The Golgi complex consists of serially stacked membrane cisternae which can be further categorized into sub-Golgi regions, including the cis -Golgi, medial-Golgi, trans -Golgi and trans -Golgi network. Cellular functions of the Golgi are determined by the characteristic distribution of its resident proteins. The spatial resolution of conventional light microscopy is too low to resolve sub-Golgi structure or cisternae. Thus, the immuno-gold electron microscopy is a method of choice to localize a protein at the sub-Golgi level. However, the technique and instrument are beyond the capability of most cell biology labs. We describe here our recently developed super-resolution method called Golgi protein localization by imaging centers of mass (GLIM) to systematically and quantitatively localize a Golgi protein. GLIM is based on standard fluorescence labeling protocols and conventional wide-field or confocal microscopes. It involves the calibration of chromatic-shift aberration of the microscopic system, the image acquisition and the post-acquisition analysis. The sub-Golgi localization of a test protein is quantitatively expressed as the localization quotient. There are four main advantages of GLIM; it is rapid, based on conventional methods and tools, the localization result is quantitative, and it affords ~ 30 nm practical resolution along the Golgi axis. Here we describe the detailed protocol of GLIM to localize a test Golgi protein.

The Golgi complex plays essential roles in secretory/endocytic trafficking of proteins and lipids (hereafter cargos) in mammalian cells ^{1 , 2 , 3} . At the Golgi, cargos are not only sorted to various sub-cellular compartments but also modified by diverse types of glycosylation. The mammalian Golgi complex comprises numerous laterally connected Golgi stacks, which typically consists of 4 – 11 tightly adjacent and flat membrane sacs called cisternae. The serially stacked Golgi cisternae are further categorized, from one end to the other, as cis , medial and trans -cisternae. At the trans -side of a Golgi stack, the trans -most membrane sac develops into a tubular and reticulum membrane network called the trans -Golgi network (TGN) ⁴ . In the secretory pathway, cargos derived from the endoplasmic reticulum (ER) enter a Golgi stack at its cis -side and then sequentially pass through medial and trans -cisternae. Cargos eventually exit the Golgi at the trans -Golgi or TGN destining to the plasma membrane, endosomes or secretory granules.

The molecular and cellular mechanisms of how cargos transit a Golgi stack and how the Golgi maintains its cisternal organization remain mysterious and are currently still under a heated debate ¹ . One of difficulties in this field is that Golgi cisternae can only be resolved under the electron microscopy (EM) since the resolution of an optical microscope (~ 200 nm) is insufficient to resolve individual Golgi cisternae (< 100 nm in both cisternal thickness and distance). Therefore, the sub-Golgi localization of resident proteins and transiting cargos are conventionally determined by the immuno-gold EM. However, the immuno-gold EM is very technically demanding and it is beyond the capability of most cell biology labs. Although the resolution of the EM can be sub-nanometer, the resolution afforded by the immuno-gold EM is greatly hampered by the size of the antibody complex (primary plus the secondary antibody) and the gold particle, and it can be worse than 20 nm. Furthermore, EM images are obtained from 2D thin-sections instead of a 3D global view of the Golgi, which can result in erroneous conclusions depending on the relative position and orientation of the 2D section ⁵ . For example, studying an EM single-section is unable to reliably differentiate a vesicle from the orthogonal view of a tubule since both can display identical round membrane profiles. The recent advent of super-resolution microscopy techniques, such as 3D-structured illumination microscopy (3D-SIM), stimulated emission depletion (STED), photoactivated localization microscopy (PALM) and stochastic optical reconstruction microscopy (STORM), makes it possible to resolve sub-Golgi structures under light microscopes ⁶ . However, there are at least four drawbacks that can significantly limit their uses in the cell biological study of the Golgi. 1) Current super-resolution techniques require expensive and special hardware configuration which is beyond most cell biology labs. 2) Special fluorescence labeling protocols are needed for some super-resolution techniques. 3) Although, under the best condition, these techniques claim 20-110 nm in spatial resolution, the practical resolution obtained in real samples can be much worse. 4) In comparison to conventional microscopy, these super-resolution techniques still have difficulties in conducting multicolor, 3D or live cell imaging, either singly or in combination. Probably most importantly, both immuno-gold EM and the super-resolution microscopy techniques yield qualitative instead of quantitative localization data.

Attempting to partially solve problems mentioned above, we have recently developed a conventional light microscopy based method, which is named Golgi protein localization by imaging centers of mass (GLIM), to systematically and quantitatively localize a Golgi protein at a resolution equivalent to that of the immuno-gold EM ⁷ . In this method, the Golgi in cultured mammalian cells is dispersed as Golgi mini-stacks by the treatment of nocodazole, a microtubule depolymerizing drug. Extensive studies have demonstrated that nocodazole-induced Golgi mini-stacks (hereafter Golgi mini-stacks) closely resemble native Golgi stacks in both organization and cellular functions ^{8 , 9 , 10 , 11} . The localization quotient (LQ) of a test protein can be acquired through GLIM and it denotes the quantitative sub-Golgi localization. The numerical values of LQs can be compared and a LQ database of more than 25 Golgi markers has been available.

In GLIM, Golgi mini-stacks are triple-labeled by endogenous or exogenously expressed GM130, GalT-mCherry and the test protein (x). GM130 and GalT-mCherry, cis – and trans -Golgi markers respectively ^{12 , 13} , provide reference points. The triple fluorescence, red (R), green (G) and far-red (B), are artificially displayed as red, green and blue, respectively. Center of fluorescence mass (hereafter center) is adopted to achieve sub-pixel resolution. The Golgi axis is defined as the vector from the center of GM130 to that of GalT-mCherry. The Golgi mini-stack is modeled as a cylindrical structure with infinite rotational symmetry around the Golgi axis. Therefore, a Golgi mini-stack can be further modeled as an one-dimensional structure along the Golgi axis. The LQ of the test protein x is defined as d _x /d ₁ , in which d _x is the distance from the center of x to that of GM130, while d ₁ is the distance from the center of GalT-mCherry to that of GM130. If the center of x is off-axis, its projection axial distance is used for the calculation. The variables, including Golgi axis, axial angle, d _x , d ₁ , angle α and angle β, for GLIM are schematically illustrated in Figure 1 . LQ is independent of the Golgi axial angle though Golgi mini-stacks orient randomly in a cell.

Golgi mini-stacks appear inhomogeneous in images. We developed three criteria to select analyzable Golgi mini-stacks for GLIM. 1) The signal-to-noise ratio criterion, in which the ratio of the total intensity of a Golgi mini-stack to the standard deviation (SD) of the background is ≥ 30 in each channel. This criterion is to ensure the positioning accuracy of the center of mass, which depends on the signal-to-noise ratios of Golgi mini-stacks. 2) The axial angle or distance criterion, which requires d ₁ ≥ 70 nm. d ₁ decreases with the increase of the Golgi axial angle. When the axial angle is approaching 90° or vertical, the mini-stack becomes non-resolvable as d ₁ is approaching 0. d ₁ ≥ 70 nm can effectively exclude near vertical Golgi mini-stacks. 3) The co-linearity criterion, in which either |tan α| or |tan β| is ≤ 0.3. This criterion ensures that the three centers of a mini-stack are sufficiently co-linear for our one-dimensional model of the Golgi mini-stack. All light microscopes suffer from chromatic aberration which can seriously distort the relative positions of red, green and far-red fluorescence centers. Chromatic aberration of microscope systems is experimentally calibrated by imaging 110 nm beads, which are triple-labeled by red, green and far-red fluorescence. For each bead image, the center of red is defined as the true position of the bead and chromatic-shifts of green and far-red centers are fitted by first-order polynomial functions. Centers of Golgi mini-stacks are subjected to the polynomial functions to correct the chromatic-shifts in green and far-red channels.

Through GLIM we can achieve a resolution of ~ 30 nm along the Golgi axis under standard conditions. Importantly, it provides a systematical method to quantitatively map any Golgi protein. GLIM can be performed by conventional microscopes, such as wide-field or confocal microscopes, using common fluorescence labeling protocols. The imaging and data processing can take as short as an hour. Through GLIM, we have directly demonstrated the progressive transition of the secretory cargo from the cis – to trans -side of the Golgi ⁷ .

Note: Below is a step-by-step protocol of GLIM for determining the LQ of EGFP-tagged tyrosylprotein sulfotransferase 1 (TPST1), a Golgi resident enzyme, in HeLa cells.

1. Preparation of Fluorescence-labeled Golgi Mini-stacks

3. Image Acquisition

NOTE: GLIM requires images of high signal-to-noise ratio (SNR) for high precision center of mass calculation. The image can be acquired by conventional microscopes such as the laser scanning confocal, spinning disk confocal or wide-field microscope. A wide-field microscope equipped with plan-apochromatic objective lenses and a low noise image sensor, such as a charge-coupled device (CCD) and scientific complementary metal-oxide semiconductor (sCMOS) can be used. Parameters for the image sensor are adjusted to ensure a low read-noise and high dynamic range. The microscope must be equipped with the optimal configuration of fluorescence filters for green, mCherry and far-red fluorophore, and it must have negligible fluorescence cross-talk. Ideally, the imaging system achieves a Nyquist sampling rate in the x, y and z axis, which typically requires the x, y and z size of a voxel to be less than 100, 100 and 200 nm, respectively. The x and y size of the voxel are always equal and are referred to as pixel_size. The pixel_size can be calculated by dividing the camera sensor size by the system magnification.

4. Image Analysis

The modern research grade light microscope equipped with a plan apochromatic lens, such as the one used in our lab, shows minimal chromatic aberration ( Figure 2 A ). However, a careful examination of the multi-color fluorescent bead image can reveal the shift of different color images of the same bead ( Figure 2 B ). We define that the red channel is free of chromatic aberration and therefore centers of red fluorescence are the true positions of beads. The relative chromatic-shifts of green and far-red fluorescence can be represented by green and blue vectors originating from the center of red fluorescence ( Figure 2 C ). The chromatic-shift within the imaging-plane is empirically illustrated by the green and blue vector for each bead ( Figure 2 D ). For our microscope, the chromatic-shifts are not uniform as both magnitudes and directions of shifts change according to x and y positions. The chromatic-shift can significantly affect the accuracy of GLIM as the shift of far-red centers can be as much as 50 nm. Therefore chromatic-shift must be corrected. Once centers of beads are acquired, we employ first-order polynomial fitting to calibrate and subsequently correct the chromatic-shifts of centers. The chromatic-shift aberration is greatly improved by this approach ( Figure 2 D-G ).

In mammalian cells, the Golgi complex is aggregated at the perinuclear area which is usually not resolvable under a conventional optical microscope ( Figure 3 A ). After nocodazole treatment for 3 hours, the perinuclear Golgi complex disappears and dozens of Golgi mini-stacks assemble at the endoplasmic reticulum exit sites throughout the cytoplasm ( Figure 3 B ). Large chunks of Golgi membrane and aggregated Golgi mini-stacks are present and they are not selected for analysis. An example image illustrating the selection of Golgi mini-stacks is shown in Figure 3 C . Using the "Macro-Golgi ROI inspection" tool, 40 selected Golgi mini-stacks are listed in Figure 3 D . After applying the three criteria, 21 Golgi mini-stacks were analyzable and chosen for the calculation of LQs of TPST1-EGFP ( Figure 3 D ; labeled as "L"), with their statistics shown in Figure 3 E . In total, 111 analyzable Golgi mini-stacks were selected from 12 cells and the LQ of TPST1-EGFP was measured to be 0.76 0.04 (mean ± SEM; n = 111) ( Figure 3 F ). The LQ of TPST1-EGFP positions it between giantin (LQ = 0.59) and EGFP-Rab6 (LQ = 1.04) ⁷ . It is near GS15 (LQ = 0.83) and ST6GalT1-AcGFP1 (LQ = 0.82). We have defined the sub-Golgi regions according to LQs by considering qualitative localization data from literature: the ERES/ERGIC, < -0.25; cis , 0.00 ± 0.25; medial, 0.50 ± 0.25; trans -Golgi, 1.00 ± 0.25 and TGN, 1.25 – 2.00. The LQ of TPST1-EGFP therefore indicates its trans -Golgi localization, which is in agreement with a previous study ¹⁴ .

SUPPLEMENTAL MATERIALS: Macro-Golgi ROI inspection.ijm, Macro-Output 3 channels data.ijm, Matlab codes, My_train.exe, My_test.exe, and Spreadsheet template.opj.

Previously, the localization of a Golgi protein under the light microscopy was mainly quantified by the degree of correlation or overlapping of the image of the protein with the image of a Golgi marker of known localization ^{15 , 16 , 17} . The resulting correlation or overlapping coefficient reflects how close the testing protein is to the Golgi marker spatially. There are at least three caveats for this approach. First, the correlation or overlapping coefficient is nonlinear and it does not directly indicate spatial distance. Second, the degree of correlation is critically dependent on the resolution of the microscopic system. Therefore, the coefficient between two Golgi proteins is not a system-independent constant. Third, two Golgi proteins with the same axial localization but different lateral distribution along cisternae can have distinct coefficients. Hence, the correlation or overlapping coefficient does not indicate the axial localization of a Golgi protein. In an alternative method proposed by Dejgaard et al ., cis , trans -Golgi markers and the protein of interest are triple-labeled in nocodazole treated Golgi mini-stacks, similar to GLIM. Within each Golgi mini-stack, positions of the three maximum intensity pixels are acquired and the relative position of the test protein is calculated as a distance ratio ¹⁸ . However, their method is unable to achieve sub-pixel resolution, which greatly limits its application. Compared to previous quantitative methods, GLIM, which was independently developed but bears a similar concept to that of Dejgaard et al ., is able to quantify the axial localization of a Golgi protein with unprecedented accuracy and consistency.

We presented the protocol of GLIM to acquire the LQ of exogenously expressed GFP-tagged protein – TPST1-EGFP. The LQ of endogenous protein can be determined if its antibody is available. Depending on the species of the primary antibody, mouse or rabbit, then a rabbit or mouse anti-GM130 antibody, respectively, can be used in the triple-labeling protocol. Both rabbit and mouse anti-GM130 antibodies for immunofluorescence labeling are commercially available. We have previously demonstrated that rabbit and mouse anti-GM130 antibodies give the same results in GLIM ⁷ . If the test protein is intrinsically red in fluorescence emission, such as mCherry fusion, a GFP-tagged Golgi marker protein with known LQ in combination with GM130 antibody labeling can be used to triple-label cells and indirectly deduce the LQ of the test protein. Assuming the marker protein's LQ is LQ _m and the distance from the center of the marker protein to that of GM130 is d _m , the LQ of the test protein x can be calculated as (d _x /d _m )×LQ _m . One of the greatest advantages of GLIM in comparison to super-resolution microscopy is that it can easily be applied to the live cell imaging in conventional microscopic setups. We have tried a combination of three fluorescence proteins, GFP-Golgin84, mCherry-GM130 and GalT-iRFP670, for dynamically monitoring the LQ of GFP-Golgin84 in single Golgi mini-stacks ⁷ .

In GLIM, the most time-consuming step is the post-acquisition analysis, especially on the selection of analyzable Golgi mini-stacks. As Golgi mini-stacks are heterogeneous in both size and molecular composition ¹⁹ , it is unclear if the distribution of LQs ( Figure 3F ) represents the heterogeneous architecture of Golgi mini-stacks or the uncertainty of our calculation of the LQ. Regardless of the causes, we found that it is important to have a large number of analyzable Golgi mini-stacks (n) to ensure the accuracy of the LQ, which is compromised when n is small. Typically, data with n ≥ 100 from multiple cells yields reliable LQs. Therefore, GLIM gives ensemble-averaged localization data, obscuring the individuality of Golgi mini-stacks. Another limitation of GLIM is that it assumes that all Golgi proteins have a narrow distribution around a single center. Some Golgi proteins, such as COPI subunits, ARF1 and soluble N-ethylmaleimide-sensitive factor attachment protein receptor (SNAREs) ^{20 , 21} , have been reported to distribute broadly from the cis to the trans -Golgi and the TGN. It is probably inappropriate to study their Golgi localization by the LQ. As GLIM currently involves manual selection of the Golgi mini-stacks, which is both laborious and subjective, the future development of GLIM will be to implement a software tool to automatically select Golgi mini-stacks with minimal human interference.

What is the spatial resolution of GLIM? Since the LQ is a ratio, which is unitless, it does not indicate a spatial distance. Here, we attempt to give a very rough estimation. The spatial resolution of GLIM can be defined as the smallest axial distance between two resolvable Golgi proteins. Examining LQs of various Golgi proteins ⁷ , we found that SEMs of datasets with n ≥ 100 range around 0.03. Assuming two Golgi proteins have the same n = 100 and SEM = 0.03, the two Golgi proteins have significant different localization by t-test (p < 0.05), i.e. , resolvable, if the difference between their LQs is ≥ 3 × SEM = 0.09. From the EM data, we can estimate that the axial length of the Golgi mini-stack, which is also the distance from GM130 to GalT-mCherry in GLIM, is ~ 300 nm ⁸ . Hence, the resolution of GLIM is estimated to be 0.09 × 300 = 30 nm along the Golgi axis. In GLIM, a larger n generally yields smaller SEM, which in turn results in higher resolution.

It is probably inappropriate to directly compare the resolution of GLIM to that of the immuno-gold EM since the latter technique does not directly yield quantitative localization data. Similar to GLIM, the resolution of the immuno-gold EM along the Golgi axis can be defined as the smallest distance between two resolvable Golgi markers. To measure its resolution requires the study of dual-immuno-gold labeling of two Golgi proteins that are closely adjacent to each other along the Golgi axis. Such systemic study is probably unavailable according to our knowledge. As discussed in the introduction, one of the limiting factors in the spatial resolution of the immuno-gold EM is the large size of the antibody complex, which makes the resolution to be worse than ~ 20 nm. Another important limiting factor of the image resolution is the labeling density of antigen molecules. According to the Nyquist sampling theory, there must be at least two gold particles per resolution unit. In most immuno-gold EM studies, the distances between neighbor gold particles are not < 15 nm due to the very low labeling efficiency; this makes it difficult for this technique to achieve an image resolution less than 30 nm. From this point of view, we estimate that the resolution of GLIM is at least comparable to that of the immuno-gold EM.

GLIM is a robust method that gives highly consistent results. It is independent of the type of microscope used. We have tested that wide-field and spinning disk confocal microscopes yielded the same results. LQs of the same Golgi proteins acquired at different batches of experiments were also in good agreement with each other. A LQ database consisting of a diverse range of Golgi resident proteins has been generated for comparison and interpretation. We expect that more usage of GLIM in the research community will significantly expand the database. Consequently, a more complete database quantitatively describing the localization of a large number of Golgi proteins will greatly help in understanding the organization and function of the Golgi complex.

We would like to thank D. Stephens (University of Bristol, Bristol, United Kingdom) for the TPST1-EGFP DNA plasmid, as well as Lakshmi Narasimhan Govindarajan for helping with the software optimization. This work was supported by grants from the National Medical Research Council (NMRC/CBRG/007/2012), Ministry of Education (AcRF Tier1 RG 18/11, RG 48/13 and RG132/15 and AcRF Tier2 MOE2015-T2-2-073) to L.L.