Supplementary Material:Dual-Color Fluorescence Cross-Correlation Spectroscopy on a Single Plane

Illumination Microscope (SPIM-FCCS)

Jan Wolfgang Krieger,1,† Anand Pratap Singh,2,†, Christoph S. Garbe,3 ThorstenWohland,2 and Jörg Langowski,1,∗

1German Cancer Research Center (DKFZ), Biophysics of Macromolecules (B040), Im Neuenheimer Feld 580,D-69120 Heidelberg, Germany,

2Departments of Biological Sciences and Chemistry and NUS Centre for Bio-Imaging Sciences, NationalUniversity of Singapore, 14 science Drive 4, Singapore 117557,

3Interdisciplinary Center for Scientific Computing, University of Heidelberg, Speyerer Straße 6, D-69115Heidelberg, Germany,

†Anand Pratap Singh and Jan Krieger contributed equally to this work.∗ jl@dkfz.de

http://www.dkfz.de/Macromol

References and links1. A. P. Singh, J. W. Krieger, J. Buchholz, E. Charbon, J. Langowski, and T. Wohland, “The performance of 2D array detectors for light

sheet based fluorescence correlation spectroscopy,” Opt. Express 21, 8652–8668 (2013).2. The matlab scripts for the bead scan evaluation is freely available at: http://www.dkfz.de/Macromol/quickfit/

beadscan.html.3. N. Bag, J. Sankaran, A. Paul, R. S. Kraut, and T. Wohland, “Calibration and limits of camera-based fluorescence correlation spectroscopy:

A supported lipid bilayer study,” ChemPhysChem 13, 2784–2794 (2012).4. J. Sankaran, N. Bag, R. S. Kraut, and T. Wohland, “Accuracy and precision in camera-based fluorescence correlation spectroscopy

measurements,” Anal. Chem. 85, 3948–3954 (2013).5. N. Baudendistel, G. Müller, W. Waldeck, P. Angel, and J. Langowski, “Two-hybrid fluorescence cross-correlation spectroscopy detects

protein–protein interactions in vivo,” ChemPhysChem 6, 984–990 (2005).6. G. Vámosi, N. Baudendistel, C.-W. von der Lieth, N. Szalóki, G. Mocsár, G. Müller, P. Brázda, W. Waldeck, S. Damjanovich,

J. Langowski, and K. Tóth, “Conformation of the c-Fos/c-Jun complex in vivo: A combined FRET, FCCS, and MD-modeling study,”Biophys. J. 94, 2859–2868 (2008).

7. P. Liu, T. Sudhaharan, R. M. Koh, L. C. Hwang, S. Ahmed, I. N. Maruyama, and T. Wohland, “Investigation of the dimerization ofproteins from the epidermal growth factor receptor family by single wavelength fluorescence cross-correlation spectroscopy,” Biophys.J. 93, 684–698 (2007).

8. C.-H. Huang, “Phosphatidylcholine vesicles. formation and physical characteristics,” Biochemistry 8, 344–352 (1969). PMID: 5777332.9. M. Wachsmuth, “Fluoreszenzfluktuationsmikroskopie: Entwicklung eines prototyps, theorie und messung der beweglichkeit von

biomolekülen im zellkern,” Ph.D. thesis, Ruprecht-Karls-Universität ,Heidelberg. (2001).10. QuickFit 3.0 can be downloaded free of charge from http://www.dkfz.de/Macromol/quickfit/.11. X. Pan, W. Foo, W. Lim, M. H. Fok, P. Liu, H. Yu, I. Maruyama, and T. Wohland, “Multifunctional fluorescence correlation microscope

for intracellular and microfluidic measurements,” Rev. Sci. Instrum. 78, 053711–053718 (2007).12. P. Kapusta, “Absolute diffusion coefficients: compilation of reference data for fcs calibration,” PicoQuant Application note (2010).13. Z. Petrášek and P. Schwille, “Precise measurement of diffusion coefficients using scanning fluorescence correlation spectroscopy,”

Biophys. J. 94, 1437–1448 (2008).14. A. P. Siegel, N. M. Hays, and R. N. Day, “Unraveling transcription factor interactions with heterochromatin protein 1 using fluorescence

lifetime imaging microscopy and fluorescence correlation spectroscopy,” J. Biomed. Opt. 18, 025002–025012 (2013).15. T. Wohland, X. Shi, J. Sankaran, and E. H. K. Stelzer, “Single plane illumination fluorescence correlation spectroscopy (SPIM-FCS)

probes inhomogeneous three-dimensional environments,” Opt. Express 10, 10627–10641 (2010).16. J. Sankaran, X. Shi, L. Ho, E. Stelzer, and T. Wohland, “ImFCS: A software for imaging FCS data analysis and visualization,” Opt.

Express 18, 25468–25481 (2010). Available at .17. L. C. Hwang and T. Wohland, “Single wavelength excitation fluorescence cross-correlation spectroscopy with spectrally similar

fluorophores: Resolution for binding studies,” J. Chem. Phys. 122, 114708 (2005).18. J. Ries, Z. Petrášek, A. J. Garcı́a-Sáez, and P. Schwille, “A comprehensive framework for fluorescence cross-correlation spectroscopy,”

New J. Phys. 12, 113009 (2010).19. J. Ries, S. Chiantia, and P. Schwille, “Accurate determination of membrane dynamics with line-scan FCS,” Biophys. J. 96, 1999–2008

(2009).20. J. Hendrix, C. Flors, P. Dedecker, J. Hofkens, and Y. Engelborghs, “Dark states in monomeric red fluorescent proteins studied by

fluorescence correlation and single molecule spectroscopy,” Biophys. J. 94, 4103–4113 (2008).21. U. Haupts, S. Maiti, P. Schwille, and W. W. Webb, “Dynamics of fluorescence fluctuations in green fluorescent protein observed by

fluorescence correlation spectroscopy,” Proc. Natl. Acad. Sci. 95, 13573–13578 (1998).22. T. Wocjan, J. Krieger, O. Krichevsky, and J. Langowski, “Dynamics of a fluorophore attached to superhelical DNA: Fcs experiments

simulated by brownian dynamics,” Phys. Chem. Chem. Phys. 11, 10671–10681 (2009).23. J. Buchholz, J. W. Krieger, G. Mocsár, B. Kreith, E. Charbon, G. Vámosi, U. Kebschull, and J. Langowski, “Fpga implementation of a

32x32 autocorrelator array for analysis of fast image series,” Opt. Express 20, 17767–17782 (2012).24. D. E. Koppel, “Analysis of macromolecular polydispersity in intensity correlation spectroscopy: The method of cumulants,” J. Chem.

Phys. 57, 4814 (1972).25. T. Wohland, R. Rigler, and H. Vogel, “The standard deviation in fluorescence correlation spectroscopy.” Biophys. J. 80, 2987–2999

(2001).

mailto:jl@dkfz.dehttp://www.dkfz.de/Macromolhttp://www.dkfz.de/Macromol/quickfit/beadscan.htmlhttp://www.dkfz.de/Macromol/quickfit/beadscan.htmlhttp://www.dkfz.de/Macromol/quickfit/http://staff.science.nus.edu.sg/~chmwt/resources/imfcs_software.html

26. S. Saffarian and E. L. Elson, “Statistical analysis of fluorescence correlation spectroscopy: the standard deviation and bias,” Biophys. J.84, 2030–2042 (2003).

27. S.-M. Guo, J. He, N. Monnier, G. Sun, T. Wohland, and M. Bathe, “Bayesian approach to the analysis of fluorescence correlationspectroscopy data ii: Application to simulated and in vitro data,” Anal. Chem. 84, 3880–3888 (2012).

28. D. W. Marquardt, “An algorithm for least-squares estimation of nonlinear parameters,” J. Soc. Ind. Appl. Math. 11, 431–441 (1963).29. K. Levenberg, “A method for the solution of certain nonlinear problems in least squares,” Quart. Appl. Math. 2, 164–168 (1944).30. J. Wuttke, “lmfit 3.2 – a c/c++ routine for levenberg-marquardt minimization with wrapper for least-squares curve fitting, based on

work by B.S. Garbow, K.E. Hillstrom, J.J. Moré, and S. Moshier, available at http://apps.jcns.fz-juelich.de/doku/sc/lmfit,” (2010).

31. A. Corana, M. Marchesi, C. Martini, and S. Ridella, “Minimizing multimodal functions of continuous variables with the ”simulatedannealing” algorithm corrigenda for this article is available here,” ACM T. Math. Software 13, 262–280 (1987).

32. J. R. Unruh and E. Gratton, “Analysis of molecular concentration and brightness from fluorescence fluctuation data with an electronmultiplied ccd camera,” Biophysical journal 95, 5385–5398 (2008).

Contents

S1 Details on the two lightsheet microscopes SPIM-1 and SPIM-2 4S1.1 Optical setup of the SPIMs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 4S1.2 Camera Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5S1.3 Sample Mounting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

S2 SPIM-FCCS alignment and calibration 6S2.1 Alignment procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6S2.2 Bead Scans . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7S2.3 PSF calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8S2.4 Stability of the setups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10S2.5 Volume overlap: objective scan . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

S3 Sample preparation 12S3.1 Cell culture protocols: FuGENE . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12S3.2 Cell culture protocols: Neon transfection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13S3.3 Small and giant unilamellar vesicle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 13

S4 Additional SPIM-FC(C)S example data 14S4.1 SPIM-FCS example measurements in buffer, lipid and live cells . . . . . . . . . . . . . . . . . . 14S4.2 Additional data from 607bp DNA measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . 14S4.3 SPIM raw data and overview images . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15S4.4 EGFR/PMT in CHO cell membranes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

S5 Confocal FCCS measurements 17S5.1 Confocal FCS measurements 1 (Heidelberg) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17S5.2 Confocal FCS measurements 2 (Singapore) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17S5.3 Sample preparation for confocal microscopes . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17S5.4 Confocal volume calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17S5.5 Confocal data Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18

S6 Derivation of the SPIM-FCCS Correlation Functions 19S6.1 Notation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19S6.2 FCCS Correlation Functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20S6.3 SPIM molecular detection functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21S6.4 SPIM correlation functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21S6.5 Confocal Microscopy correlation functions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22S6.6 FCCS correlation functions with different mobility modes . . . . . . . . . . . . . . . . . . . . . . 22S6.7 Multiple components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22S6.8 Background Correction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23S6.9 Triplet and other Blinking Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

S7 Testing SPIM-FCCS with simulations 24S7.1 Simulation code . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24S7.2 Test of the SPIM-FCCS models and fitting routines . . . . . . . . . . . . . . . . . . . . . . . . . 26S7.3 Crosscorrelation amplitude error in dependence of alignment accuracy . . . . . . . . . . . . . . . 26

http://apps.jcns.fz-juelich.de/doku/sc/lmfithttp://apps.jcns.fz-juelich.de/doku/sc/lmfit

S8 Data Evaluation 27S8.1 Background Correction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27S8.2 Bleach Correction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28S8.3 Correlation & Fitting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30S8.4 Global Model Fitting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30S8.5 Performance . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31S8.6 Calibration of the measured concentration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31

S1. Details on the two lightsheet microscopes SPIM-1 and SPIM-2

S1.1. Optical setup of the SPIMs

Tab. ST1 summarizes the components used in SPIM-1 and SPIM-2. An explanation of the optical setup is given inthe main text. The sketch in the main text shows the setup of SPIM-1, whereas SPIM-2 is shown in Fig. S1.

We found it to be advantageous to mount the beam dichroic beam combiner onto a piezo-driven kinematic mount(customized MDI-H, Radiant Dyes, Wermelskirchen, Germany) as this allows a very fine adjustment of the twolightsheets with respect to each other.

component SPIM1 SPIM2blue laser Cobolt Calypso 491nm, 25mW Coherent Lasers OBIS488 LX, 100mWblue beam expander Sill S6ASS2075/067, 1x-8x 6x custom builtgreen laser Cobolt Jive 561nm, 25mW Oxxius, SLIM-561green beam expander Qioptiq bm.x VIS-YAG 5x 6x custom builtdochroic combiner Chroma zt488/594rpc Semrock 552 nm edge LaserMUXcylindrical lens Newport CKX18-C, f = 100 mm Edmund Optics, f = 75 mmlightsheet objective Nikon Plan Fluor 10x/0.3 Olympus SLMPlan 20x/0.25detection objective Nikon CFI Apo-W NIR 60x/1.0 Olympus LUMPLFLN 60x/1.0 Wbeam splitter 565 mm 565 mmgreen filter Semrock BrightLine HC525/50 Semrock BrightLine HC525/50red filter Semrock Edge Basic 561LP Semrock Razor Edge 568 nm

or Semrock Brightline HC 593LPtube lens Nikon MXA20696, f = 200 mm Olympus LU074700, f = 180 mmcamera Andor iXon X3 860 Andor iXon X3 860

Table ST1. Important components of the two microscopes used in this paper. The focal lengths of the different lenses aregiven as f = ....

EMCCD

100 mW MAXIMUM AT 488nm

MELLES GRIOT 561 nm

LASER

100 m

W

MAXIM

UM

AT 488nm

M

ELLES GRIO

T

48

8 n

m

LA

SE

R

Y

XZ

1.0

0.6

0.2

[µm] -4 0 4

σz

=1280561

σz

=1220488

nm

Side View

TopView

[5]

[1]

[2]

[3]

[4][6]

[10]

[7]

[8]

[9]

488nm

561nm

[11] 20 µm

Fig. S1. Schematic for dual color SPIM microscope (Singapore), [1-2]- Laser lines 488/561 nm, [3]- Dichroic mirror to combinelasers, [4]- Adjustable slit to avoid any back reflected light, [5]- Cylindrical lens to project laser light in direction of curvedsurface of the cylindrical lens, [6]- Projection objective to create micrometer thick light-sheet, [7]- Sample camber filled withimaging buffer or 1xPBS, [8]- High NA1.0 water dipping detection objective to collect the fluorescence, [9]- Tube lens, [10]-Image splitter optics (DualView) to separate the green and red fluorescence and focus on to the same camera sensor, [11]-EMCCD camera sensor to acquire image series. And inset image shows typical light sheet image on to the camera with it’sintensity profile.

S1.2. Camera Properties

Both setups use an Andor iXon X3 860 electron-multiplying CCD camera. We used the frame-transfer mode ofthis camera to gain high acquisition times for the measurements. The following Tab. ST2 summarizes the typicalacquisition settings used for out SPIM-FCCS measurements.

sample: ROI EM-gain preamp. gain # frames exposure time readout time frame timebeads 128×20 100 4.7× 100.000 470 µs 60 µs 530 µsSUVs 128×20 100 4.7× 100.000 1040 µs 60 µs 980 µsDNA 128×6 300 4.7× 700.000 230 µs 60 µs 290 µsDNA 128×4 300 4.7× 700.000 210 µs 60 µs 270 µscells 128×20 300 4.7× 100.000 470 µs 60 µs 530 µs

Table ST2. typical camera settings

S1.3. Sample Mounting

Fig. S2 illustrates the different types of sample mounting. For FC(C)S calibration, different samples (as e.g.described in Ref. [1]) were filled into small, heat-sealed sample bags made from thin transparent foil with arefractive index matching that of water (fluorinated polyethylene propylene films, thickness 13.0 µm, refractiveindex 1.341− 1.347, Katco Ltd., United Kingdom, or Lumox Folie 25 M, thickness 25 µm, Sarstedt AG & Co,Nümbrecht, Germany). Between 20 µl and 50 µl were filled into each of these bags.

Cells were grown (as described in the main text and section S3) on small glass pieces cut from No. 3 cover slips(0.28 mm−0.32 mm thick, No. 16301, Neolab, Heidelberg, Germany). The glass pieces were washed with acetoneor 70% ethanol and the with deionized water. Finally they were sterilized before use.

Fig. S2. Photographs of samples, as used for the SPIM-FCS/FCCS measurements. Left: A sample bag from thin foil. Right: acover slip with cells in front of the detection objective (no sample chamber). The self-clamping tweezers are the same on bothsides.

S2. SPIM-FCCS alignment and calibration

S2.1. Alignment procedure

The alignment process for SPIM-FCCS is more complicated than that for SPIM-FCS, still with some simple toolsit can be routinely done within a few minutes. First the blue laser is aligned as usual for a SPIM. A mirror ismounted under 45◦ in the sample chamber, which allows to directly observe the light sheet. The second (green)laser is then overlay to the first by means of the beam combining dichroic mirror and the separate beam expander.The first allows to set the beam position perpendicular to the observation plane and the latter allows to position thetwo beam waists over each other along the x-direction (light sheet’s direction of propagation). The alignment canbe checked by scanning the 45◦ mirror along the x-axis. See Fig. S6A for a sketch of the setup and Fig. S6B,Cexemplary results. The two light sheets can be aligned typically with a distance in z-direction δ z of less than30−100 nm (at a typical light sheet width of ∼ 1.2−1.3 µm).

In a second step the DualView optics was aligned, by imaging an electron microscopy grid with 16.9 µm or12.7 µm grid spacing (1500 or 2000 lines per inch; Latech Scientific Supply Pte. Ltd, Singapore) in transmissionillumination mode. This step is crucial, as the MDE of the two camera pixels for the red and green fluorescencechannel have to be overlay as perfectly as possible. The DualView allows to position both of the two color channelsindependently on the camera chip. During the alignment, the image cross-correlation coefficient IC betweenthe two image halves {Lx,y} and {Lx,y} of width w and height h is maximized to position the images with subpixel-accuracy:

IC =∑x,y(Lxy−L) · (Rxy−R)

w ·h ·√

σ2L ·σ2R, (S1)

where the average of an image L is L = ∑x,y

Lxy/(w ·h) and its variance is given by σ2L = ∑x,y(Lxy−L)2/(w ·h). The

results of the alignment were checked by a z-scan of fluorescent microspheres embedded in a gel cylinder (seeFig. S6D and section S1.3 for details on the preparation). Fitting a 3D Gaussian model function to each beadin both channels allowed us to measure the displacement of the MDEs in all directions (see section S2.2). Weroutinely reach a lateral and longitudinal displacement δx,δy and δ z of better than 100 nm. Exemplary results areshown in Fig. S6D,E.

Non overlapping spilt view images

Overlapping spilt view images

(A)

(B)

0 255Red

Red 255

25

5G

ree

nG

ree

n2

55

Fig. S3. Image splitter alignmen with co-localizationt: Transmitted overlapping (green and red channel) image of TEM gridand its intensity based co-localization of green and red channels for the alignment of image splitter. This estimates maximumoverlap of green and red channel for camera based FCCS measurements.

Fig. S4. Image splitter alignment with image cross-correlation: The images show the difference image Lxy−Rxy in false-colors,as implemented in QuickFit 3.0. The left most image shows the true image (also in false colors) of an EM-grid (1500 lines/inch)in transmission illumination mode. Pixel size if 400×400 nm2.

S2.2. Bead Scans

Beads in a gel cylinder for the PSF determination were prepared as follows:

1. Dissolve 0.5% Phytagel and 0.1% MgSO4 in 20−40 ml of deionized water

2. Heat until gel has dissolved

3. Let cool down to around 40◦C and mix with beads (e.g. 7 µl of 100 nm-diameter TetaSpec Microspherestock with 1 ml of gel) by vortexing the gel in an Eppendorf tube

4. Cut the tip of a standard 1 ml syringe (inner diameter 4.6 mm) and drawn up∼ 400 µl of the fluid gel. Ensurethat no air is trapped bewteen the gel and the plug.

Data was then acquired and evaluated as follows (a more detailed description of these methods can be found in thesupplementary information of [1]):

1. After the gel has solidified, extrude a few millimeters of the gel in front of the detection objective and imagein the first millimeter of the gel. Let the gel settle inside the sample chamber for a few minutes.

2. z-stacks were recorded with 100−200 nm step size and 2000 frames using both lasers for illumination.

3. a custom Matlab script (Matlab 2012a, MathWorks, Ismaning, Germany) evaluated the stacks automaticallyto yield the results shown in the main text [2]:

(a) One image half was segmented according to intensity and from every connected set of pixels only theone pixel with the highest intensity was used as first estimate for the bead position. The script also tookcare to keep a minimum distance of 3 pixels (in x-, y- and z-direction) between any two initial beadpositions.

(b) A region-of-interest (ROI) around each bead position was cut from both channels (i. e. the beadpositions determined in one channel are used to evaluate both) an:

(c) Fit three 1D Gaussian functions to x-, y- and z-cuts through the brightest pixel of each bead.(d) Fit a 3D Gaussian function to the whole bead: From the position~r0,g and~r0,r of the fit 3D gaussian in

both channels the displacement ~d :=~r0,g-~r0,r between the color channels could be calculated.(e) Only those beads were used for further evaluations, where the inter-channel displacement was in a

given range: −1 µm ≤ |~d| ≤ 1 µm. Also beads with unreasonably high or low widths were excluded,as these indicate bad fits or that no bead was inside the ROI.

(f) Histograms of the different bead widths and the displacements ~d were created.

The 1/e2-widths ψxy(wxy) obtained in the fits described above had to be corrected for the finite pixel size: Forthis correction a 1D molecular detection efficiency function (MDE) was set up:

MDE(x;a,wxy) = Rect(x;a)?PSF(x;wxy) =1

N

a/2∫−a/2

exp

(−2 · (x−ξ )

2

w2xy

)dξ (S2)

where Rect(x;a) is 1 for −a/2 ≤ x ≤ a/2 and 0 anywhere else (i. e. a is the pixel width), wxy is the 1/e2-widthof the Gaussian PSF PSF(x;wxy), ? denotes a convolution and N is a normalization constant. Then a simple1D-Gaussian PSF(x;ψxy) was fitted to Eq. (S2), using a least-squares scheme. From the resulting plot ψxy(wxy)(see Fig. S5) the corrected wxy can be read.

Fig. S5. Plot of the width of a Gaussian function fitted to the pixel MDE in Eq. (S2) for different pixel-widths a. The curveψxy(wxy) = wxy is shown in green. On the RHS two example MDE curves (blue) and fits (red, dotted) are shown.

Fig. S6. Exemplary results of a full characterization of the lightsheet and the molecular detection efficiency of SPIM-1. (A)Mounting of a mirror under 45deg to align and measure the light sheets. (B) Light sheet width (1/e2-half width) of the blue andgreen light sheet as determined with the setup in (A). (C) displacement between the central peaks of the lightsheets from (B).(D) Mounting of a gel cylinder (0.5% PhytaGel) with 100 nm fluorescent beads. (E) example PSF (F) Distribution of the lateral(δx,δy) and longitudinal (δ z) displacements of multi-colored beads in a gel between the green and red detection channel.

S2.3. PSF calibration

As already discussed in [1,3] SPIM-FCS needs a calibration of the PSF-size wg and wr, but in contrast to confocalFCS, the calibration does not need a sample of known diffusion coefficient. We use the known pixel size of thecamera and magnification of the optical system as a ruler to determine the absolute diffusion coefficient of acalibration sample. Then this diffusion coefficient can be used in a second step to estimate the PSF-sizes wg andwr. The calibration is performed in these steps:

1. acquire a measurement of any calibration sample sealed in sample bags, as described in section S1.3, we

used either:

• 100 nm fluorescent microspheres (FluoSpheres YG [F-8803] or FluoSpheres RT [F-8801], Invitrogen)diluted to ∼ 0.5 nM in a buffer solution (pH7.5)

• Atto-488 & Atto-568 dyes dissolved in 1xPBS (Fluka BioChemika, Sigma-Aldrich, Singapore)

2. The longitudinal widths zg and zr were estimated from bead scans or the measured light sheet width, takinginto account the depth of focus of the detection objective.

3. Autocorrelation functions were calculated using ImFCS or QuickFit 3.0 for different binnings of the pixels.Usually a binning between 1×1 and 5×5 was used, giving a pixel size between 0.4× .4 µm2 and 2×2 µm2in the object plane (taking into account the 24 µm pixels of our Andor iXon X3 860 and the magnification of60×).

4. At large pixel sizes a� wg,wr (higher binnings), the diffusion coefficient measured with SPIM-FCS ismostly independent of the value of wg,wr (see Fig. S7). The different sets of ACFs were fitted with thestandard SPIM-FCS autocorrelation model:

gρ(τ) =1√

π · zρ a2C

√

4Dτ +w2ρ√

π ·a·

e−(

a2

4Dτ+w2ρ

)−1

++ erf

a√4Dτ +w2ρ

2 ·(1+ 4Dτz2ρ

)−1/2+G∞, ρ ∈ {g, r}

the parameters a and zg,zr were fixed to their known values (from the respective binning and step 2). Thefits were performed for different values of wg,wr chosen around the expected value of wg,wr (e.g. between400 nm and 800 nm). For increasing binning the curves wg,wr against the pixel size a converge against thetrue diffusion coefficient D. Finally the value of D is calculated by averaging the fit results of all differentwg,wr at the largest binning. See Fig. S7(A,D).

5. Finally the unbinned data was fitted again, now using the diffusion coefficient D determined in the last stepand a and zg,zr still fixed. From this final fit a good estimate of the lateral focal size wg,wr can be extractedat the lowest binning. See Fig. S7(B,E).

6. As a second method of SPIM calibration, here first time we used fluorescent organic dyes to determinethe lateral PSF. It should be noted that, the current EMCCD camera frame rate could achieve ∼ 3,700 fpsfor 4× 128 pixel camera resolution, and this temporal resolution would not be sufficient to get accuratediffusion coefficient or concentration of small-fast diffusing organic dyes Ref. [4]. However, the fluorescentsignal can be easily auto-correlate (at different binning of camera pixels, above method step 1-5) and lateralPSF of the microscope can be determined. As a demonstration, here we used mixed solution of 5− 10 nmAtto-488/Atto-565 dyes and recorded their fluorescence signal at 270 µs camera exposure (4× 128 pixel,150,000−200,000 frames). Acquired image frames can be easily analyzed, to get reliable lateral PSF of themicroscope (stated above; step 1-5)Fig. S7(C-E).

250 500 750 1000 1250 1500 1750 2000

pixel size [nm]

2.5

5

7.5

10

diffusion

coefficient

[µm²/s]

wxy=100nm

wxy=200nm

wxy=300nm

wxy=400nm

wxy=500nm

250 500 750 1000 1250 1500 1750 2000

pixel size [nm]

100

200

300

400

500

lateralfocussize

wxy[nm]

calibration D=(5.66611 ± 0.801874)µm²/s

wxy=(300.705 ± 15.2252)nm

(A ) (B)

D [µ

m2/s

]

Pixel size [nm]

900

800

700

600

16001200800400

ωx

y [

nm

]

σxy

= 780±110610

σxy

= 690±125510

[nm]

400

300

200

100

16001200800400Pixel size [nm]

400500

900

[nm]

(C) (D) (E)

Atto 565 Bin-1 Bin-2 Bin-3 Bin-4

G(τ

)

τ [sec]

2.0

1.0

1.5

10-3 10-2 10-1 10-0

Fig. S7. Example calibration from SPIM-1 for 100 nm beads acquired at 37◦C. (A) diffusion coefficient fitted for differentvalues of wxy = wg and at different pixel sizes. (B) lateral PSF size wxy = wg fitted at different pixel sizes, fixing the diffusioncoefficient D(a= 200 nm) = (5.7±0.8) µm2/s, acquired from (A); (C) ACFs of Atto-565 dye (Atto-488 ACFs not shown here)at different camera binning, (D-E) Lateral PSF size determination for both green (Atto-448 dye) and red channel (Atto-565 dye).

Tab. ST3 summarizes typical focus sizes we obtained for both SPIM setups.

Sample Parameter SPIM-1 SPIM-2TetraSpec in gel zg[nm] (1150±100) (1220±120)

zr[nm] (1180±100) (1280±140)TetraSpec in solut. wg[nm] (604±50) (680±110)

wr[nm] (616±50) (760±125)

Table ST3. Results of the SPIM-FCCS calibration. The focus heights were determined by bead scans and the widths bySPIM-FCS calibrations as described in the text.

S2.4. Stability of the setups

As already discussed before the SPIM setups are very stable over time. Fig. S8 shows that the PSF size of a SPIMdoes not change significantly over the course of more than half a year.

Fig. S8. Results of PSF-calibrations on SPIM-1 over nine months.

S2.5. Volume overlap: objective scan

Light sheet microscopes typically have two objectives in orthogonal direction, one for the illumination (low NA)and a second one (high NA) for imaging the fluorescence signal onto the camera sensor (see figure Fig. S1). Thisconfiguration for the objectives requires a perfect overlap of the illumination light sheet plane and the detectionplane created by high NA detection objective. So the proper alignment of a SPIM for FCS/FCCS measurements caneasily be tested by measuring the diffusion coefficient D and the particle number N in the same sample at differentdisplacements of the detection and projection objective from their ideal (after alignment) positions. Fig. S9 showsthe results of such a measurement. Here the measurements are shown for the green detection channel. The relativepositions of both illumination (5 µm step, shown in blue circles) and detection (1 µm step, green circles) objectivesrespectively were changed systematically. The optimal alignment will be in the region with lowest particle number(empty circles) and highest diffusion coefficient (filled circles), i. e. smallest (and best) detection volume.

projection

objective

detection

objective

1 µm step

200 50 100200N D µm2/s

5 µm step

200

25

30

20

100

ND

µm

2/s

Fig. S9. Diffusion coefficient D and particle number N measured at different displacements of the two objectives from the idealposition, determined with our alignment procedure.

S3. Sample preparation

S3.1. Cell culture protocols: FuGENE

Adherent HeLa cells (provided by F. Rösl, DKFZ, Heidelberg, Germany) were grown in a 5% CO2 humidifiedatmosphere at 37◦C in a phenol red-free DMEM growth medium (Invitrogen Life Technologies, Carlsbad, USA)supplemented with 10% fetal calf serum and 1% Glutamine. First, the growth medium was removed from the flaskand the cells are washed with 5 ml Hanks balanced salts solution (PAN-Biotech, Aidenbach, Germany). Cells weretrypsinized, by incubating the cells for ∼ 1 min with 5 ml of Trypsin/EDTA solution. To stop the trypsinizationprocess we add 10 ml of DMEM medium. The cells were diluted (between 1:5 and 1:80) in fresh medium andseeded in new cell culture flask, depending on the requirements. For the in vivo SPIM measurements the cellsare seeded on small glass pieces in a 35 mm petri dish. Transfection with the mammalian expression vectors iscarried out with FuGENE HD transfection reagent (Roche Diagnostics, Mannheim, Germany) as proposed by themanufacturer. The cells were transfected 24−48 h before the measurements. Details on the used amounts of DNAand transfection reagent are given in Tab. ST4.

plasmid amount of plasmid transfection cells

eGFP-mRFP1fusion protein 100 ng

45 µl medium4 µl FuGENE HD HeLa

IRES: eGFP andmRFP1 monomers

100−140 ng — ” — HeLa

PMT-eGFP +PMT-mRFP1

200 ng + 400 ng45 µl medium6 µl FuGENE HD CHO

GFP-EGFR-mRFP1 1−1.5 µg — ” — CHO

eGFP 100−150 ng Neon Transfection RBL-2H3

c-Fos-eGFP 1−1.5 µg — ” — HeLa

Table ST4. Detailed protocols used for the transfection of cells in this paper. The amounts in the table are given forstandard 35 mm petri dishes.

As a positive control for maximum cross-correlation in vivo the pSV-eGFP-mRFP1 was used. It is a two-colorfusion protein of eGFP and mRFP1 separated by a 7 amino acid linker. As a control for no cross-correlation weused the pIRES2-eGFP-mRFP1. It is an internal ribosomal entry site (IRES) vector, which expresses the dyesseparately. The plasmid construction is described in more detail in Ref. [5]. Note that due to the close proximity ofthe two fluorophores, our eGFP-mRFP1 fusion construct, shows ∼ 30% Förster resonance energy transfer (FRET)efficiency [6].

To show interaction in cell membranes, we used three constructs already described in [7]: As negative controla red or green fluorescent protein (GFP or mRFP) was fused to a plasma membrane targeting sequence (PMT) onit’s N-terminus. As positive control we used an epidermal growth factor receptor (EGFR) fused to an mRFP at it’sextracellular and and a GFP on the intracellular side.

Fig. S10. (A) Schematic of cell sectioning by the light sheet, (B) Sketch of a SPIM image of a cell with labeled molecules inthe cytoplasm and nucleus and (C) Sketch of a SPIM image of a cell with labeled molecules in the membrane only.

S3.2. Cell culture protocols: Neon transfection

HeLa and RBL-2H3 (gift from Min Wu, CBIS, NUS, Singapore) cells were maintained as described in theprevious section. For transfection we used a Neon-Transfection system (Invitrogen, Singapore) with 10 µl goldtips (resuspension buffer R and electrolyte buffer E) using the protocoll recommendet by the manufacturer (NeonTransfection protocols). The used amounts of plasmid are given in Tab. ST4. Following transfection, the cells werespread on the cover slips (as described above) and finally used for the live cell measurements.

S3.3. Small and giant unilamellar vesicle

As a test sample that can easily be created in any lab, we used small unilamellar vesicle (SUVs). Single- anddouble-labeled vesicles were prepared according to the protocol given in [8]. All lipids, fluorescently labeled lipidsand cholestrol were purchased from Avanti Polar Lipids (Alabaster, Alabama, USA). Briefly, a stock solutionof POPG lipid (1-hexadecanoyl-2-(9Z-octadecenoyl)-sn-glycero-3-phospho-(1’-rac-glycerol), sodium salt) andhead group labelled lipids Rhod-PE (1,2-dimyristoyl-sn-glycero-3-phosphoethanolamine-N-(lissamine rhodamineB sulfonyl), ammonium salt) and/or Bdp-Chol (23-(dipyrrometheneboron difluoride)-24-norcholesterol) wereprepared individually in chloroform (Sigma Aldrich, Singapore) and mixed thoroughly in a round bottom flask.The chloroform was evaporated (∼ 3 hrs) on a rotary evaporator (Rotavap R-210, Buchi, Switzerland). Lipids werethen hydrated in 500 µl buffer containing 10 mM HEPES and 150 mM NaCl (Buffer A, pH 7.4) and sonicatedfor ∼ 30 min in a water bath sonicator (FB15051 Model, Fisher Scientific, Singapore). To avoid large variationsin vesicle size, we centrifuged the vesicles at 10,000− 15,000 g for 30− 40 min and supernatant used for themeasurements.

Giant unilamellar vesicles (GUVs, POPC 89%, POPG 10% and PI(4,5)P2 1% and TopFluor PI(4,5)P2, Avantipolar lipids) were prepared by gentle hydration. Then the GUV solution was mixed with a low melting agar solution(0.5-0.8 %, ∼ 40°C temperature) and sealed in sample bag for the measurements.

S4. Additional SPIM-FC(C)S example data

Here in this section first we give additional test sample measurements that show what is possible with SPIMmicroscope, when used for fluorescence correlation spectroscopy. Later we present additional SPIM-FCCSmeasurements, in order to give an overview of the un- and pre-processed raw data for FCCS analysis.

S4.1. SPIM-FCS example measurements in buffer, lipid and live cells

0.5

0.3

0.1

02.5 10 1 4 8

2.5

2.0

1.5

0

G(τ

)G

(τ)

x 1

0-4

0

2

50 100 150

50 100 150

D [µm2/s]

C [nM]

G(τ

) x

10

-3

2.57.5 7.52.5

Atto 488

Bin-2

G(τ

)

τ [sec] τ [sec]

2.0

0

1.5

10-3 10-2 10-1 100

10-3 10-2 10-1 100

10-2 10-1 100

250 350

20 30 40

D [µm2/s]

D [µm2/s] C [µM]

C [nM]

10-2 10-1 100

D [µm2/s] C [µM]

(Α) (Β)

(D)(C)

Bin-2

GUV: POPC-POPG-

PI(4,5)P2

EGFP in

RBL-2H3 cell

c-Fos-EGFP in

HeLa cell

Fig. S11. SPIM-FCS measurements: (A) Atto-488 dye in buffer, (B) GUV embedded in Agar, (C) monomeric EGFP expressedin RBL cells and (D) c-Fos-EGFP expressed in HeLa cells.

S4.2. Additional data from 607bp DNA measurement

Fig. S12. (A) Comparison between SPIM-FCCS and confocal FCCS using a single/double-labeled 607bp dsDNA with fourdifferent robust linear regressions, (B) example CFs for the negative sample, (C) example CFs for the sample with mostdouble-labeled DNA stands. In (B,C) the dashed blue line is the cross-correlation amplitude explained by crosstalk only(κgr = 11.8%). Colored dashed lines are SPIM-FCCS fits using the models in the main text and DA = DB = DAB.

S4.3. SPIM raw data and overview images

Fig. S13. Example images from a typical SPIM-FCCS measurement in a HeLa cell expressing the eGFP-mRFP1-dimer (thesame cell, as shown in the main text).

S4.4. EGFR/PMT in CHO cell membranes

EGFR PMT

0

1

2

3

rel. d

imer

concentr

ati

on p

AB

statistics

raw data

0.5 1 1.5 2 2.5 3 3.5

relative dimer concentration pAB

=cAB

/min(cA,c

B)

5

10

15

frequency

PMT

EGFR(A) (B)

Fig. S14. Statistics of the results obtained from CHO cells transfected with single-labeled PMT or double-labeled EGFR. (A)boxplot for each sample (circle=average, line=median, box=25% to 75%-quantile, whiskers=min/max) and as blue trianglesthe average and standard deviation over all pixels each cell (triangles are scattered randomly in x-direction to deparate them abit). (B) histograms of the data shown in (A).

S5. Confocal FCCS measurements

S5.1. Confocal FCS measurements 1 (Heidelberg)

Confocal FCCS measurements for comparison to the SPIM-FCCS measurements were performed on a customFCCS setup [9], based on an inverted Olympus IX-70 microscope (Olympus, Hamburg, Germany), a galvanometricscanner to position the focus and a 60x/NA1.2 objective. Light from the 488 nm and 568 nm lines of an ArKr-Laser(CVI Melles Griot, Bensheim, Germany), filtered by an AOTF (AOTF Nc, AA Opto Electronic, France), wasreflected into the microscope resulting in about 3 µW of laser power above the objective. Fluorescence was detectedwith two avalanche photodiodes (SPCM-AQR-13, Perkin-Elmer, Wellesley, USA) and correlated by a hardwarecorrelator card (ALV-5000/E, ALV Laser GmbH, Langen, Germany). The two color channels were split with a*** dichroic mirror and the filtered with a 535DF30 bandpass filter in the green and a OG590 longpass filter inthe red channel. Data evaluation was performed using QuickFit 3.0 [10]. As a FCCS-calibration standard, we useda 170 bp dsDNA (carrying an Alexa488 and an Alexa594 dye at the two ends). With this sample we achieved amaximum relative crosscorrelation of (see also Fig. S15)

ggr(10−5 s)min

[ggg(10−5 s),grr(10−5 s)

] ≈ 55%Measurements were carried out at room temperature (22− 24◦C) with laser powers of ∼ 3− 4 µW for cells

and ∼ 7− 8 µW for in-vitro samples on both laser lines. For in-vitro samples First 6 runs á 30 s each weremeasured. For cell measurements, a confocal image of each cell was first acquired, then in that image pointsfor FCS measurements were selected. On each point an FCS measurement with 6 runs á 10 s each was taken.The model was fit to the average of these runs and their standard deviation was used for weighting. Runs withdeviations due to e.g. cell movement were excluded before the fit.

Fig. S15. Auto- and crosscorrelation curves of a confocal measurement of a 170bp DNA sample. The data points represent theaverage and the transparent ranges are the standard deviations of 10 consecutive runs á 30 s.

S5.2. Confocal FCS measurements 2 (Singapore)

The confocal FCCS setup used in Singapore was described previously [11] and we will provide only a briefdescription of the instrument. The confocal FCS system is based on a modified Olympus FV 300 confocalmicroscope (Olympus, Tokyo, Japan). Fluorescence was excited with the 488 nm and 543 nm lines of an argon ionlaser (Melles Griot, Albuquerque, NM, USA), which was focused by a water-immersion objective (60x, NA 1.2;Olympus, Tokyo, Japan) into the sample. The fluorescence light emitted from the sample was collected by the sameobjective and passed through a 560DCLP dichroic mirror (Omega Optical, Brattleboro, VT) followed by band-passfilters (510AF23/615DF45, Omega Optical, VT, USA). Further it passes through 3x magnification system and wasspatially filtered by a 150 µm pinhole. The light from the pinhole was imaged onto an avalanche photodiode whichoperated in photon counting mode (SPCM-AQR-14-FC; Pacer, Berkshire, UK). The autocorrelation curves werecomputed online by a hardware correlator (Flex02-01D; Correlator.com, Bridgewater, NJ, USA). The laser powerwas adjusted according to the samples (0.2−20 µW), as measured in front of the microscope objective.

S5.3. Sample preparation for confocal microscopes

In-vitro samples were filled into 8-well chambered coverslides (Lab-Tek chambered #1.0 borosilicate coverglass,No. 155411, NUNC, Rochester, USA) or 18-well slides (µ-Slide 18 well, No. 81821, Ibidi, Martinsried, Germany).

S5.4. Confocal volume calibration

Calibration of the two confocal microscopes was done with solutions of different small dye molecules. Theassumed diffusion coefficients of these are given in Tab. ST5.

Dye D20,W [ µm2/s] ReferenceAtto488 (370±9) [12]Alexa488 407 [13]Alexa568 317 [14]

Table ST5. Diffusion coefficients D20,W at 20◦C of dye molecules dissolved in water, as they were used for calibration

S5.5. Confocal data Evaluation

FCS measurements were evaluated using a normal or anomalous diffusion confocal (Gaussian) fitting model withC components:

normal diffusion : g(τ) = G∞ +1N· 1−Θt +Θt · e

−t/τt

1−Θt·

C

∑i=1

ρi

(1+

4Di · τw2xy

)−1·

(1+

4Di · τK2 ·w2xy

)−1/2(S3)

anomalous diffusion : g(τ) = G∞ +1N· 1−Θt +Θt · e

−t/τt

1−Θt·

(1+

4Γ · τα

w2xy

)−1·

(1+

4Γ · τα

K2 ·w2xy

)−1/2(S4)

where G∞ is the convergence value for long correlation times, Θt is the triplet fraction and τt its correlation time,N is the number of particles in the focal volume, Di is the normal diffusion coefficient of the particles of speciesi = 1..C, Γ is the anomalous diffusion coefficient and α the anomality parameter (α = 1 is normal diffusion). Theaxial ratio of the Gaussian focal volume is denoted by K = wz/wxy. The lateral focal 1/e2-half width wxy wascalibrated as described above. The axial ratio K was fixed to 5 or 8, depending on the microscope. The fractionof the i-th component is denoted by ρi with ∑Ci=0 ρi = 1. The overall particle concentration in the sample can becalculated using the effective focal volume V (confocal)eff for the confocal case as:

c =N

V (confocal)eff=

Nπ3/2 ·w3xy ·K

(S5)

S6. Derivation of the SPIM-FCCS Correlation Functions

S6.1. Notation

This section extends the SPIM-FCCS theory given in the main text. Extending our previous work [1, 15, 16] onSPIM-FCS, we derived the SPIM-FCCS model functions within the framework described in [7, 17, 18], as thisallows to easily incorporate effects, as e. g. channel crosstalk or Förster resonance energy transfer (FRET). Thenotation used in the main text and this section is summarized in Tab. ST6 and is defined throughout the text, whereneeded.

Symbol explanation~r = (x,y,z)t position in spacet position in timeg,r channel indices for channel green and red. The index r/g is used when an

expression applies to either channel r or channel gA,B,AB species indices (a=green fluorophore, b=red fluorophore,

ab=double-labeled particle)gchannels(τ) normalized correlation function the given channel, e.g. ggg is the

crosscorrelation function between the green and red channel. (channels:gg=green autocorrelation, rr=red autocorrelation, gr=grenn-redcrosscorrelation)

gspecieschannels(τ) normalized correlation function of the given species in the givenchannel, e.g. gabgr is the crosscorrelation function between the green (g)and red(r) channel of species ab.

g∞ offset value of a normalized correlation functionGspecieschannels(τ) non-normalized correlation function between the given channels and for

the given speciesIchannel(t), Ichannel(t,~r) fluorescence intensity timetrace in the given channelIrawchannel(t,~r) raw fluorescence intensity timetrace after background and before bleach

correction in the given channel and at the given positionδ Ichannel(t) fluorescence fluctuations in the given channelMDEchannel(~r) molecular detection efficiency function of the given channelδ (~r) is the Dirac-δ distributionPSFchannel(~r) point spread function of the given channelηspecieschannel brightness of the given species in the given channelηchannel brightness of the given channelcchannel average particle concentration in the given channelcspecies concentration of a given molecular speciesκgr crostalk coefficient from the green into the red channel〈

I(m)channel〉

measured fluorescence intensity in the given channel〈B(m)channel

〉measured background signal intensity in the given channel

ILS(z) z intensity profile of the lightsheetfpixel,channel(~r) function that describes the position and size of a camera pixel in the

object plane (it is 1 inside the pixel and 0 outside)a,b pixel size in x- and y-directionzLS,channel lightsheet 1/e2-halfwidth in z-driectionzPSF,channel PSF 1/e2-halfwidth in z-driectionzchannel MDE 1/e2-halfwidth in z-driectionwchannel MDE 1/e2-halfwidth in x- and y-driectionDspecies normal diffusion coefficient of the given speciesΓspecies anomalous diffusion coefficient of the given speciesα anomality parameter~d = (δx,δy,δ z)t shift between the green and the red focus~v = (vx,vy,vz)t drift velocity of particles

Table ST6. Notation used for SPIM-FC(C)S theory

S6.2. FCCS Correlation Functions

We start again with the fluorescence signal emitted by a spot~r of a sample over time t. The fluorescence is splitinto two color channels (green and red) Ig(t;~r) and Ir(t;~r′), which may have a spatial displacement ~d =~r′−~r dueto e.g. optical abberations or missalignments. The normalized crosscorrelation function can now be written as:

gγρ(τ;~r) =〈Iγ(t;~r) · Iρ(t + τ;~r′)

〉〈Iγ(t;~r)

〉·〈Iρ(t;~r′)

〉 −1 = Gγρ(τ,~r′−~r)〈Iγ(t;~r)

〉·〈Iρ(t;~r′)

〉 −1 (S6)are calculated. Here Gγρ(τ) is the non-normalized correlation function between channels γ and ρ and 〈·〉 denotesa temporal average:

〈 f 〉= limT→∞

1T

T∫0

f (t) dt

To derive the FCCS correlation functions, we follow the same approach as in [7, 17, 18] and start from thefluorescence signals, which can be written for any color channel γ (typically for two-color FCCS γ ∈ {r,g}) andthe set of molecular species S= {A,B,AB, . . .} as:

Iγ(t) =∞∫∫∫−∞

MDEγ(~r) · ∑χ∈S

ηχγ cχ(t,~r) dV (S7)

Here ηχγ represents the fluorescence signal caused by a fluorophore bound to species χ in channel γ . The symbolcχ(t,~r) denotes the local particle concentration of species χ at time t and position~r and MDEγ(~r) is the moleculardetection efficiency of channel γ for an emitting particle at position~r, with the normalization condition:

∞∫∫∫−∞

MDEγ(~r) dV = 1 (S8)

Using these definitions, the correlation function Eq. (S6) can be written as:

gγρ(τ) =∑

χ∈Sηχγ η

χρ ·G

χγρ(τ)(

∑χ∈S

ηχγ · cχ

)·

(∑

χ∈Sηχρ · cχ

) . (S9)or explicitly for the case of a dimeric interaction A+B AB:

ggg(τ) =(ηAg )2GA(τ)+(ηABg )2GAB(τ)+(ηBg )2GB(τ)

(ηAg cA +ηABg cAB +ηBg cB)2

grr(τ) =(ηAr )2GA(τ)+(ηABr )2GAB(τ)+(ηBr )2GB(τ)

(ηAr cA +ηABr cAB +ηBr cB)2

ggr(τ) = grg(τ) =ηAg ηAr GA(τ)+ηABg ηABr GAB(τ)+ηBg ηBr GB(τ)

(ηAg cA +ηABg cAB +ηBg cB) · (ηAr cA +ηABr cAB +ηBr cB)

(S10)

Here we use the simplification (as usual in FCCS, cf. [7, 18]) that the time-averaged concentrations arehomogeneous on the scale of the MDEsizes (also note the normalization of the MDEEq. (S8)):

cχ =〈∫∫∫

MDEg(~r)cχ(t,~r) dV〉=

〈∫∫∫MDEr(~r)cχ(t,~r) dV

〉. (S11)

In Eq. (S9), Gχγρ(τ) are the unnormalized (cross-)correlation functions of species χ between channels γ and ρ:

Gχγρ(τ) =〈Iγ(t) · Iρ(t + τ)

〉ηχγ η

χρ

−1 = cχ ·∞∫∫∫−∞

∞∫∫∫−∞

MDEγ(~r) ·MDEρ(~r ′) ·φχ(~r,~r ′,τ) dV dV ′. (S12)

Here φχ(~r,~r ′,τ) is the Green’s function describing the motion of species χ . In the case of normal diffusion it canbe written as:

φχ(~r,~r ′,τ) =1(

4π ·Dχ τ)3/2 · exp[− (~r−~r ′)24Dχ τ

](S13)

S6.3. SPIM molecular detection functions

As in our previous publications [1,15,16], we model the SPIM-FC(C)S MDEas the convolution of a Gaussian PSFand a rectangular pixel of size a×b:

MDEρ(x,y,z) =1

N· ILS(z) ·

{[fpixel,ρ(x,y) ·δ (z)

]?PSFρ(x,y,z)

}=

=1

N· ILS(z) ·

a∫0

b∫0

PSFρ(x−µ,y−ν ,z) dν dµ, (S14)

where ? denotes a convolution, PSFρ(x,y,z) is the point-spread function describing the detection optics and N isa normalization constant determined, so:

∞∫∫∫−∞

MDEρ(~r) dV = 1 (S15)

The camera pixel in the object plane is described by

fpixel,ρ(x,y) =

{1 0≤ x≤ a and 0≤ y≤ b0 else

(S16)

The Gaussian PSF is defined as

PSFρ(x,y,z) =1

(π)3/2 ·wρ√zρ· exp

[−2 · x

2 + y2

w2ρ−2 · z

2

z2ρ

](S17)

where wρ is the lateral 1/e2-width of the rotational symmetric PSF and the axial 1/e2-extent of the MDEis givenby the axial extent of the lightsheet zLS,ρ and the PSF zPSF,ρ [1]:

1z2ρ

=1

z2LS,ρ+

1z2PSF,ρ

.

The parameters wρ and zρ can be calibrated using e.g. a measurement of fluorescent beads dissolved in water andbead-scans, as described in sections S2.2 and S2.3.

The MDE can now be written as:

MDEρ(x,y,z) =1

2ab ·√

2π · zρ·

[erf

(√2 · (a− x)

wρ

)+ erf

(√2 · x

wρ

)]·

·

[erf

(√2 · (b− y)

wρ

)+ erf

(√2 · y

wρ

)]· exp

[−2 · z

2

z2ρ

](S18)

S6.4. SPIM correlation functions

With the SPIM MDE as defined in the last section, the integrals in Eq. (S12) can be solved analytically. They canbe separated into three directional components Gχγρ(τ) = cχ ·G

χγρ;x(τ) ·G

χγρ;y(τ) ·G

χγρ;z(τ):

Gχγ,x(τ) =∞∫−∞

∞∫−∞

MDEγ(x) ·MDEγ ′(x′−δx) ·1√

4πDχ τ· exp

[− (x− x

′)2

4Dχ τ

]dx dx′ =

12a2·

(a−δx) · erf

√2 · (a−δx)√8Dχ +w2g +w2r

−2δx · erf √2 ·δx√

8Dχ +w2g +w2r

++ (a+δx) · erf

√2 · (a+δx)√8Dχ +w2g +w2r

++

√8Dχ +w2g +w2r√

2π·

[e− 2·(a−δx)

2

8Dχ τ+w2g+w2r −2 · e− 2δx

2

8Dχ τ+w2g+w2r + e− 2·(a+δx)

2

8Dχ τ+w2g+w2r

] . (S19)

Here MDEγ(x) is the part of the MDE-factor in the x-direction, assuming a separable MDEas in Eq. (S18) and forchannel γ,γ ′ ∈ {g, r}. The z-factor is the same as usually used in confocal FCS:

Gχγ,z(τ) =∞∫−∞

∞∫−∞

MDEγ(z) ·MDEγ ′(z′) ·1√

4πDχ τ· exp

[− (z− z

′)2

4Dχ τ

]dz dz′ =

=

√2π·

exp[− 2·δ z28Dχ τ+z2g+z2r

]√

8Dχ τ + z2g + z2r. (S20)

In the main text only correlation functions for the simple case of a vanishing shifts |~d| = |~r′−~r| → 0 betweenthe MDEs was given.

S6.5. Confocal Microscopy correlation functions

For sake of completenes we also state the non-normalized correlation functions for a purely Gaussian MDE

MDEγ(~r) =1

N· exp

[−2 · x

2 + y2

w2γ−2 · z

2

z2γ

](S21)

where N is a normalization factor to fulfill condition Eq. (S15). For normal diffusion, we then get:

Gχγγ ′(τ) = ηgηr · cX ·(

2π

)3/2·

exp[− 2·(δx

2+δy2)8Dχ τ+w2g+w2r

− 2·δ z28Dχ τ+z2g+z2r

](8Dχ τ +w2g +w2r

)·√

8Dχ τ + z2g + z2r. (S22)

For the simple case GAgg(τ) (autocorrelation in the green channel) and setting ~d = 0 this reduces to the wellknown form of the normalized confocal autocorrelation function:

ggg(τ) = g∞ +1

cA ·π3/2w2gwz·

(1+

4DAτw2g

)−1·

(1+

4DAτz2g

)−1/2=

= g∞ +1N·(

1+τ

τA

)−1·(

1+τ

K2 · τA

)−1/2where we used the definitions N = cA ·Veff = cA ·π3/2w2gwz for the particle number N in the effective volume Veff,K = zg/wg for the structure factor K and the correlation time τA = w2g/(4DA).

S6.6. FCCS correlation functions with different mobility modes

If an additional directed flow~v = (vx,vy,vz)t is present in the sample, this can also be incorporated into the model,by replacing:

δx→ δx+ vxτ, δy→ δy+ vyτ, δ z→ δ z+ vzτAlso anomalous diffusion can easily be incorporated by using the modified Green’s function:

φanomalous(~r,~r′,τ) =1(

4π ·ΓXτα)3/2 · exp[− (~r−~r′)24ΓXτα

], (S23)

Basically the expression derived in the main text and the last subsection can be used when the followingreplacement is performed:

Dχ τ → Γχ τα ,where Γχ is the anomalous diffusion coefficient and α the anomality parameter. Note that generally Γχ is a functionof α and has unit length2/timeα !

S6.7. Multiple components

Multiple Diffusion components can also be incorporated in a per-species manner, i.e. we assign several diffusioncoefficients Dχ,1,Dχ,2, ... to every species χ . For NC components, NC fractions ρχ,1, ... with the normalizationcondition

NC

∑i=1

ρχ,i = 1

are used. So far the single-species, non-normalized correlation functions Gχ(τ) ≡ Gχ(τ;Dχ) only depended on asingle diffusion coefficient. Now we add the additional diffusion components by replacing:

Gχ(τ)→NC

∑i=1

ρχ,i ·Gχ(τ;Dχ,i)

The Gχ(τ;Dχ) in this equation are the single-species correlation functions used throughout the paper and derivedin the previous sections S6.4-S6.5

S6.8. Background Correction

If the background contribution has not been corrected for before the correlation (in that case Bmr,g = 0), thenormalized correlation functions have to be corrected for this artifact by the transformation:

ggr(τ)→ ggr(τ) ·Fmg −B

mg

Fmg· F

mr −B

mr

Fmr(S24)

S6.9. Triplet and other Blinking Dynamics

Most dyes used in FCS and FCCS show on-off dynamics (”blinking”) where the fluorophore now and then getstrapped in a dark state for a certain time. In all dyes used in this paper one of these dark states is the triplet statewith its lifetime τT. This fast blinking leads to an additional decay term in the autocorrelation functions, that canbe incorporated into the model functions described in this section by extending the non-normalized autocorrelationfunctions for species χ [19]:

Gχgg(τ) → T χ(τ) ·Gχgg(τ) Gχrr(τ) → T χ(τ) ·Gχrr(τ)Gχgr(τ) → Gχgr(τ) Gχrg(τ) → Gχrg(τ).

(S25)

The triplet decay term T χ(τ) can be written as:

T χ(τ) =1−ΘT +ΘT · e−τ/τT

1−ΘT(S26)

where ΘT ∈ [0...1] is the fraction of molecules currently in the triplet state.Some of the fluorescent proteins used in this paper show an additional blinking dynamics on a longer timescale

due to conformational changes (e.g. mRFP, see Ref. [20]) or chemical reactions ( e.g. a protonation of GFP, seeRef. [21]). The parameters for such a second reaction will be denoted by ΘRand τRand can also be incorporatedinto Eq. (S26):

T χ(τ) =1−ΘT−ΘR +ΘT · e−τ/τT +ΘR · e−τ/τR

1−ΘT−ΘR(S27)

Note however that these dynamics are typically too fast (1 µs ≤ τT ≤ 5 µs and 10 µs ≤ τR ≤ 100 µs) to becaptured in SPIM-FC(C)S, so these corrections are not necessary there. But they have to be accounted for inconfocal FC(C)S (see also section S6.5).

S7. Testing SPIM-FCCS with simulations

S7.1. Simulation code

We tested several aspects of the models presented in this paper using an FCS simulation system already describedin [1,22,23]. We extended this simulation code with methods that allow to measure crosscorrelations between twoarbitrarily positioned foci. Here we will give only a short description of the code, as it was used for the resultsshown below:

We simulate the brownian random walk trajectories of three sets of particles A, B and AB. They carry eithera green (A), a red (B) or both (AB) fluorophores and all move with the same diffusion coefficient Dsim ≡ DA =DB = DAB. Different relative concentrations ranging between cAB/(cA + cB+ cAB) = 0..100% were used. Thenfor each relative concentration a set of foci was positioned in the simulational box For illumination we assume aflat lightsheet

ILS(x,y,z) = I0 · exp(−2 · z

2

w2LS

)with 1/e2-width wLS. For detection we use a function of the form of the MDEEq. (S18), also used in the SPIMcorrelation function derivation. The exact parameters are summarized in Tab. ST7 and model foci as typicallyfound in our SPIMs. For each simulation run (i. e. set of trajectories), we position one red detection focus in theorigin and several green detection foci were positioned a distance δx = 0 nm..2000 nm away in x-direction. Ineach timestep ti = i ·∆tsim and for each focus γ the average number of detected photons Nγ(ti) was estimated,based on the position of all random walkers, the illumination light distribution and the MDE. Then the numberof detected photons Nγ(ti) was drawn from a Poissonian distribution with average/variance Nγ(ti) to mimic thephoton noise. We performed simulations for two different values (κgr = 3.5% and κgr = 11.2%) of the green→redcrosstalk parameter κgr. Finally the auotorrelation function of each intensity stream and also the crosscorrelationsbetween the red and all green foci were calculated using a multi-τ-scheme. Data was evaulated with QuickFit3.0, as described in the main paper (linked fit, assuming DA = DB = DAB and focus parameters matching thesimulation parameters). From these fits we report the diffusion coefficient, the absolute concentrations and therelative concentration

p′AB =cAB

cA + cBcAB. (S28)

paramter valuegreen lightsheet width wLS,g = 1200 nmred lightsheet width wLS,r = 1300 nmpixel size a = 400 nmgreen MDEheight zg = 2000 nmgreen focus height from fita zg,fit = 1029 nmgreen MDEwidth wg = 500 nmred MDEheight zr = 2000 nmred focus height from fita zg,fit = 1029 nmred MDEwidth wr = 600 nmgreen→red spectral crosstalk κgr = 3.5%

κgr = 11.2%

green→red focus separation δx ∈ {0,25,50,75,100,200,300,400,600,1000,2000} nmdiameter of simulational sphere 14 µm

diffusion coefficient Dsim = 30 µm2/ssimulation duration Tsim = 10 ssimulation timestep ∆tsim = 10−5 sminimum lagtime for correlation τmin = 2 ·∆tsimconcentrationsb for:

p′AB = 0 cA = cB = 0.5 nMcAB = 0 nM

p′AB = 0.125 cA = 0.5 nMcB = 0.2 nMcAB = 0.1 nM

p′AB = 0.31 cA = 0.2 nMcB = 0.5 nMcAB = 0.3 nM

p′AB = 0.5 cA = cB = 0.25 nMcAB = 0.5 nM

p′AB = 0.78 cA = cB = 0.1 nMcAB = 0.7 nM

p′AB = 1 cA = cB = 0 nMcAB = 1 nM

a measured from a Gaussian fit to the product of the MDE and the lightsheetb c = 1 nM leads to n = 865 particles in the simulational box

Table ST7. SPIM-FCCS Simulation parameters

S7.2. Test of the SPIM-FCCS models and fitting routines

For the shift δx = 0 nm, we checked whether the correct relative concentrations could be measured at all. Fig. S16shows the results, which are in good agreement with the simulation parameters.

Fig. S16. Result of SPIM-FCCS simulations with different relative concentrations p′AB ≡ cAB/cALL. (A) measured relativeconcentration vs. the used simulation parameter for κgr = 3.5% and (B) for κgr = 11.2%, (C) measured diffusion coefficient(D = (34.9±1.5) µm2/s).

S7.3. Crosscorrelation amplitude error in dependence of alignment accuracy

When moving the red and green detection focus apart by a distance δx in lateral direction, we get a slowlydecreasing correlation amplitude. Fig. S17 shows example correlation curves from the simulation for differentseparations and relative concentrations. For small separations δx < a, the crosscorrelation function still hasapproximately the form of the unshifted case. For larger displacement (see especially δx = 1000 nm in Fig. S17),a distinct peak is visible in the crosscorrelation function. To test the influence of different displacements ofthe real foci on the fit results (assuming non-shifted foci), we fitted all these curves with an assumed shift ofδx = δy = 0δ z = 0 nm. The results are shown in Fig. S18. The relative error of the relative concentration

|p′AB(δx)− p′AB(δx = 0)|p′AB(δx = 0)

(S29)

stays below 5% (orange line in Fig. S18) for shifts of up to ∆x = 200 nm, so an alignment accuracy of better than100 nm for our SPIM setups is enough to obtain accurate results from the fits.

Fig. S17. Examples of ACFs (green, red, according to detection channel) and CCFs (blue) for the relative concentrationsp′AB = 1 and p

′AB = 0.5 (κgr=11.2%) and different displacements δx of the red and green foci. The top row depicts the two

MDEs.

Fig. S18. Fit results when assuming no shift in the fit model, when a shift is present in the microscope setup. (A) relative errorof the measured relative concentration

S8. Data Evaluation

S8.1. Background Correction

First the camera offset and any other static background signal is corrected by subtracting the static per-pixelbackground signal:

I(t;~r) = Ĩ(t;~r)−B(~r) with B(~r) = 1NB

NB−1

∑n=0

B(n∆̇t;~r)

where ∆t is the frame time of the acquisition and NB is the number of background frames acquired before themeasurement.

S8.2. Bleach Correction

If bleaching was visible in the image series, it was corrected after the background correction and before thecorrelation, as described in Ref. [16, 19]. If the bleaching rates are not too high, the fluorescence decay in eachpixel can be described by a simple exponential function

f (t) = f0 · exp(−t/τB). (S30)

For longer measurements (typically > 75000 frames), the mono-exponential decay is not sufficient, so we are usinga heuristically modified function to describe the decay:

f (t) = f0 · exp(− t + f2t

2

τB

). (S31)

f (t) = f0 · exp(− t + f2t

2 + f3t3

τB

). (S32)

This function is related (as an early cut-off) to the cumulant analysis often used to dynamic light scattering toanalyze poly-exponential decay curves [24]. We fit one of the above functions (free parameter f0, f2, f3,τB) to asubset of the intensity time series Irawγ (t,~r) in each pixel~r of each channel γ . For the fitted subset, equally distributedblocks of 50−80 consecutive intensity samples are averaged to yield one estimate of the count rate each. Then theintensity is pixel-wise corrected with the operation:

Iγ(t,~r) =Irawγ (t,~r)√f (t)/ f (0)

+ f (0) ·(1−√

f (t)/ f (0))

(S33)

0 25 50 75 100 125 150 175 200

time t [seconds]

0

500

1000

1.5⨯103

2⨯103

2.5⨯103

3⨯103

inte

nsity

I(

t) [A

DU

]

0 25 50 75 100 125 150 175 200

time t [seconds]

500

1000

1.5⨯103

2⨯103

2.5⨯103

inte

nsi

ty

I(t)

[A

DU

]

0

(F) (G)

(C) (D)

0 25 50 75 100 125 150 175 200

time t [seconds]

0

500

1000

1.5⨯103

2⨯103

2.5⨯103

3⨯103

inte

nsi

ty

I(t)

[A

DU

]

0 25 50 75 100 125 150 175 200

time t [seconds]

0

500

1000

1.5⨯103

2⨯103

2.5⨯103

3⨯103

inte

nsi

ty

I(t)

[A

DU

]

intensity before bleach correction

intensity after bleach correction

f(t)

= A

·ex

p(-t/t)

Bf(t)

= A

·ex

p(-

(t+f·t²)

/t)

2B

0 25 50 75 100 125 150 175 200

time t [seconds]

0

500

1000

1.5⨯103

2⨯103

2.5⨯103

3⨯103

inte

nsi

ty

I(t)

[A

DU

]

intensity after bleach correction resulting autocorrelation curves(A)

(H)

(E)

(B)

10lag time t [seconds]

0.1-3

10 0.01 1

10lag time t [seconds]

0.1-310 0.01 1

10lag time t [seconds]

0.1-3

10 0.01 1

corr

ela

tion f

un

ctio

n g

(t)

-210

0

corr

ela

tion f

unctio

n g

(t)

-310

0

corr

ela

tion funct

ion g

(t)

-47·10

0

no

co

rre

cti

on

Fig. S19. Example result of our bleach correction from the eGFP-mRFP1-fusion cell shown in the main text (with 2×2-binning). (A,C,F) show the non-corrected intensity time traces for one pixel (green/red curve for green/red color channel)with bleach correction fits (black curves) for the simple exponential model Eq. (S30) in (C) and the modified model Eq. (S31)in (F). (A,D,G) show the intensity time traces after correction, (B,E,H) show the autocorrelation functions obtained after thebleach correction (red line is the average and the error polygons show the standard deviation over 5 consecutive segments of∼ 10.5 s length each).

It can be shown that with this transformation the corrected signal has a mean and a variance of f (0), which isexpected if the intensity scales linearly with the number of particles in the observation volume and the number ofparticles is distributed according to a Poissonian distribution. In principle this transformation can also be used witha more complex model f (t).

Fig. S19 compares the effect of different bleach corrections. The first row shows the count rate and correlationfunction (CF) with no bleach correction. In the second row, model Eq. (S30) was applied and in the third rowmodel Eq. (S31). The CFs are shown as average and standard deviatio over 5 consecutive runs. The maximum CFamplitude descreases with improving bleach correction and the prominent offset at higher lag times vanishes. Alsothe standard deviation of the run CFs is reduced, as different runs do no longer differ in visible particle number orabsolute intensity.

We also checked this bleach correction method using the FCS simulation tool described above. A set of randomwalk trajectories were created. In addition the walkers were sucessively removed from the simulation with a pertimestep bleaching probability pbleach for each walker. This leads to an exponentially decaying particle numberin the simulation box, resembling the situation in our samples. The simulation used the parameters shown inTab. ST8. Fig. S20 summarizes the results. The diffusion coefficient obtained from the uncorrected curve is (23.1±0.3) µm2/s and from the bleach-corrected curves we got (22.8± 0.3) µm2/s. Thus no systematic deviation ofthe mobility parameters obtained from a fit to a bleach-corrected autocorrelation curves could be detected, whenbleaching up to at least ∼ 50% of the initially available particles.

paramter valuegreen light sheet width wLS,g = 1200 nmpixel size a = 400 nmgreen MDEheight zg = 2000 nmgreen focus height from fita zg,fit = 1029 nmgreen MDEwidth wg = 500 nm

diameter of simulational sphere 12 µm

walker concentrationb c = 1 nMdiffusion coefficient Dsim = 30 µm2/ssimulation duration Tsim ≈ 55 ssimulation time step ∆tsim = 1 ·10−5 sminimum lag time for correlation τmin = 2 ·∆tsimbleach probabilities (per simulation 10 µs step): pbleach =

{10−7,3 ·10−7

}a measured from a Gaussian fit to the product of the MDE and the lightsheet

b c = 1 nM leads to n = 545 particles in the simulational box

Table ST8. SPIM-FCCS Simulation parameters

ACFs

10-4

0.001 0.01 0.1 1

lag time τ [seconds]

0

0.5

1

corr

ela

tion functi

on g(τ)

no bleach

rbleach

=0.005s-1

countrates

0 10 20 30 40 50

time t [seconds]

0

2

4

6

8

10in

tensit

y I(t

) [k

cps]

no bleaching

rbleach

=0.005s-1

fit: f(t)=7.68·exp(-t/62.01s)

ACFs

10-4

0.001 0.01 0.1 1

lag time τ [seconds]

0

0.5

1

corr

ela

tion functi

on g(τ)

no bleaching

rbleach

=0.005s-1

countrates

0 10 20 30 40 50

time t [seconds]

0

2.5

5

7.5

10

12.5

15

17.5

20

inte

nsit

y I(t)[A

.U.

or

kH

z]

no bleaching

rbleach

=0.005s-1

(A) (B)

(C) (D)

Fig. S20. SPIM-FCS simulations with reservoir depletion (bleaching). (A) shows countrate curves and (B) the correspondingautocorrelation curves without bleach correction. (C) shows the bleach-corected countrates and (D) again the correspondingautocorrelation curves. Note: The shown autocorrelation curves have a similar noise signature, as they are all created from thesame set of particle trajectories.

S8.3. Correlation & Fitting

Finally the full frame is split into two halves, representing the two color channels Ig(t;~r) and Ir(t;~r). Subsequentlythe auto- and crosscorrelation functions are calculated using a multi-τ software correlator. The resulting sets{(τi, ĝγρ,i)} of estimates of the correlation function for semi-logarithmically spaced lag-times τi, can be used toestimate the parameter ~βof a theoretical model function gγρ(τ;~β ) using a least-squares fit:

~β ∗ = argmin~β

∑i

[ĝγρ,i−gγρ(τi;~β )

σγρ,i

]2, (S34)

where ~β ∗ is the optimal set of model parameters. The weighting factors σγρ,i allow to incorporate knowledge aboutthe accuracy of the estimates {(τi, ĝγρ,i)}. They can either be calculated theoretically [25,26], or be estimated fromthe measurement: For each pixel we split the complete timeseries into 3− 10 segments of equal length that eachyield a correlation function. The models are then fit to the per-lag average of these functions. The correspondingper-lag standard deviations can be used as weighting factor σγρ,i. In [27] more advanced methods have beenproposed that allow to estimate also the off-diagonal elements of the covariance-matrix.

S8.4. Global Model Fitting

In the main text we derived model functions ggg(τ;cA,cAB,DA,DAB, ...), grr(τ;cA,cB,cAB,DA,DB,DAB, ...)and ggr(τ;cA,cB,cAB,DA,DB,DAB, ...) for SPIM-FCCS that are linked through a set of global parameters

(e.g. the concentrations). Therefore these functions should not be fitted to the measurements (τi, ĝgg,i, ĝgr,i, ...)independently. To extract the global parameters, we choose a global fitting method that simultaneously minimizesthe least-squares deviations of the fit functions from the measurements. The optimal parameter vector ~πγ(~β ) is thenthe solution of this least-squares optimization problem:

~β ∗ = argmin~β

∑γρ={gg,rr,gr,rg}

∑i

[ĝγρ,i−gγρ

(τi;~πγρ(~β )

)σ̂γρ,i

]2, (S35)

where ~πγρ(~β ) maps the ”global” parameter vector ~β to the ”local” vector containing only the parametersused by the model function gγρ(·; ·). To solve this optimization problems we used either a version of theLevenberg-Marquardt algorithm [28,29], implemented in the software library lmfit [30] or our own implementationof the simulated annealing stochastic optimizer [31].

S8.5. Performance

task typical processing timebackground correction + correlation, no binning 50 s∗

background correction & bleach correction + correlation, no binning 120 s∗

background correction & bleach correction + correlation, 2× binning 40 s∗single ACF or CCF 1-component model fits 30 s for 1280 fits (∼ 40−50 fits/s)FCCS global fits, 1-component (DA = DB = DAB linked) 190 s for 1280 fits (∼ 7 fits/s)FCCS global fits, 2-component (D1,D2 per channel) 500 s for 1280 fits (∼ 2.5 fits/s)

∗ About 20−30% of the time required for calculating the correlation functions is used to read and decode the TIFF files, 30% for backgroundand bleach correction, as well as estimating several statistics. The remaining 40−50% are used for correlation.

Table ST9. Typical time consumption/performance for data evaluation a time series of 128× 20 pixels and 100,000frames. All data was evaluated on an AMD Phenom II X6 1090T, 3 GHz, 16 GB RAM running Linux. QuickFit 3.0 wasoptimized for this architecture during compilation. A 64− bit built was used. Processing times on Windows 7 (64-bitbuild of QuickFit 3.0 from our homepage) are typically 20−30% slower on a comparable computer. A single thread isused for one tasks, but several measurements can be evaluated in parallel.

S8.6. Calibration of the measured concentration

As already discussed in Refs. [1, 32], absolute concentrations cannot be measured with camera-basedSPIM-FCS/FCCS. But as we already showed in Ref. [1], the dependence between the true concentration andthe measured concentration is linear, so a calibration factor can be derived, if the true concentration is known.To calibrate the concentrations in SPIM-FCCS a modified procedure can be used. We prepared a 170 bp dsDNAsample, containing a mixture of Alexa-488 and Alexa-594 single-labeled, as well as double-labeled molecules.Then a concentration series of this sample was measured using SPIM-FCCS and a confocal FCCS setup. Bothmeasurements were evaluated with a global fitting model (DA = DB = DAB), as described in the main text. Thena calibration factor can be calculated by a (robust) regression analysis. Fig. S21 shows the measured data as wellas the resulting regression coefficients. Note that this calibration depends on the camera properties and should beredone for any new image sensor (or set of camera settings) and SPIM-FCCS setup.

confocal vs. SPIM FCCS

0 5 10 15 20 25

confocal FCCS concentration cc

[nM]

0

50

100

150

200

250

SPIM

-FC

CS concentr

ati

on c

SPIM

[nM

]

species A

species B

species AB

fit: cSPIM

(cc) = 9.22 · c

c, R

2=0.996

fit: cSPIM

(cc) = 8.82 · c

c, R

2=0.983

fit: cSPIM

(cc) = 11.16 · c

c, R

2=0.971

Fig. S21. calibration of the SPIM-FCCS concentration against concentrations measured in a confocal microscope.Concentrations measured in a dilution series of a 170 bp dsDNA sample. Species A is Alexa-488 single-labeled, species Bis Alexa-594 single-labeled and species AB is double-labeled. Datapoints (circles) and errorbars are averages and standarddviations over fits to 6 runs (20 s each) in the confocal case and all pixels in 2− 4 measurements in the SPIM-FCCS case(700,000 frames, 330 µs frametime). A global fit with DA = DB = DAB was used in both cases. Linear functions (dotted lines)were fit with robust regression and forced to cSPIM(0) = 0.

Details on the two lightsheet microscopes SPIM-1 and SPIM-2Optical setup of the SPIMsCamera PropertiesSample Mounting

SPIM-FCCS alignment and calibrationAlignment procedureBead ScansPSF calibrationStability of the setupsVolume overlap: objective scan

Sample preparationCell culture protocols: FuGENECell culture protocols: Neon transfectionSmall and giant unilamellar vesicle

Additional SPIM-FC(C)S example dataSPIM-FCS example measurements in buffer, lipid and live cellsAdditional data from 607bp DNA measurementSPIM raw data and overview imagesEGFR/PMT in CHO cell membranes

Confocal FCCS measurementsConfocal FCS measurements 1 (Heidelberg)Confocal FCS measurements 2 (Singapore)Sample preparation for confocal microscopesConfocal volume calibrationConfocal data Evaluation

Derivation of the SPIM-FCCS Correlation FunctionsNotationFCCS Correlation FunctionsSPIM molecular detection functionsSPIM correlation functionsConfocal Microscopy correlation functionsFCCS correlation functions with different mobility modesMultiple componentsBackground CorrectionTriplet and other Blinking Dynamics

Testing SPIM-FCCS with simulationsSimulation codeTest of the SPIM-FCCS models and fitting routinesCrosscorrelation amplitude error in dependence of alignment accuracy

Data EvaluationBackground CorrectionBleach CorrectionCorrelation & FittingGlobal Model FittingPerformanceCalibration of the measured concentration