (Life Imaging Services) using a USB TC-08 Thermocouple Data Logger (Pico Technology) or on a GP-168 incubator (EMBL, Heidelberg, Germany) coupled with a THERM 2290-2/3 multifunctional measuring instruments (Ahlborn Mess- und Regelungstechnik GmbH, Holzkirchen, Germany)
Preparation of fluorophore solutions
Fluorophore solutions were diluted in de-ionized water and deposited into a homemade measurement chamber 
before acquisitions using FFM microscope. Compared to disposable observation chambers, our chamber uses high quality coverslips permitting a better adjustment of the objective correction collar. The low variance in the coverslips width allows obtaining a similar optical path for both reference and biological samples. Two widely used fluorophores were tested: alexa488 and rhodamine B.
In vitro measurements of eGFP-multimers
The same soluble protein extracts as for the native polyacrylamide gel were used for in vitro measurements of the eGFP-multimers. Measurements were carried out at 37°C, in a humidified 5% CO2 atmosphere. The same measurement chamber as for in vivo experiments was used, with a working volume of about 2 ml.
Intranuclear data acquisition and sorting
For data acquisition, a confocal image of the mid-section of the whole cell was recorded using CLSM. On this image, up to 5 random positions were selected in the nucleus of the cell, avoiding the nucleoli. We then performed autocorrelation measurements of six runs of ten seconds each for every single measurement point. A second image was recorded after the completion of the FCS measurements in order to exclude measurements in which the cell moved during the experiment. The laser power was about 5 kW·cm−2 in order to minimize photobleaching as well as cell damages, stress and thus movements. The crosstalk from the red channel into the green and vice-versa was negligible. We studied at least 44 representative points for each construct in each cell line.
Before data analysis, a first selection was done on the basis of the recorded image: each cell that showed a motion between the first and the second picture was discarded. The second selection criterion was the study of the recorded data set for the histone fluorescence intensity channel: the recorded intensities should decrease regularly due to photobleaching effect. Data showing an abnormal behavior of the fluorescence intensity (e.g. a rapid decrease followed by a slow one, or even an increase of the signal intensity) were discarded as well, these being indicators for conformational changes in the cell nucleus. This procedure ensures that the recorded and analyzed fluorescence fluctuations are due to particle diffusion and not to cell movement or reorganization.
The setup was calibrated using Alexa 488 (Molecular Probes, Eugene, Oregon) for the green channel and Alexa 568 for the red one. The focal volume was determined for every work session using a 20 nM Alexa 488 solution, diluted in sterile de-ionized water.
For collecting the diffusions maps we used only one laser line (488 nm) to excite both eGFP and mRFP1. This was done in order to reduce the cell's exposure to laser energy and therefore allowed for a maximal number of measurement points in the cells. At 488 nm, mRFP1 still shows over 35% excitation efficiency compared to its maximum absorption at 584 nm 
, which does not constitute a major drawback to the measurement other than a globally reduced intensity of the mRFP1 fluorescence.
The red channel data gives the fluorescence intensity from the histones, and therefore the chromatin density. We normalized the data using the average value of the red channel for the whole nucleus, allowing comparison of cells that showed different overall expression of the tagged histone. For this normalization, we assumed that every cell from a given cell line contains the same amount of DNA and that it is equally distributed over different cross-sections in a cell. The autocorrelation curves of each individual measured point are then fitted by a normal diffusion model, as described in 
. We chose to fit the data to a normal diffusion model rather to an obstructed diffusion model because it is more robust and allows the fit of two distinct diffusive fluorescent populations 
In the normal diffusion model, the mean square displacement <x2
> depends linearly on the time t
is the diffusion coefficient.
To account for non-fluorescent processes and diffusing components of different sizes, we fitted the model function (Eq. 2) for two fluorescent diffusive components and one non-fluorescent component by the program Quickfit
, written in our laboratory and based on the Marquardt-Levenberg algorithm 
In equation (2), the nonfluorescent components are due to transitions of the fluorescent molecules into the triplet state, where
is the fraction of particles in this state, ai
the relative fraction of each diffusive species, N
the average number of particles in the focus volume and ksp
the structure factor, which depends on the focus volume and is given by:
is the axial and w0
the lateral dimension of the detection volume. The diffusion time ιdiff
is related to the diffusion coefficient D
For all proteins in all cell lines, we obtained the fraction of the fast and the slow components as well as their respective diffusion times. From the latter we calculated diffusion constants using the measured diffusion times of Alexa 488 at 25°C and 37°C and its published diffusion constant 
. The structure factor ksp
of the focus volume was determined using Alexa 488 in aqueous solution and was found to be consistent with the point spread function obtained with embedded fluorescent beads (200 nm diameter). In living cells, however, there is a refractive index mismatch that distorts the confocal volume and which cannot be corrected in a straightforward way. The effect of refractive index mismatch on FCS measurements was recently estimated by Müller et al. 
. With the refractive index of the cellular environment of 1.36±0.004 
, this mismatch would lead to an absolute error in diffusion coefficient of less than 10% at the focal depth used here (<20 micrometers). This is much smaller than the variations in diffusion time actually measured in our experiments, so it would be very unlikely that small variations in refractive index mismatch would cause the changes observed here. However, absolute values of the diffusion coefficients calculated from the measured diffusion times may be slightly underestimated. Since we compare diffusion coefficients which have all been measured inside living human cells under the same conditions, such deviations are not critical.
Diffusion maps were generated by interpolation from the FCS measurement points using the program Origin (OriginLab Corporation, Northampton, MA, USA).
Diffusion simulations of eGFP-constructs
For computing the theoretical diffusion coefficients of the eGFP-constructs, assuming a rod shape, we used the latest available version of the public domain software SEDNTERP 
allows calculating the viscosity η and the density ρ for the used buffer as well as computing the partial specific volume,
from the amino acid composition of the protein. It also allows building oligomers from monomers of a given composition, thus enabling to compute diffusion coefficients of rod-shape oligomers.
We first determined the diffusion coefficient of monomeric eGFP assuming a globular shape, using equation (5) and (6) with V
as the volume of the protein, M
the molecular weight,
the partial specific volume (from SEDNTERP), Na
the Avogadro number and r
the radius of the sphere, for calculating the radius of the sphere.
The diffusion coefficient of a spherical particle is given by the Stokes-Einstein equation:
is the diffusion coefficient, kB
the Boltzmann constant, T
the absolute temperature, r
the radius of the sphere and η
the viscosity of the solvent. The value for the radius, calculated from equation (6) was used in equation (7), with η
equal to the viscosity of water, to get the theoretical diffusion coefficient of the globular shaped protein. The diffusion coefficient was then corrected for 37°C using
and estimated for oligomers of molecular masses Mα
(assumed spherical) with
These correspond to the maximum possible diffusion coefficients for perfect spheres.