In our experimental conditions, we are? determining """"""""colocalization"""""""" in the resolution limit of light microscopy. In this PPG, the """"""""colocalization"""""""" term? will imply that two molecules may share the same microenvironment and/or can be elements of the same set? of macromolecular complexes which are resolved by the optical system. Forexample, two proteins forming part? of two different molecular complexes may show positive """"""""colocalization"""""""" tests due to the lack of the optical resolution? to detect each molecular complex. Still, the positive test will indicate that the two proteins are located proximally? within the optical resolution. Thus, in """"""""colocalization"""""""" determinations it is critical to achieve the maximal x,y,z? resolution. In this PPG, we will use developed methodologies in the laboratory to achieve the highest resolution? attainable with optical microscopy. We have maximized the optical linearity and the quantity of light recorded by the? detectors in conjunction with image restoration analysis to recover the light lost by the optical imperfections of the? system. This methodology has allowed us to reach a resolution in the x,y plane of about 100-200 nm (see Figs. 1-3).? Definition of """"""""colocalization"""""""" and random """"""""colocalization"""""""". In a cell labeled with two fluorophores, one red (R)? and one green (G), the number of voxels above threshold that are labeled with the red fluorophore is nRand the? number labeled with the green fluorophore is nG. The number of colocalized voxels, i.e. those voxels containing? intensity signals above threshold from both fluorophores is ncoioc. The percentage of """"""""colocalization"""""""" of G with R is? given by 100*^^- (eq. 1); and the percentage of """"""""colocalization"""""""" of R with G is given by 100*^^ (eq. 2). In? n* nc? these measurements we will estimate the probability that the measured """"""""colocalization"""""""" could occur by chance by? calculating % random """"""""colocalization"""""""" from ((area_1x area_2)/total areaA2) x100, where area_1 and area_2 are the? areas above threshold (eq. 3).? The """"""""colocalization"""""""" method is generally based on the user ability to set the intensity threshold, and has the major? draw back of the lack of a mathematical model to adjust the threshold of both images that will define the degree of? paired pixel overlap. Most importantly the pixel overlap method has the limitation of being a binary test that? determines whether the two paired pixels in two images have or not intensities above the intensity threshold and does? not consider whether the two stained proteins have a landscape of intensity staining with a high degree of correlation,? as it would be expected if they are elements of a common complex. The application of a correlation measurement? was recently described by Li et al. (2004)7.? In the following sections, I will discuss and apply developed algorithms to quantify the degree of protein-protein? association by two methods, the INTENSITY CORRELATION ANALYSIS and the INTENSITY THRESHOLD? """"""""COLOCALIZATION""""""""ANALYSIS.? Intensity correlation analysis. In the present analysis, we acquire high resolution images to compare the? correlation of pixel intensities in equivalent x,y coordinates of paired images from cells double stained for two different? proteins. The prediction is that if two proteins are elements of the same macromolecular complex, the intensity? staining landscape of the two images should have a x,y pixel to pixel positive correlation. On the contrary, if the two? proteins are localized in distinct compartments the result will be a negative correlation. Finally, if the proteins in the? two images are labeled in a diffuse non structured pattern (random), the correlation will tend to 0.? This method is based on the principle that for any set of values the sum of the differences from the mean equal zero,? i.e., ?N(A-a)=0, where a is the mean of the distribution with N values of Al. In the experiment N is the number of? pixels, and A is the intensity for each pixel. If we have two set of values in two arrays 1 and 2 with N pixels per array? having a random distribution of intensities A-,and 6, for arrays 1 and 2 , the sum of the product of their differences will? also tend to zero, thus IN(A-a)(S,-Jb)~0. On the other hand, if the two intensities are positively correlated, the product? will tend to be a positive value (^(Ara)(Brb}>G) and if they are negatively correlated the product will tend to a? negative value (LN(Ara)(Brb)<0). To perform the analysis, we generated two intensity arrays 1 and 2 of equivalent? x,y coordinates of the paired images. The pixel intensity values of the arrays were normalized to their individual? maximum and a 3rd array was obtained from (Ara)(Brb). Rows with 0 values with the same x,y coordinates in both? PHS 398/2590 (Rev. 05/01) Page 382 Continuation Format Page? CONTINUATION PAGE Principal Investigator/Program Director (Last, first, middle): Ping, Peipei (Stefani, Heart Biology Core)? images were eliminated. We generated the correlation plot between the first two arrays and between array 3, and? arrays 1 and 2. As indicated, positive, negative or 0 values of EN(A-a)(S,-6) would be an indication of positive,? negative, and 0 correlation. To evaluate the statistical? significance of the data we used the non-parametric? Psign test by counting the number of positive N+ and? negative N .numbers of the 3rd array (Ara)(Brb). We? generated the intensity correlation quotient (ICQ) from? N+/(N+-N.)-0.5. Values of ICQ are 0> for positive? correlation, <0 for negative correlation and - 0 for? random correlation. The correlation coefficient r? (COR) between arrays 1 and 2 was calculated from? r=2xy/(NSxSy) wher r = correlation coefficient, xy =? product of deviation scores, N = sample size, Sx =? standard deviation of X (intensities in first image), and? Sy = standard deviation of Y (intensities in second? image).? We initially performed intensity correlation analysis? using as an experimental """"""""colocalization"""""""" model the? same protein tagged by two different antibodies at? different sites. Confocal images were at 0.038? urn/pixel in the x, y axis, and every 0.1 urn in the z? plane. We used as the target protein caveolins (1-3).? Caveolin 3 (CAV-3) was immunostained with anti-? CAV3 antibody and all the caveolins with a generic? anti-caveolin antibody (anti-CAVg). Since in the heart,? CAV3 is the most abundant caveolin, anti-CAVg? should mainly stain CAV3. As a model for non-? """"""""colocalization"""""""" (segregated localization), we? displaced by 4 pixels the x,y plane of the anti-CAVg? labeled image. In Figure 6, panels A,B are single? confocal sections of cardiomyocytes immunostained? for CAV3 and caveolin generic CAVg, respectively.? Panels C and D are the regions marked with squares? in (A) and (B) at higher magnification with the? corresponding overlay in E. Note the strikingly similar Fig. 6. Intensity correlation analysis using """"""""colocalization"""""""" and non? pattern of pixel intensity distribution for CAV3 (C) and """"""""colocalization"""""""" experimental models. (A,B) Single confocal sections? CAVg (D) that generates the yellow signal in E. In of cardiomyocytes immunostained for CAV3 (A) and caveolin generic? CAVg (B). (C, D) Regions depicted in (A) and (B) at higher? contrast, when the region D was displaced by 4 pixels? magnification with their overlay (E). (F) overlay of (C) and (D) but with? in the x,y plane, the overlay shows a clear separation? the region D displaced 4 pixels in the x,y plane. (E1-E3) Intensity? between green and red signals (F). correlation analysis for (C, D). (E1) Correlation plot of CAV-g vs.? CAV3. (E2.E3) Plots of CAV-g and CAV3 vs. (A,-a)(Brb). (F1-F3)? In the same figure, graphs E1-E3 illustrate the? Intensity correlation analysis of F. (D) (CAVg) was displaced by 4? intensity correlation analysis for (C) and (D). E1 is the pixels in the x,y plane; the analysis shows a negative correlation. (G1-? correlation plot of CAV-g vs. CAV3 pixel intensities. G3) Intensity correlation analysis of two arrays with random numbers.? The points are not randomly distributed (compare with? G1 for a random plot) with a correlation coefficient of 0.84. E2 and E3 are the plots of CAV-g and CAV3 vs. (Ara)(Br? b). As expected for a positive correlation, the majority of the (A,-a)(Brb) have positive values (red dotted line marks? the zero value of the (Ara)(Brb) axis). The ICQ was 0.36 with Psign test<0.001. We can conclude from these data? analysis that the pixel intensities are highly correlated as expected from the dual labeling of the same protein. Graphs? F1-F3 show an equivalent analysis for the pair of images generating the F overlay. As indicated, in this case the? region D (CAVg) was displaced by 4 pixels (0.038 um/pixel) in the x,y plane. As result of the displacement, the? analysis should give a negative correlation. In fact, the correlation coefficient of the plot in F1 was -0.24 and ICQ -0.15? with PSign <0.005. Graphs G1-G3 show the results of a random distribution. In this case, the data arrays were made? with random numbers. G1 shows the random distribution of pixel intensity in the correlation and G2, G3 quasi equal? number of positive and negative (Ara)(B,-b) values.? PHS 398/2590 (Rev. 05/01) Continuation Format Page? CONTINUATION PAGE Principal Investigator/Program Director (Last, first, middle): Ping, Peipei (Stefani, Heart Biology Core)? Intensity threshold """"""""colocalization"""""""". The most difficult task to measure """"""""colocalization"""""""" by changing the intensity? threshold in both images is to establish criteria of setting the intensity threshold values which define the areas above? threshold to be compared. The intensity threshold should eliminate non-specific signals as detector noise, non-? specific antibody binding, autofluorescence, and blur? from optical imperfections. In """"""""cotocalization"""""""" studies.? 100 D CAVg OVER CAV3? the intensity threshold is typically set in an arbitrary? manner bv the user and """"""""colocalization""""""""? measurements are questionable. To overcome this? caveat, we have developed mathematical criteria to? set the threshold levels. Most importantly, in the? same set of paired images, curves are constructed to u? measure """"""""colocalization"""""""" level as function of the 3? intensity thresholds selected. We found convenient to? plot % """"""""colocalization"""""""" level vs. random % 8? OCAVgOVERCAV3? """"""""colocalization"""""""" (calculated from eq. 2 taking into CAV3OVERCAVg? account the total area and the two sets of thresholded (4 Pixel shift)? areas), using various threshold levels in paired? images. In this manner, we can have an estimate of 100.00 10.00 1.00 0.10? the level of contamination random """"""""colocalization"""""""". in % RANDOM COLOCALIZATION? the """"""""colocalization"""""""" measurements, (see below). We? have calculated intensity threshold values in two Fig. 7. Threshold intensity """"""""colocalization"""""""" analysis as function oi? different manners: METHOD 1. Varying the intensity threshold intensity. % """"""""colocalization""""""""vs. % random """"""""cotocalization"""""""" is? threshold as a function of the average intensity. plotted for random """"""""colocalization"""""""" (r), CAVg over CAV3 (a),and? CAV3 over CAVg (b) (for panels C,D, Fig. 6), and CAVg over CAV3? independently in both images, and in a sequential? (c), and CAV3 over CAVg (d) (overlapped images, Fig. 6F). The? manner, and METHOD 2. Adjusting the threshold? CAVg image in panel F was 4 pixel x,y shifted. Note the high CAVg? intensities of both images to attain the same area over CAV3 of ca. 85% at 1 % random """"""""colocalization"""""""". The 4 units x,y? (number of pixels) above threshold in both images. In pixel shift dramatically reduced the """"""""colocalization"""""""" level.? METHOD 1 the threshold in each image is first Calibration=0.038 urn/pixel.? calculated from the average intensity of all the pixels in each image, thereafter the same procedure is applied in the? resulting images and so on. In this manner, a set of images are generated where the threshold values are the? average intensities of the respective previous images. This method has the advantage that all the pixel intensity? values are considered to calculate the threshold values, reducing asymmetries and differences of the total point? histogram between both images. In METHOD 2, intensity threshold values are calculated in each image with the? constrain that the number of pixels above threshold (thresholded area) is the same in both images and areas in? decrementing steps from 100% to ca. 1% of the total area are calculated. In a given area having an identical value? for both images, one subset of pixels in the same x, y coordinates will have intensities in both images while a second? set of pixels will have intensity values only in one image. The """"""""percent colocalization"""""""" is calculated from the ratio of? the number of pixels with intensities (>0)in both images with the number of pixels with intensities (>0)in one image.? In this algorithm, the colocalization level of Image A over Image and of Image A over A is the same. Minor? differences can arise from the computational limitations to define the right threshold to obtain two identical areas in? both images. As in METHOD 1,""""""""%colocalization"""""""" will be plotted vs.""""""""% random colocalization"""""""". These two methods? are testing pixel overlap in a binary manner at various threshold intensities and thresholded areas. Therefore, this? analysis is also taking into consideration the degree of correlation of the landscape of intensity staining in both? images.? Figure 7 illustrates METHOD 1 by evaluating the """"""""colocalization"""""""" level as function of the threshold intensity? determined sequentially by the average intensity of the individual images. The graph illustrates the relationship? between percent """"""""colocalization"""""""" and percent random """"""""colocalization"""""""". As previously discussed, a positive? """"""""colocalization"""""""" has to be much higher than the random """"""""colocalization"""""""". Plots (a and b) are the CAVg over CAV3 and? CAV3 over CAVg """"""""colocalization"""""""" levels, respectively. Plot (r) is the theoretical percent random """"""""colocalization"""""""". It is? clear that at 1% random """"""""colocalization"""""""", the percent """"""""colocalization"""""""" of CAVg over CAV3, and CAV3 over CAVg is? much higher being 85% and 64%,respectively. Plots (c and d) correspond to panel F of Fig.6, where 4 pixel units? (0.038 urn/pixel) were shifted in the x,y plane. As for the intensity correlation analysis, the pixel shift practically? eliminated the """"""""colocalization"""""""" level between CAVg over CAV3 and vice versa. Identical results were obtained by? using METHOD 2 (not shown). Equivalent results are illustrated with the three methods in Fig. 16, showing their? application to investigate '""""""""'colocalization'"""""""" of cytochrome C-GFP fusion protein expressed in cardiomyocytes via? adenovirus and a mitochondria marker (mitotracker TMRE).? PHS 398/2590 (Rev.05/01) Continuation Format Page? CONTINUATION PAGE Principal Investigator/Program Director (Last, first, middle): Ping, Peipei (Stefan!, Heart Biology Core)? Figure 8 shows another example of """"""""colocalization"""""""" measurement varying the threshold as a function of the average? intensity (METHOD 1) by comparing the """"""""colocalization"""""""" level of a transfected labeled protein with a native? immunolabeled protein. Confocal images were at 0.054 urn/pixel in the x, y axis, and every 0.1 urn in the z plane.? Panels A, B and C are single confocal sections of a cultured neonatal rabbit myocyte transfected with p38-GFP fusion? protein (green) and immunostained with anti-TAB1 antibody (red). Regions 1 and 2 are displayed at a higher gain in? D,E and H,l, respectively. Note the distribution outline in defined clusters of TAB1 and P38 which has a filamentous? pattern in the cytoplasm. Graphs (G) and (K) show the relationship between percent """"""""colocalization"""""""" and percent? random """"""""colocalization"""""""" for regions 1 and 2, respectively.? In the graph, each data point was obtained by? sequentially setting the threshold as function of the? average intensity. The percent """"""""colocalization"""""""" and? percent random """"""""colocalization""""""""was calculated from eqs.? 1-3. The theoretical curve of random """"""""colocalization""""""""was? plotted as a reference (labeled RANDOM). With a .RcgioncV c ? Region 1 I |_| Region 2 | Region2? minimal threshold value, the area above threshold equals? the total area of the region; thus, the values of percent? """"""""colocalization"""""""" and percent random """"""""colocalization"""""""" are? 100. As the intensity threshold values are increased as a? function of the average intensity, the thresholded areas? are reduced resulting in diminished """"""""colocalization"""""""". The? arrow marks the """"""""colocalization"""""""" level when the percent? random """"""""colocalization"""""""" is 3% (see table at the right? bottom). Panels F and J show the corresponding masked XUM1K00DOM1C0OLOC1A.U0ZATO0H.1? DCAV3 OVER UVg? thresholded areas (yellow shows the overlap). The ? NCX OVER CAVg COLOCALIZATION ANALYSIS (Arrow in G and K)? """"""""colocalization"""""""" level (40-50%) is much higher than the CYTOPLASM (Region 1) NUCLEUS(Region Z)? Total Area = 131 rn 2 Total Area = 104 (im2? expected random """"""""colocalization""""""""for those areas. These P38Area =25 prn2 P38Area =20 ym2? TAB1Area =21 pm2 TAB1 Area = 20 Mm2? determinations indicate a high level of """"""""colocalization"""""""" in P3B Over TAB1 =41% P38 Over TAB1= 50%? the nucleus and cytoplasm for TAB1 and P38. Similar <-> 100%.0RAN1D0.O0 M CO1L.0OCALIZ0A.1TION 0.01 TRAanBd1omOvCeorloPc3.8==35.0% RTaAndBo1mOCvoelor cP.3=8=3.502%? values were obtained using METHOD 2. Panel L shows Fig. 8. """"""""Colocalization"""""""" measurements in neonatal heart cells? examples of minimal and maximal """"""""colocalization""""""""in adult transfected with p38GFP (green) and immunolabeled with anti-? heart myocytes. Minimal """"""""colocalization"""""""" was observed TAB19 antibody (red} (seetexO. _? between caveolin generic (CAVg) and Na/Ca2+ exchanger (NCX). Note that the curve for NCX over CAVg follows? closely the theoretical curve for random """"""""colocalization"""""""". In contrast, maximum """"""""colocalization"""""""" was obtained when? the same protein (caveolin) was immunostained with monoclonal (CAV3) and polvclonal (CAVg) antibodies. In the? latter condition, one should expect 100%""""""""colocalization"""""""" if the antibodies would have 100%efficiency to stain the? target protein. Evidently, this is not the case, and only a fraction of caveolin can be simultaneously stained by both? antibodies.? Evaluation of intensity correlation and intensity threshold analyses. The major advancementin our studies? is the achievement of high spatial x.v.z resolution bv image restoration in conjunction with the use of? analytical user-independent colocalization methods (intensity correlation analysis and intensity threshold? colocalization). These methods will provide a strong argument to define whether two proteins are elements or not of? the same macromolecular complex, and,in particular for this PPG, to quantify with rigorous algorithms protein-protein? """"""""colocalization"""""""" and association of proteins with mitochondria. Both """"""""colocalization"""""""" methods, intensity correlation? analysis and intensity threshold """"""""colocalization"""""""" have advantages and drawbacks. The intensity correlation is a? powerful statistical method that evaluates the correlation between the landscapes of paired images. One draw back? is that not clearly interpretable results are obtained if the images have regions of positive and negative correlation. To? circumvent this deficiency the user should evaluate different cytoplasmic and membrane regions. Another drawback? of this analysis is that it does not give the """"""""colocalization"""""""" level as the overlap factor as in the intensity threshold? analysis. On the other hand, the major draw back of the intensity threshold method is the mathematical definition of? random """"""""colocalization"""""""" due to the lack of a valid estimate of the total area where the protein can be distributed? randomly. This is not the case for cytosolic proteins that can randomly distribute in the cytoplasm and the total area is? directly measured from the region of interest. In general, proteins are targeted to regions making very difficult to? measure the surface of the targeted area. For example, the total area that one has to consider for membrane proteins? is difficult to estimate. Nevertheless, the fact that when positive, """"""""colocalization"""""""" gives a """"""""co/oca//zaton""""""""/tandom? """"""""colocalization"""""""" ratio of several orders of magnitude higher which makes it easy to define the """"""""colocalization"""""""" level. In? fact, if """"""""colocalization"""""""" remains significant in the limit of random """"""""colocalization"""""""" (-0.1%), it is a strong argument of a? PHS 398/2590 (Rev. 05/01} Continuation Format Page? CONTINUATION PAGE Principal Investigator/Program Director (Last,first,middle): Ping, Peipei (Stefani, Heart Biology Core)? positive finding. Moreover, as control. I have been able to detect lack of """"""""colocalization"""""""" of two membrane proteins? that do not co-immunoprecipitate such as NCX and caveolin as shown in Fig.8L. In view of these considerations,? one of the major advantages of the intensity correlation analysis is that it is not dependent on the estimate of random? """"""""colocalization"""""""". Thus, by using both types of analysis. I am confident that reliable """"""""colocalization"""""""" measurements will? be obtained in the limit of the microscope spatial resolution. The implementation of the intensity correlation analysis is? very laborious and time consuming. For example, the analysis of 1024x1024 pixel image takes about 6 hrs.At? present, I am developing a program to increase the processing speed and to automatically analyze various? predetermined regions in the image. Alternatively, I am also developing a program that will determine the degree of? correlation from the x and y first derivative of the landscape. This approach may increase the sensitivity and? discrimination of the analysis, since it will take also into account the slopes of the valleys in the landscape.?
Showing the most recent 10 out of 82 publications
