• deconvolution;
  • endoplasmic reticulum;
  • fluorescence microscopy;
  • Golgi apparatus;
  • immuno electron microscopy;
  • modeling;
  • protein quantification


  1. Top of page
  2. Abstract
  3. Results
  4. Discussion
  5. Materials and Methods
  6. Acknowledgments
  7. References

Several lines of evidence support a novel model for Golgi protein residency in which these proteins cycle between the Golgi apparatus and the endoplasmic reticulum (ER). However, to preserve the functional distinction between the two organelles, this pool of ER-resident Golgi enzymes must be small. We quantified the distribution for two Golgi glycosyltransferases in HeLa cells to test this prediction. We reasoned that best-practice, quantitative solutions would come from treating images as data arrays rather than pictures. Using deconvolution and computer calculated organellar boundaries, the Golgi fraction for both endogenous β1,4-galactosyltransferase and UDP-N-acetylgalactosamine:polypeptide N-acetylgalactosaminyltransferase 2 fused with green fluorescent protein (GFP) was 91% by fluorescence microscopy. Immunogold labeling followed by electron microscopy and model analysis yielded a similar value. Values reflect steady-state conditions, as inclusion of a protein synthesis inhibitor had no effect. These data strongly suggest that the fluorescence of a GFP chimera with an organellar protein can be a valid indicator of protein distribution and more generally that fluorescent microscopy can provide a valid, rapid approach for protein quantification. In conclusion, we find the ER pool of cycling Golgi glycosyltransferases is small and approximately 1/100 the concentration found in the Golgi apparatus.

Several lines of evidence support a novel model for Golgi protein residency in which these proteins dynamically cycle between the Golgi apparatus and the endoplasmic reticulum (ER). Chief amongst these was the outcome of ER exit block experiments in which Golgi-resident proteins were found surprisingly to accumulate in the ER (1–5). This slow recycling, halftime of about 1.5 h, was by a coat protein I (COP I)-independent pathway (6). The evidence for the COP I-independent trafficking of some bacterial and plant toxins from the Golgi apparatus to ER gave further credence to a COP I-independent pathway for resident Golgi protein recycling (6,7). The ability of Golgi-associated rab protein isomers to induce recycling of Golgi-resident proteins to the ER also supports the possibility that Golgi protein residency may be apportioned normally between the Golgi apparatus and ER (8–12). Perhaps the earliest indication of a COP I-independent recycling pathway came from the brefeldin A (BFA) experiments in which the Golgi apparatus upon drug treatment collapsed into the ER (13). In addition to a COP I-independent pathway(s) for resident Golgi protein recycling to the ER, there exists a COP I-dependent pathway that is functionally important in the retrieval of leaked ER proteins back to the ER. For recent reviews of Golgi protein recycling, please see Storrie et al. (14,15).

The most fundamental prediction of any model of Golgi-resident protein recycling to the ER is the existence of a distinct, quantifiable, steady-state ER pool of Golgi-resident proteins. However, the existence of such a pool poses distinct problems in maintaining separation of function between the Golgi apparatus and ER. Golgi-specific glycosyltransferases and glycosidases do not normally modify ER-resident glycoproteins [for review, see (16)]. Only when Golgi proteins are induced to accumulate in the ER in response to BFA treatment (13), rab protein overexpression (11), or ER exit block (5) do such modifications occur. Thus, any such pool must be small. Actual values reported for the distribution of Golgi-resident proteins between the organelle and ER vary from as much as 33% [e.g. (1)] to 8–9% [e.g. (17)] to as low as none detectable [e.g. (18)], making this a controversial issue. Techniques used to evaluate this distribution include fluorescence microscopy, cell fractionation and electron microscopy. Determination of this value is important both to substantiate the recycling model but also in the broader conceptual sense of providing a framework into which the distinctive protein and lipid-processing properties of the Golgi apparatus and ER may be reconciled with the dynamics of protein cycling between the two.

In the present work, we test this prediction through the quantification of distribution of two different Golgi glycosyltransferases, UDP N-acetylgalactosamine:polypeptide N-acetylgalactosaminyltransferase 2 (GalNAcT2) and β1,4-galactosyltransferase (GalT), between the Golgi apparatus and ER. GalNAcT2 and GalT were chosen on the basis of being representative Golgi resident, type II, transmembrane proteins for which either high-quality antibodies or well-characterized green fluorescent protein (GFP) chimeric proteins exist as tools for quantification. In the case of GalNAcT2, the protein samples the entire cisternal space of the Golgi apparatus being distributed from cis to trans (19). GalNAcT2 was either epitope-tagged or fused as a truncated polypeptide with GFP. In either case, the overexpressed protein was stably expressed and has been shown previously to have a normal distribution across the Golgi apparatus (3,19). GalT is concentrated in the trans Golgi cisternae to trans Golgi network, and the endogenous protein distribution was characterized. Taking a range of approaches including quantitative light and electron microscopy and kinetic modeling, we find that both proteins are distributed in an approximately 90:10 ratio, Golgi apparatus to ER at steady state. Furthermore, our methodologies for analyzing light-microscopy images as data arrays provide a quantitative approach applicable to a variety of intracellular protein studies. The analytical validation of GFP chimeric protein distribution as representative of the endogenous distribution of Golgi glycosyltransferases provides a rapid and general approach to distributional quantification of proteins.


  1. Top of page
  2. Abstract
  3. Results
  4. Discussion
  5. Materials and Methods
  6. Acknowledgments
  7. References

Golgi-to-ER exchange kinetics indicate an approximately 90:10 Golgi-to-ER distribution

The Cisternal Maturation model requires that glycosyltransferases ‘resident’ in the Golgi apparatus actively recycle during each round of transport. Much of this recycling must be within the Golgi cisternal stack. However, a portion may well be recycled between the Golgi apparatus and ER as part of an ongoing need to balance membrane flows [for review, see (14)]. Hence, a quantifiable fraction of fully processed Golgi enzymes would be normally found in the ER. Fluorescence microscopy data from Zaal et al. (1) suggest that approximately 67% would be Golgi localized and the remaining 33% in the ER. Such a high ER value was surprising given that Golgi-specific glycosyltransferases and glycosidases do not normally modify ER-resident glycoproteins [for review, see (16)]. We, therefore, looked at quantifying this distribution using direct kinetic rate measurements.

Assuming the transport time between the Golgi and the ER is much shorter than the residence times within, we consider the Golgi and the ER to be two separate compartments connected by first-order transport processes (Figure 1A). The change in ER enzymes as a function of time can be described by

  • image(1)

where E is the relative concentration of proteins in the ER and kGA and kER are the rate constants for transport between the Golgi and the ER in the retrograde and anterograde directions, respectively. Degradation and synthesis are neglected in our analysis due to experimental conditions investigated. The general solution of the differential equation can be written as

  • image(2)

where the integration constant C is determined from the initial condition. The steady-state solution, Ess, is given by a ratio of the rate constants:

  • image(3)

Using GalNAcT2-GFP transfected HeLa cells, we previously collected fluorescent microscopy data in the presence of cycloheximide (CHX), with and without mSar1pdn protein as an ER exit block, that can now be used to determine the steady-state distribution (5). In the presence of mSar1pdn, GalNAcT2-GFP loss from the Golgi and accumulation in the ER were recorded and quantified as previously described. For this analysis, kER was set equal to zero by definition, and the final steady state was reached when all proteins relocated at the rate kGA into the ER (Ess = 1). A value of 0.57 ± 0.04/h for kGA was reported. In the absence of mSar1pdn, FRAP experiments were done following photobleaching of approximately 35% of ER fluorescence. The intensity recovery data from the time-lapse image set was then fit to equation 2, and from three separate measurements, we found kGA + kER to be 4.00 ± 0.42/h (Figure 1B). Substituting these measured values into equation 3 yields an ER pool value for GalNAcT2-GFP of 14 ± 2% and, hence, a distribution of approximately 90:10 Golgi to ER. Given these measured rate constants, one can say that GalNAcT2-GFP would exit the Golgi apparatus at a rate of 1.0%/min and exit the ER at a rate of 5.7%/min. The reciprocals of these values yield the mean residence times of 106 ± 8 min in the GA and 17.8 ± 2.2 min in the ER for a total cycle time of 124 ± 8 min for GalNAcT2, in contrast with previously estimated total Golgi and ER residence times of 84.6 ± 11.3 min for GalT (1).


Figure 1. Two-compartment model for Golgi glycosyltransferase cycling.A)Schematic diagram of the two-compartment kinetic model. The endoplasmic reticulum (ER) and the Golgi apparatus are considered as two separate compartments connected by first-order transport processes. The rate constants kER and kGA are the ER-to-Golgi (anterograde) and Golgi-to-ER (retrograde) transport rates, respectively. Golgi-to-ER transport rate kGA = 0.57 ± 0.04/h was measured from an ER exit block experiment using mutant Sar1pdn proteins by Miles et al. (5).B)Measured fluorescence recovery after 35% photobleaching of the ER in HeLa cells expressing GalNAcT2 green fluorescent protein. Relative ER fluorescence was averaged over three separate HeLa cells. Data are redrawn from Miles et al. (5). The solid curve reflects the fit of the data points to the exponential equation shown. The fitted parameter m2 in the table inset represents the sum of rate constants kER + kGA. Both rate measurements were done in the presence of protein synthesis inhibitor cycloheximide.

Download figure to PowerPoint

Immunogold electron microscopy indicates approximately 90:10 distribution of GalNacT2 between the Golgi apparatus and ER

Given the difference in distribution values between the kinetic model analysis of fluorescent microscopy data for GalNAc-T2 and the previously reported direct fluorescent microscopy results for GalT (1), we chose to determine the distribution between the Golgi and ER of GalNAcT2 using immunogold labeling of thawed cryosections combined with electron microscopy. In contrast to light microscopy, the resolution of the electron microscope is high relative to the underlying structures. On the basis of our kinetic modeling results, we expected the level of ER labeling to be low for any given glycosyltransferase. Therefore, for the immunogold experiments, we chose to use stably transfected GalNAcT2-VSV (VSV, P5D4 epitope tag from vesicular stomatitis virus G Protein). At fivefold overexpression, GalNAcT2-VSV has the same cellular distribution as that of an endogenous protein (19) and has been found repeatedly to have the same cycling kinetics between the Golgi apparatus and ER as other Golgi enzymes (3,5,6).

Two sets of electron micrographs were taken, one at magnification of 34 000 to quantify immunogold-labeling density and a second at lower magnification to score overall Golgi and ER areas. For immunogold labeling, cryosections were incubated sequentially with anti-VSV primary antibody/10 nm immunogold to localize GalNAcT2 and with anti-protein disulfide isomerase (PDI)/5 nm immunogold to identify ER. GalNAcT2, positive regions were photographed at random and scored morphometrically (3). For overall area stereology, cells were photographed at random at a magnification of 10 000 or 16 000 and organelles identified by their characteristic morphology. Consistent with previous results (3,19), 10-nm GalNAcT2 labeling was heavily concentrated over Golgi cisternae (arrowhead, Figure 2A) and associated tubules (arrows, Figure 2A) with occasional 10-nm GalNAcT2 labeling (arrowhead, Figure 2B) in association with the PDI-positive ER (arrow, Figure 2B). As shown in Figure 2C, Golgi cisternal stacks (arrowheads) and ER (arrow) could be readily distinguished morphologically in the lower magnification micrographs.


Figure 2. Electron micrographs of UDP-N-acetylgalactosamine:polypeptide N-acetylgalactosaminyltransferase 2 (GalNAcT2)-VSV HeLa cells with immunogold labeling (A, B) and without (C).A)Electron micrograph acquired at 34 000 magnification showing Golgi cisternae (arrowhead) and Golgi tubules (arrow) where 10-nm gold particles indicate VSV labeling as described in the Materials and Methods.B)Thirty-four thousand magnification-acquired image depicting endoplasmic reticulum (ER) immunolabeled for GalNAcT2 with 10-nm (arrowheads) and 5-nm gold particles labeling protein disulfide isomerase, an ER marker (arrow).C)An example of a randomly selected image acquired at 16 000 magnification and illustrating Golgi stacks (arrowheads) and ER ribbon (arrow). Note that the Golgi tubules are not readily apparent, and spacing between cisternae is difficult to reliably distinguish at this magnification.

Download figure to PowerPoint

As expected, the 10-nm GalNAcT2 labeling density was much higher over Golgi cisternae and associated tubules than over the ER, mitochondria, or nucleus. We took the average of both mitochondrial and nuclear labeling as an indicator of non-specific reactivity. The level of non-specific labeling was approximately 30% of that observed over the ER. The mitochondrial/nuclear labeling density was subtracted from all values to give a corrected Golgi apparatus- and ER-labeling density for GalNAcT2-VSV. Labeling density over the Golgi apparatus was approximately 80–100 higher than that over the ER. In order to convert labeling density to the total distribution of GalNAcT2-VSV between Golgi apparatus and ER, we determined the relative area of ER and Golgi apparatus (cisternae plus tubules) in low- magnification, random cell sections. As summarized in Table 1, the Golgi apparatus area was approximately 12% of the ER. In total, 90% of the GalNAcT2 labeling was associated with the Golgi apparatus and 10% with the ER (Table 1).

Table 1. GalNAcT2-VSV relative protein distribution by immunogold labeling
 Golgi cisternaeGolgi tubulesER
  1. GalNacT2-VSV gold particle-labeling density over Golgi cisternae, associated Golgi tubules and endoplasmic reticulum (ER) was determined as described in Materials and Methods. Values were corrected for background, non-specific labeling by subtraction using the average of the labeling density over mitochondria (2.00 particles per µm2) and nucleus (1.09 particles per µm2). Similarly, stereology was used to determine the relative area of the Golgi cisternae, Golgi tubules and ER. n = 2 for the number of determinations. The ± values indicate range of the data. Number of gold particles counted: Golgi cisternae, 1339; Golgi tubules, 1248; ER, 24; mitochondria, 8; and nucleus, 26.

Particle density (number/µm2)2271662.86
Relative area average (n = 2)5.57 ± 0.47.09 ± 0.5100
Relative protein average (n = 2)90 ± 1.510 ± 1.5

Best-practice widefield fluorescence microscopy yields approximately 90:10 Golgi-to-ER steady-state distribution for Golgi glycosyltransferases

The EM and kinetic analysis distribution values indicated an approximately 90:10 distribution for GalNAcT2 leading us to investigate whether a similar steady-state distribution of GalNAcT2 between the Golgi apparatus and ER would be obtained using fluorescence microscopy. HeLa cells are fairly thin, approximately 6–7-µm thick, when imaged as fixed, plastic-mounted samples. As such, widefield images focused for bright Golgi apparatus sampling should contain intensity information from the full cell volume; the Golgi apparatus is located at approximately mid-cell height. This was essentially the approach used by Zaal et al. (1) in quantifying GalT-GFP distribution in a single plane image collected with a laser-scanning fluorescence microscope operated with the pinholes wide open. For our studies, all widefield images were collected at a resolution sufficient for deconvolution analysis (1.4 numerical aperture objectives and approximately 2× oversampling to give images that met Nyquist sampling criteria).

GalNAcT2-VSV, GalNAcT2-GFP and GalT all distributed in a similar manner by widefield fluorescence microscopy (Figure 3). Antibody staining was used to reveal GalNAcT2-VSV and GalT distributions and the inherent fluorescence of the GFP moiety to reveal the distribution of GalNAcT2-GFP. In images displayed with a normal grayscale range (100–3000 grayscale levels, 12 bit camera), all had a distinct juxtanuclear Golgi-like fluorescence distribution with little detectable fluorescence observed over the cytoplasm (Figure 3A,B,C). When the same images were displayed with a compressed grayscale range that accentuated the display of low-intensity fluorescence (100–300 grayscale levels), three striking distributional features were now apparent (Figure 3A′,B′,C′). First, the bright, juxtanuclear Golgi apparatus area was now much larger. This was an expected outcome for out-of-focus plane Golgi intensity (i.e. blur) and light spread by the objective, also known as point-spread function. Second, there was cytoplasmic fluorescence apparent for all three. Third, particularly obvious for GalNAcT2-GFP (Figure 3B′) and GalT (Figure 3C′), the fluorescence pattern gave a rim-like fluorescence staining at the nuclear envelope, typical of an ER distribution. For GalNAcT2-VSV, this was less obvious. However, this is not unusual for antibody staining as even antibodies against bona fide ER proteins such as p63 often give a somewhat granular staining pattern [see e.g. Figure 1 in (6)]. In double label experiments with Sec61p, an ER marker, near-complete overlap between the cytoplasmic GalNAcT2-GFP fluorescence and Sec61p staining was observed (data not shown). We conclude from this that there is a low quantifiable level of both GalNAcT2 and GalT present in the ER. At first glance, the intensity of GalNAcT2-VSV staining over the ER was slightly more intense than that of GalNAcT2-GFP or GalT (compare Figure 3A′,B′,C′). The overall background brightness of the GalNAcT2-VSV labeling, however, was slightly higher than that for GalNAcT2-GFP or GalT, and the net distribution is the same (summarized in Table 3).


Figure 3. Golgi enzymes are found in the endoplasmic reticulum (ER) at lower concentrations by widefield light microscopy. HeLa cells stably expressing GalNAcT2-VSV (A, A′) stained for VSV, expressing fluorescent GalNAcT2-GFP (B, B′) fixed and mounted, and wild-type cells stained for endogenous GalT (C, C′) are shown as single plane widefield images. Identical images are shown at normal (A – C, linear mapping of 100–3000 grayscale levels of 0–4095) and high(A′–C′, 100–300 grayscale levels of 0–4095) brightness. Note that the nuclear envelope (arrowheads) typical of the ER distribution only becomes visible under high brightness (A′–C′). Bar, 10 µm.

Download figure to PowerPoint

Table 3. Comparison of grayscale fluorescence intensity parameters by three fluorescence microscopy methods
MethodMarkerWildtypeERGolgi (vis)Signal-to-noise (Golgi)n
  1. Average grayscale pixel intensities of the wild-type (WT) and ER signal as well as average net intensity of visually identified Golgi apparatus are corrected as shown in the equations below. The noncell background was approximately 100 (12-bit camera, 4095 maximum grayscale value) for widefield and spinning-disk images and was approximately 2 (8-bit photomultiplier scaling, 255 maximum grayscale) for laser-scanning images. Each of these values corresponds to what can be considered as the dark current sensor set point. For widefield and spinning-disk images, the signal-to-noise ratio for average Golgi intensity was calculated by dividing the signal with the background noise in units of photoelectrons. The conversion factor of electrons/(analog-to-digital unit) was obtained from the mean and variance of two flat-field images. For laser-scanning images, signal-to-noise ratio was calculated from the square root of signal divided by average single-photon hit in the background. All numbers are calculated for n individual cells and shown as mean ± SEM.

  2. Equations:

  3. WT = average grayscale value − intercellular background (BG). GalT is an endogenous, wildtype protein. For GalT, 0 = background corrected intercellular intensity value.

  4. ER = average grayscale value − BG.

  5. Golgi = average grayscale value over visually outlined Golgi apparatus − (ER + BG).

WidefieldT2-VSV20.7 ± 7.848.4 ± 211392 ± 383190 ± 5030
T2-GFP4.1 ± 2.013.7 ± 5.9732 ± 508110 ± 7030
GalT015.3 ± 6.71260 ± 391200 ± 6030
Laser scanningT2-VSV0.83 ± 0.501.30 ± 0.2931.6 ± 6.63.0 ± 0.610
Spinning diskT2-VSV13.2 ± 4.325.8 ± 8.6701 ± 9030 ± 625
GalT04.0 ± 1.1409 ± 6854 ± 930

As a first approximation to determining the apportionment of GalNAcT2 and GalT between the Golgi apparatus and ER, wildtype (WT) and tagged HeLa cells were imaged in the same field at a constant illumination intensity and exposure that did not saturate the CCD camera (Figure 4A). To draw organellar and cell boundaries, we displayed images nonlinearly (technically, gamma = 0.4), and cell boundaries and the Golgi apparatus were outlined by eye (visual threshold, Figure 4B). Under these conditions, the boundaries of cytoplasmic fluorescence were apparent, and little detail was lost due to image display saturation. We then calculated the fraction of total fluorescence intensity from the Golgi apparatus compared with that of the whole cell to be approximately 63–64% for either GalNAcT2 (epitope or GFP tagged) or GalT (Table 2). Values were the same whether or not the cells had been cultured for 4 h in the presence of CHX to inhibit protein synthesis. Note that all values were corrected for non-specific background fluorescence. For tagged cells, the correction was based on the mean fluorescence level found in co-cultured and co-imaged WT cells (Figure 4A,B). For endogenous GalT, the level of intensity found between cells was subtracted. Quantitatively, these results are essentially the same as those found by Zaal et al. (1) for the visual quantification of a concatenated GalT-GFP-GFP-GFP chimera using widefield microscopy (laser-scanning microscope operated with wide-open pinholes). Wild-type cytoplasmic intensity was about 40% of the total signal found over the ER region of GalNAcT2-VSV and 30% of GalNAcT2-GFP (Table 3). As widefield microscopy is a high signal-to-noise ratio technique (>100, Table 3), the background-corrected ER values are significant.


Figure 4. Single plane widefield images were analyzed before and after deconvolution using visual and calculated thresholds.A)Shown is a fluorescence image of HeLa cells stably expressing GalNAcT2-VSV polypeptide N-acetylgalactosaminyltransferase 2-VSV stained for VSV, as well as a wild-type HeLa (WT). A nonlinear intensity display (gamma = 0.4) was used to bring out dimmer structures. Bar, 10 µm.B)Cell outlines are shown as well as the Golgi apparatus, as identified visually (visual threshold, arrowhead).C)Shown is the surface plot of the fluorescence intensity. The visual threshold (arrowhead) underestimates the Golgi intensity peak, and the calculated threshold (arrow) is needed to account for the total intensity of the peak.D)The average ER intensity + 2SD in the region (arrowhead) far from the Golgi was used for the calculated threshold (arrow). E, F) Shown are the image and its surface plot after five iterations of deconvolution. The Golgi apparatus as determined from the visual (arrowhead) and calculated (arrow) thresholds of intensity is outlined or marked. Note the overlap between visual and calculated thresholds. All fluorescence images are shown with gamma = 0.4 while surface plots are drawn with actual pixel values (gamma = 1.0).

Download figure to PowerPoint

Table 2. Percentage of glycosyltransferases found in the Golgi apparatus measured from single plane widefield images
  GalNAcT2-VSVGalNAcT2-GFPEndogenous GalT
  1. HeLa cells expressing GalNAcT2-GFP or GalNAcT2-VSV and wild-type cells were cultured ± cycloheximide (CHX, 50 µg/mL, 4 h), fixed and stained [(VSV and endogenous GalT and imaged for fluorescence at Nyquist criterion (100 nm/pixel) as single plane widefield images as described in the Materials and Methods. At least 30 cells were analyzed for each group using raw and deconvolved images with visual and calculated thresholds as shown in Figure 2. Percentage of fluorescence found in the Golgi fraction is given in the table (mean ± SEM). Results from best possible practice of using calculated threshold on deconvolved images are shown in bold.

RawVisual61 ± 562 ± 564 ± 564 ± 565 ± 766 ± 4
Calculated85 ± 584 ± 690 ± 388 ± 487 ± 489 ± 3
DeconvolvedVisual85 ± 684 ± 889 ± 487 ± 588 ± 489 ± 3
Calculated88 ± 787 ± 891 ± 491 ± 491 ± 391 ± 3

As indicated in Figure 3A′,B′,C′, the actual spread of Golgi-specific fluorescence in widefield images was greater than that apparent to the eye. This was readily apparent in a quantitative manner when the image was reformatted as a surface plot of fluorescence intensity versus XY coordinates (Figure 4C). The Golgi signal surface plotted was a clustered set of intensity peaks much like a mountain range, and the visual threshold marked by an arrowhead considerably underestimated the perimeter of the mountain range. To include the whole Golgi signal within the boundary, we measured the average ER intensity within a region far from the Golgi (arrowhead, Figure 4D), calculated a standard deviation (SD) and used the average intensity value + 2SD as a calculated threshold. Negligible amounts of the ER in dispersed spots were above this calculated threshold as tested in HeLa cells stained for Sec61p, an ER marker (data not shown). Moreover, the Sec61p distribution indicated that the ER distribution was uniform about the cytoplasm (data not shown). Such scattered spots were excluded from a Golgi intensity calculation by setting a minimum area limit. The Golgi intensity within the area defined by the calculated threshold was measured and corrected for maximal underlying ER contribution by subtracting the average ER intensity value from each pixel. This approach yielded a Golgi fraction of almost 90% for both the transfected/overexpressing GalNAcT2-VSV and GalNAcT2-GFP and the endogenous GalT (Table 2, raw, calculated). Addition of CHX did not affect the results.

We thus reasoned that the discrepancy between visual and calculated threshold was an outcome of the objective point-spread function. If so, it should be resolved by deconvolution, a mathematical method to reassign light to its source (20). After applying iterative maximum likelihood estimation deconvolution to the raw image, we observed that the gap between visual and calculated thresholds was greatly reduced and the two almost overlapped in the XY fluorescence image (Figure 4E). The surface plot showed sharper peaks with reduced perimeter and near overlap of the two thresholds (Figure 4F). Golgi fractions obtained from the deconvolved widefield images using a visual threshold were 85–89% for GalNAcT2 whether VSV- or GFP-tagged and endogenous GalT, plus and minus CHX, and 87–91% using a calculated threshold (Table 2, deconvolved). This confirmed our hypothesis that objective light spread was responsible for the difference in two different methods for raw images and more importantly indicates that the correct ratio for distribution of Golgi glycosyltransferases between the Golgi apparatus and ER was approximately 90:10. As the same result is obtained in the presence of CHX, we infer that this is a steady-state measurement and only a negligible portion of the ER pool is from new synthesis.

Technically, these results strongly indicate that visually identifying the Golgi apparatus from raw images underestimates the Golgi fraction of the glycosyltransferases at approximately 65% [present work and Zaal et al. (1)]. The use of a calculated threshold and/or deconvolution produced a decidedly higher outcome, approximately 90%, which agreed with both the EM and kinetic-modeling quantification. On the basis of our results, the best practice for widefield microscopy would be to determine a calculated threshold on a deconvolved image, as it applies an objective method to the corrected image (Table 2, bold). Using a calculated threshold on raw images provided a good practical approximation, as it gave only a slightly lower number than the best-practice result and can be applied to undersampled images. In support of the assumption of these experiments that single plane, widefield images focused on bright Golgi apparatus captured fluorescence intensity in a representative manner from the full depth of fixed HeLa cells (approximately 6–7 mm), we found no statistically significant change in total fluorescence intensity following BFA treatment, a condition that disperses Golgi glycosyltransferases into the ER (data not shown).

Best-practice confocal microscopy also yields an approximately 90:10 distribution

Having shown that the light spread has to be corrected for best-practice scoring of Golgi apparatus and ER fluorescence in widefield images, we decided to test whether the light rejection properties of confocal microscopy reduced and/or eliminated this problem. Moreover, with the optical sectioning properties of confocal microscopy, we can test directly the assumption that widefield fluorescence microscopy provides representative intensity information for the entire cell. Therefore, we imaged HeLa cells as stacks of optical sections at spatial intervals that provide near optimal oversampling for three-dimensional deconvolution (approximately 70-nm XY pixel spacing and 150-nm Z spacing).

Laser-scanning confocal microscopy

Initial optical sectioning experiments were performed with a Zeiss LSM510 microscope set to give optimal stray light rejection, 1-Airy unit pinholes, and little to no photobleaching. As expected (21), these conditions resulted in a low signal-to-noise ratio image (Figure 5 and Table 3). As shown in Figure 5A,A′, bright Golgi apparatus staining was observed for GalNAcT2-VSV in raw, unprocessed images. The average Golgi signal-to-noise ratio was three (Table 3). When the intensity display was remapped to show dim fluorescence, slightly brighter fluorescence was observed over the cytoplasm of tagged than WT cells (Figure 5C). This putative ER fluorescence was approximately 50% brighter than the WT background (Table 3). Using either visual or calculated thresholds, the wild-type corrected, stack-summated Golgi fraction was approximately 80%. (Table 4). GalNAcT2-VSV distributions became slightly sharper and had more contrast after deconvolution (Figure 5B,B′,D). However, as shown in Figure 5C,D, the visual and calculated thresholds were approximately the same before and after deconvolution, and the stack-summated Golgi fraction was again approximately 80% (Table 4). As indicated by the low signal-to-noise ratio, the information content embodied in these average values is limited, and therefore, further accumulation of cell data by laser-scanning microscopy was not pursued.


Figure 5. Laser-scanning confocal images. A – B′) Shown are slices in XY-plane (A, B) and XZ-plane(A′,B′) from image stacks of one HeLa-cell expressing GalNAcT2-VSV and staining for VSV, imaged with a laser-scanning confocal microscope (63×, 1.4-NA oil immersion objective, voxel volume: 60 × 60 × 150 nm). Identical cells are shown before (A, A′) and after (B, B′) deconvolution. The arrowheads in A and B correspond to the Y-position of the slices in A′ and B′. To illustrate light spread better, we changed the grayscale mapping to 0–85 in A′ and B′ from 0–255 in A and B; Bars, 5 µm. C, D) Shown are identical slices from A and B with cell outlines and visual (arrowheads) and calculated (arrows) thresholds at gamma = 0.4. Note the overlap between the two thresholds in both images (C, D). Typically, 40 XY-slices constituting a complete stack (approximately 6 µm in cell thickness) were analyzed.

Download figure to PowerPoint

Table 4. Percentage of glycosyltransferases found in the Golgi measured from laser-scanning image stacks
  1. HeLa cells expressing GalNAcT2-VSV were fixed and stained for VSV. Cells (n = 10) were imaged and analyzed as confocal stacks as described in Materials and Methods. Results are the percentage of fluorescence intensity found in the Golgi fraction (mean ± SEM).

RawVisual77 ± 9
Calculated81 ± 11
DeconvolvedVisual78 ± 8
Calculated81 ± 11
Spinning-disk confocal microscopy

Having eliminated laser-scanning microscopy optical sectioning due to low signal-to-noise ratio, we next investigated spinning-disk confocal microscopy as a potentially much higher signal-to-noise technique. The pinhole size of the ATTO CARV unit at 1.22 Airy units approached that of the Zeiss LSM510, and other optical properties were similar. We found that complete HeLa-cell image stacks for GalNAcT2-VSV and GalT staining could be collected without bleaching. As expected (21), spinning-disk confocal microscopy at values of 30–54 gave a considerably higher average Golgi signal-to-noise ratio than laser-scanning confocal microscopy (Table 3). As shown in Figure 6, an individual optical section at approximately mid-Golgi apparatus for GalNAcT2-VSV appeared relatively blur-free compared with widefield microscopy (Figure 2A versus Figure 5A). Similar results were seen for GalT (data not shown).


Figure 6. Spinning-disk confocal images. A – B′) Shown are representative slices in XY-plane (A, B) and XZ-plane (A′, B′) from image stacks of one HeLa-cell expressing GalNAcT2-VSV and a neighboring wild-type HeLa fixed and stained for VSV, imaged with a Nipkow confocal microscope (100×, 1.4-NA oil immersion objective, voxel volume: 72 × 72 × 150 nm). Identical cells are shown before (A, A′) and after (B, B′) deconvolution (10 iterations). The arrowheads in (A, B) correspond to the Y-position of the slices in A′ and B′. Higher brightness (grayscale mapping of 100–1200 from 0–4095) was used in A′ and B′ to show light spread in the Z direction; Bars, 5 µm. C, D) Shown are identical slices from A and B with cell outlines and visual arrowheads and calculated (arrows) thresholds at gamma = 0.4. Typically 40 XY-slices constituting a complete stack (approximately 6 µm in cell thickness) were analyzed.

Download figure to PowerPoint

When viewed with a nonlinear grayscale range (technically, gamma = 0.4), cytoplasmic fluorescence due to ER staining for GalNAcT2-VSV was obvious compared with weak background fluorescence in adjacent WT cells (Figure 6C), see also the surface plot (Figure 6E). For GalNAcT2-VSV, the average ER fluorescence intensity was twofold that of background WT fluorescence, a visually and statistically significant value (Table 3, Figure 6). Using raw, unprocessed image stacks and a visual threshold, the Golgi fraction for both GalNAcT2-VSV and GalT staining was approximately 65% (Table 5), very similar to that found for widefield microscopy in a single image plane. When a calculated ER threshold was applied to the raw, unprocessed image stacks, the Golgi fraction for GalNAcT2-VSV and GalT was now approximately 85%, again similar to that found for widefield microscopy in a single image plane. These values confirm the supposition that the single plane widefield images contain intensity information from the entire HeLa cell. Moreover, comparison of the widefield images with the spinning-disk confocal images (Figures 4 and 6) leads to the conclusion that reduced blur with confocal imaging does lead to the calculated threshold being closer to the visual threshold in XY images. With deconvolution, as expected, the brightness and contrast of the juxtanuclear Golgi apparatus staining increased and Z dimension blur (Figure 5A′,B′), in particular, decreased. As summarized in Table 5, the Golgi fraction for GalNAcT2-VSV and GalT was approximately 65% with visual thresholding and approximately 90% with calculated thresholding of deconvolved image stacks. The deconvolved, calculated threshold values in Table 5 (bold) indicate best-practice values.

Table 5. Percentage of glycosyltransferases found in the Golgi measured from spinning-disk image stacks
ImageThresholdGalNAcT2-VSVEndogenous GalT
  1. HeLa cells expressing GalNAcT2-VSV and wild-type HeLa were fixed and stained for VSV and endogenous GalT. Cells (n = 25, 30) were imaged and analyzed as confocal stacks as described in Materials and Methods. Results are the percentage of fluorescence intensity found in the Golgi fraction (mean ± SEM). Best-practice results are highlighted in bold.

RawVisual64 ± 665 ± 7
Calculated83 ± 489 ± 5
DeconvolvedVisual81 ± 587 ± 5
Calculated88 ± 691 ± 4


  1. Top of page
  2. Abstract
  3. Results
  4. Discussion
  5. Materials and Methods
  6. Acknowledgments
  7. References

We measured the level of Golgi glycosyltransferases in the Golgi apparatus and ER to provide a framework in which the functional distinctiveness of the two organelles and the dynamics of protein cycling between them can be reconciled. At the same time, we provide validated, generally applicable and rapid approaches for the quantification of protein distribution within cells. Despite the general acceptance that many Golgi proteins cycle through the ER, the extent of Golgi protein residence within the ER and more importantly how such residency can be compatible with the functional properties of the two organelles remain highly controversial. In terms of quantification, the controversy is particularly acute for GFP chimeric proteins with Golgi glycosyltransferases [compare (1,18,22)]. Fluorescent protein chimeras are the only approach to many live-cell experiments and the most practical approach to high-throughput screening of protein distribution. Quantitative resolution of these points is key to the application of such methods. We reasoned that the solution to the biological problem would come from treating images as data arrays rather than pictures and analyzing them as such. Taking this approach, we found that endogenous, epitope-tagged and GFP-tagged Golgi glycosyltransferases all had an approximately 90:10 steady-state distribution between the Golgi apparatus and ER. These results compared well with the same measurements made by immunogold labeling and kinetic modeling of protein cycling. The concentration of the two glycosyltransferases in the Golgi apparatus is almost 100-fold higher than that of the same proteins in the ER. The fluorescence approach provides not only a validated quantification of ER versus Golgi protein levels but also validation of GFP chimera as quantitative markers. Conceptually, the low concentration of Golgi enzymes in the ER provides a framework for the distinct functional properties of the organelles.

We used three different approaches to quantify the distribution of GalT and GalNAcT2 between the Golgi apparatus and ER. These were fluorescence microscopy, electron microscopy and kinetic modeling. In the fluorescence microscopy studies, the use of stably expressed, epitope- or GFP-fused GalNAcT2 was particularly important because co-cultured wild-type HeLa cells provided an internal reference for quantifying non-specific background fluorescence. GalNAcT2-VSV is overexpressed approximately fivefold in the HeLa cells used (19). This greatly facilitated its localization by either light microscopy of antibody-stained cells or electron microscopy of immunogold-labeled cryosections. In the analysis of the light micrographs, particular effort was given to analyzing the images as two dimensional data arrays of intensity versus Golgi-specific X,Y coordinates. This provided a solution to the problem of how to objectively draw a perimeter to delimit the Golgi apparatus in light micrographs. We used calculated thresholds based on average ER intensity. In optimized best-practice analysis, the data arrays were sharpened by deconvolution. There were two, perhaps, surprising outcomes. First, single plane widefield images sampled the cellular fluorescence intensity almost as well as confocal image stacks. Presumably, this is a consequence of plastic-mounted HeLa cells being relatively thin, approximately 6–7 µm. Second, spinning-disk confocal imaging was the method of choice for the collection of full cell optical sections. Laser-scanning confocal microscopy was too bleach-prone and too signal-to-noise ratio limited.

From a comprehensive analysis of fluorescent images, we concluded that there was approximately a 90% distribution of Golgi-resident proteins between the organelle and ER. For endogenous GalT, the Golgi fraction was 91%, and for GalNAcT2-GFP, the value was also 91%. Using immunogold labeling followed by electron microscopy, the value was 90% for GalNAcT2-VSV. A similar number came from mathematical modeling. That the distributions reported reflect steady-state conditions in interphase cells is strongly indicated by the lack of effect of a protein synthesis inhibitor on fluorescence distributions. Farmaki et al. (23) find in mitotic cells that CHX treatment inhibits the glycosylation of ER proteins by Golgi enzymes. From this, they infer that the ER pool of Golgi enzymes in mitotic cells is due to accumulation of newly synthesized proteins rather than recycling. We note that this conclusion is based on indirect rather than direct observation. For interphase cells, we find by both electron and fluorescence microscopy that the concentration of GalNAcT2 was approximately 100-fold higher in the Golgi apparatus than ER. By fluorescence microscopy, we found that endogenous GalT showed a similar difference in concentration between the two organelles. We suggest that these low values for ER residency of Golgi glycosyltransferases are compatible with the functional distinctions between the two organelles. In comparison with other approaches to quantify the partitioning of Golgi protein between the organelle and ER, our values are similar to the cell fractionation results of Puri and Linstedt (17) for GM130 and considerably higher than those of the analytical microscopy results of Jokitalo et al. (18).

On a technical note, our data lead to the conclusion that the fluorescence of GalNAcT2 fused to a single GFP is a valid indicator of protein distribution. Moreover, we note that singly GFP-conjugated GalNAcT2 fluorescence, when analyzed by widefield microscopy using visually drawn Golgi perimeters, gave a Golgi value of only 65%. This is very similar to the value calculated by Zaal et al. (1) using the same approach for GalT concatenated to three GFP moieties. Given our distribution value for endogenous GalT coupled with the visually drawn distribution for GFP-labeled GalNAcT2, we conclude that the fluorescence yielded by triply concatenated GalT-GFP is likely a good indicator of Golgi protein distribution if best-practice analytical procedures are applied. This conclusion, however, remains to be tested experimentally. The agreement between electron microscopy and best-practice fluorescence microscopy confirms that fluorescence microscopy is an acceptable and flexible method for pursuing quantitative results. This is fortunate. Although quantification by electron microscopy is attractive because of the ease in identifying organelles at high resolution, based on our results, we conclude that the analysis of endogenous Golgi protein distribution to the ER by electron microscopy is virtually impossible due to the low level of its expression. The immunogold labeling of GalNAcT2-VSV that is overexpresed by fivefold in the cells used was only about threefold higher than background labeling.

In conclusion, our results provide a mutually consistent, multiapproach analysis of the distribution of Golgi glycosyltransferases between the organelle and ER. We conclude that this distribution is approximately 90:10 with the concentration of these proteins in the Golgi apparatus being about 100-fold greater than the ER. The HeLa cell Golgi apparatus has about 12% of the membrane area of the ER. We speculate this large difference in concentration results in a lack of Golgi glycosylation of ER proteins. We conclude that fluorescence microscopy when combined with best-practice analysis, i.e. treating the images as data arrays, computer-calculated organelle boundaries and deconvolution, is a valid and appropriate approach to quantifying protein distributions in cells. Hence, the automated, high-throughput analysis of protein distributions using fluorescent protein chimeras should be valid, and quantifiable analysis of transient effects is feasible. Finally, the most fundamental prediction of Golgi-resident protein cycling to ER, i.e. there is a measurable, but small ER pool, does indeed hold true. At steady state, Golgi-resident glycosyltransferases do partition between the Golgi apparatus and ER.

Materials and Methods

  1. Top of page
  2. Abstract
  3. Results
  4. Discussion
  5. Materials and Methods
  6. Acknowledgments
  7. References

Cell culture

Wild-type HeLa cells were cultured in Dulbecco's Modified Eagle Medium supplemented with 10% fetal bovine serum (Hyclone, Logan, UT, USA). HeLa cells stably expressing GalNAcT2-GFP and GalNAcT2-VSV were cultured in the presence of 0.45 mg/mL of geneticin sulfate (Sigma-Aldrich, St. Louis, MO, USA). One or two days before studies were initiated, transfected cells were co-plated with WT cells in medium without geneticin and grown to approximately 70% in either 100-mm polystyrene culture dishes containing 11-mm glass cover slips for fixation. For CHX treatments, cover slips from one 100-mm dish were transferred into two 35-mm dishes, one of which had 50 µg/mL of CHX, and both were incubated for 4 h before being fixed and stained.

Kinetic modeling

Images collected in an earlier publication from this laboratory (5) were analyzed. For FRAP experiments, GalNAcT2-GFP relative fluorescence over the ER for three cells was recorded after 35% photobleaching of the ER with a Zeiss LSM 510 laser-scanning microscope (Carl Zeiss, Jena, Germany). For ER exit block experiment, Sar1pdn protein was directly injected into the cytoplasm in the presence of 100 µg/mL CHX causing an acute block. Visually identifiable Golgi area was scored and averaged for 10 cells at each time point, and the decrease in area was obtained as a rate from an exponential fit using kaleidagraph 3.5 software (Synergy Software, Reading, PA, USA). Note that exact knowledge of the total ER/Golgi fluorescence intensity or further image enhancement through deconvolution was not required for these rate measurements.

Electron microscopy

Electron microscope images collected in an earlier publication from this laboratory (3) were analyzed. Labeling densities of GalNAcT2 (10 nm gold) over Golgi cisternae, tubules, ER and mitochondria were tabulated for these immunogold-labeled images (34 000 magnification, n = 15). Note that at 34 000 magnification, the area between the individual cisternae can be readily distinguished. The area of each organelle was determined using stereology (3,24). The particle density (number/µm2) of each organelle was calculated by dividing the gold particle count (number) by the relative area (µm2). The labeling density of the cisternae was calculated with (block Golgi) and without inclusion of the area between the cisternae.

The relative areas of the Golgi and ER were found by stereology using randomly photographed electron microscope images at 10 000 and 16 000 magnification. A 0.5 × 0.5 mm grid was used for all images (n = 33, 10 000 microscope magnification; n = 61, 16 000 microscope magnification). The grid was overlaid for point counting on images printed with a laser printer to fill the margins of a US letter paper sheet. Note that at the lower magnification, the area between Golgi cisternae could not be distinguished. Therefore, the Golgi area measured at the lower magnifications included both the cisternae and the area between (block Golgi). In the absence of an immunogold marker, we did not attempt to identify Golgi tubular areas separately but rather they were calculated based on the ratio of the tubular and block Golgi area found in the immunogold-labeled micrographs (data not shown). Cisternal area was calculated from block Golgi area by correcting for the area between Golgi cisterna found with the 34 000 microscope magnification images. The relative protein distribution for GalNAcT2 labeling between the Golgi stack and associated tubules and ER was determined by multiplying the protein densities determined from the 34 000 microscope magnification images by the relative organelle areas measured and calculated using the lower-magnification, randomly photographed images (see Table 1). The combined contributions from the cisternae and tubules yield the overall raw Golgi apparatus labeling. Raw organelle labeling was corrected for non-specific labeling as indicated by mitochondrially and nuclear-associated 10-nm gold particles.

Conventional widefield microscopy and spinning-disk confocal microscopy

For light microscopy, cells cultured on cover slips were fixed with 3% formaldehyde in PBS and permeabilized with 0.1% saponin. Fish skin gelatin was used as blocking reagent in all steps (12). Wild-type or GalNAcT2-VSV-expressing HeLa cells were stained with monoclonal antibodies against endogenous GalT (a gift from Dr Tommy Nilsson, Cell Biology and Biophysics Programme, EMBL-Heidelberg) at 1:10 dilution or affinity-purified rabbit polyclonal antibodies directed against the VSV-G epitope (19) at 1:1000 dilution and Cy3-conjugated donkey anti-rabbit or mouse immunoglobulin G antibodies (Jackson ImmunoResearch Laboratories, West Grove, PA, USA) at 1:1000 dilution were used as second antibodies. The cover slips were mounted in buffered solution of the water-soluble plastic, Mowiol.

Single plane widefield images were taken using a Zeiss Axiovert 200M microscope with 63 × 1.4 NA objective, ×1.6 optovar and a Roper CoolSNAP HQ CCD camera (RoperPhotometrics, Tucson, AZ, USA). The camera was operated at 1 × 1 binning. Spinning-disk confocal image stacks were obtained using a CARV accessory (Atto Bioscience, Rockville, MD, USA) mounted to the sideport of the Zeiss Axiovert 200M microscope. A 100×, 1.4 NA objective was used and images were captured to a Retiga EXi camera (QImaging, Burnaby, British Columbia, Canada). Illumination was with an X-Cite 120 light source (Hg-halide lamp, Exfo Life Sciences, Mississauga, Ontario, Canada). Nyquist oversampling criteria were met at 100-nm pixel size for single plane widefield images and a voxel size of 72 × 72 × 150 nm for spinning-disk confocal image stacks. Each image in the confocal stack was taken at 1.5-second exposure time and a camera gain of 400 for endogenous GalT and 1000 for GalNAcT2-VSV. Fluorescence bleaching while capturing a typical stack consisting of 40 slices was <3% based on capturing an equivalent stack a second time. All images were acquired as 12-bit data utilizing most of the available 4096 grayscales. Image capture software was iplab 3.9 (Scanalytics, Fairfax, VA, USA) for Mac OS X.

Laser-scanning confocal microscopy

GalNAcT2-VSV cells were co-plated with WT cells and fixed on glass cover slips and antibody stained as described for conventional microscopy. Images were taken at the Keck Imaging Center, Virginia Polytechnic Institute and State University, Blacksburg, VA, USA, on a Zeiss LSM 510 microscope using 543-nm laser excitation with pinhole diameters set at 1 Airy unit. A 63×, 1.4-NA objective was used with a zoom factor of 4.5 to achieve a voxel size of 60 × 60 × 150 nm. Eight-bit single pass scanning with pixel dwell time of 2.56 µs was used to minimize photobleaching.

Fluorescence image processing and analysis

Images were analyzed for intensity as raw and deconvolved data arrays. Raw spinning-disk image was shading corrected. A 16-frame average of a blank field image was subtracted from the spinning-disk images with an added offset value of 128 to avoid zero-intensity pixels. The raw images were then deconvolved with huygens essential 2.7 software (Scientific Volume Imaging, Hilversum, the Netherlands). An iterated maximum likelihood estimation algorithm was used. To avoid amplification of specking, we limited deconvolution iteration to five for widefield GalNAcT2-VSV images and 10 for other widefield images or spinning-disk confocal VSV images. All other spinning-disk or laser-scanning confocal images were deconvolved for the number of iterations required to meet the preset Huygens software quality criterion.

Cell boundaries were outlined visually with gamma = 0.4 to show dimmer structures. The total fluorescence intensity of the pixels within the boundary was then summed using iplab software and corrected for the mean WT-cell intensity or noncell background in the case of endogenous GalT. Wild-type cells were, with the exception of GalT, intermixed in the image field with the tagged cells. A visual Golgi apparatus threshold was drawn with the iplab software segmentation-drawing tool and determined as an intensity value by sliding the selection bar in the segmentation menu of iplab to flood the outlined area. A calculated Golgi apparatus threshold was determined using the mean ER intensity (average value in a region from the juxtanuclear Golgi apparatus) + 2SD of the intensity for that pixel region. Pixels within the cell boundary were then tested for brightness over a (visual or calculated) threshold value, and the clusters of these bright pixels larger than a cutoff area (1.0 µm2 for widefield images and 0.5 µm2 for slices of confocal sets) were considered as Golgi pixels. The total intensity of the Golgi pixels was summed and corrected for the mean ER value. In short, the Golgi fraction was calculated by the following equation:

  • image

For confocal image stacks, the mean and standard deviation of the ER varied little between different slices within a stack. The mean of those numbers was used to calculate the Golgi threshold. Surface plots were done with imagej software (freeware, Research Resources, NIH, Bethesda, MD, USA) in most cases. If not, iplab was used.


  1. Top of page
  2. Abstract
  3. Results
  4. Discussion
  5. Materials and Methods
  6. Acknowledgments
  7. References

We express our appreciation to Anja Habermann and Gareth Griffiths, EMBL-Heidelberg, for cryosectioning and discussions on stereology. Kristi DeCourcy hosted our use of the Zeiss LSM510 microscope at Virginia Polytechnic Institute and State University. We appreciate help and input from Hans van der Voort and Gitta Himel at Scientific Volume Imaging and Jim Paladino at BioVision Technologies on deconvolution. We acknowledge helpful discussions with the Lupashin and Baldini laboratories at the University of Arkansas for Medical Sciences. This work was supported in part by a grant from the National Institutes of Health, GM65233.


  1. Top of page
  2. Abstract
  3. Results
  4. Discussion
  5. Materials and Methods
  6. Acknowledgments
  7. References
  • 1
    Zaal KJ, Smith CL, Polishchuk RS, Altan N, Cole NB, Ellenberg J, Hirschberg K, Presley JF, Roberts TH, Siggia E, Phair RD, Lippincott-Schwartz J. Golgi membranes are absorbed into and reemerge from the ER during mitosis. Cell 1999;99: 589601.
  • 2
    Ward TH, Polishchuk RS, Caplan S, Hirschberg K, Lippincott-Schwartz J. Maintenance of Golgi structure and function depends on the integrity of ER export. J Cell Biol 2001;155: 557570.
  • 3
    Storrie B, White J, Rottger S, Stelzer EH, Suganuma T, Nilsson T. Recycling of golgi-resident glycosyltransferases through the ER reveals a novel pathway and provides an explanation for nocodazole-induced Golgi scattering. J Cell Biol 1998;143: 15051521.
  • 4
    Seemann J, Jokitalo E, Pypaert M, Warren G. Matrix proteins can generate the higher order architecture of the Golgi apparatus. Nature 2000;407: 10221026.
  • 5
    Miles S, McManus H, Forsten KE, Storrie B. Evidence that the entire Golgi apparatus cycles in interphase HeLa cells: sensitivity of Golgi matrix proteins to an ER exit block. J Cell Biol 2001;155: 543555.
  • 6
    Girod A, Storrie B, Simpson JC, Johannes L, Goud B, Roberts LM, Lord JM, Nilsson T, Pepperkok R. Evidence for a COP-I-independent transport route from the Golgi complex to the endoplasmic reticulum. Nat Cell Biol 1999;1: 423430.
  • 7
    Chen A, AbuJarour RJ, Draper RK. Evidence that the transport of ricin to the cytoplasm is independent of both Rab6A and COPI. J Cell Sci 2003;116: 35033510.
  • 8
    Young J, Stauber T, del Nery E, Vernos I, Pepperkok R, Nilsson T. Regulation of microtubule-dependent recycling at the trans-golgi network by Rab6A and Rab6A′. Mol Biol Cell 2005;16: 162177.
  • 9
    White J, Johannes L, Mallard F, Girod A, Grill S, Reinsch S, Keller P, Tzschaschel B, Echard A, Goud B, Stelzer EH. Rab6 coordinates a novel Golgi to ER retrograde transport pathway in live cells. J Cell Biol 1999;147: 743760.
  • 10
    Valsdottir R, Hashimoto H, Ashman K, Koda T, Storrie B, Nilsson T. Identification of rabaptin-5, rabex-5, and GM130 as putative effectors of rab33b, a regulator of retrograde traffic between the Golgi apparatus and ER. FEBS Lett 2001;508: 201209.
  • 11
    Martinez O, Antony C, Pehau-Arnaudet G, Berger EG, Salamero J, Goud B. GTP-bound forms of rab6 induce the redistribution of Golgi proteins into the endoplasmic reticulum. Proc Natl Acad Sci USA 1997;94: 18281833.
  • 12
    Jiang S, Storrie B. Cisternal rab proteins regulate Golgi apparatus redistribution in response to hypotonic stress. Mol Biol Cell 2005;16: 25862596.
  • 13
    Lippincott-Schwartz J, Yuan LC, Bonifacino JS, Klausner RD. Rapid redistribution of Golgi proteins into the ER in cells treated with brefeldin A: evidence for membrane cycling from Golgi to ER. Cell 1989;56: 801813.
  • 14
    Storrie B, Pepperkok R, Nilsson T. Breaking the COPI monopoly on Golgi recycling. Trends Cell Biol 2000;10: 385391.
  • 15
    Storrie B. Maintenance of Golgi apparatus structure in the face of continuous protein recycling to the endoplasmic reticulum: making ends meet. Int Rev Cytol 2005;244: 7196.
  • 16
    Kornfeld R, Kornfeld S. Assembly of asparagine-linked oligosaccharides. Annu Rev Biochem 1985;54: 631664.
  • 17
    Puri S, Linstedt AD. Capacity of the Golgi apparatus for biogenesis from the endoplasmic reticulum. Mol Biol Cell 2003;14: 50115018.
  • 18
    Jokitalo E, Cabrera-Poch N, Warren G, Shima DT. Golgi clusters and vesicles mediate mitotic inheritance independently of the endoplasmic reticulum. J Cell Biol 2001;154: 317330.
  • 19
    Röttger S, White J, Wandall HH, Olivo JC, Stark A, Bennett EP, Whitehouse C, Berger EG, Clausen H, Nilsson T. Localization of three human polypeptide GalNAc-transferases in HeLa cells suggests initiation of O-linked glycosylation throughout the Golgi apparatus. J Cell Sci 1998;111 (Pt 1):4560.
  • 20
    Swedlow JR. Quantitative fluorescence microscopy and image deconvolution. Methods Cell Biol 2003;72: 349367.
  • 21
    Murphy DB. Fundamentals of Light Microscopy and Electronic Imaging. New York: John Wiley & Sons; 2001.
  • 22
    Axelsson MA, Warren G. Rapid, endoplasmic reticulum-independent diffusion of the mitotic Golgi haze. Mol Biol Cell 2004;15: 18431852.
  • 23
    Farmaki T, Ponnambalam S, Prescott AR, Clausen H, Tang BL, Hong W, Lucocq JM. Forward and retrograde trafficking in mitotic animal cells. ER-Golgi transport arrest restricts protein export from the ER into COPII-coated structures. J Cell Sci 1999;112: 589600.
  • 24
    Griffiths G.Fine structure immunocytochemistry.Berlin: Springer-Verlag; 1993.