The endoplasmic reticulum (ER) network is a key membrane-bound organelle in eukaryotic cells, which plays an important role in processes such as biosynthesis of secretory and membrane proteins, lipid synthesis, and Ca²⁺ storage [1]. ER continuously undergoes reorganization during differentiation and the cell cycle, which is crucial for structure maintenance and organelle functions. In this process, new ER tubules extend from existing ones and contact with other ER tubules, or old ER tubules shrink, and then new junctions are formed [2].

The ER network spreads throughout the whole cell and regulates inter-organelle communications [3]. In addition, the highly dynamic rearrangements of the ER are strongly dependent on interactions with microtubules (MTs), which are assumed to synergistically execute various physiological functions [4]. Thus, quantitative and accurate analysis of the ER reorganization and formation is important for the study of the interaction mechanisms between the ER and other compositions within cells. However, quantification techniques which are able to capture subtle alterations of morphological structures during ER rearrangements are lacking.

In this study, we proposed a multi-parametric, quantitative method combining pixel-wise orientation and waviness to characterize the morphological structure of the ER and MT. We applied this quantification method to the time-dependent, super-resolution images of co-labeled ER and MT from U2OS cells acquired from two-dimensional structured illumination microscopy (2D SIM, as detailed in Supplement 1). Morphological quantification of orientation and waviness was applied to the time-dependent ER-MT images and both waviness value itself and variation of waviness were proved to be sensitive indicators for ER reorganization. Finally, a new mechanism of MT-associated ER formation, named hooking, was established based on the distinct quantification output from other known mechanisms.

U2OS cells were seeded in glass bottom dishes and cultured at 37°C with 5% CO₂ for 12 h. For MT labeling, we followed a previously described protocol [5], in which the cells were co-incubated with 4 μM PV-1 and 5 μM tubulin-Atto 488 at 37°C for 1 h, then washed three times with growth medium before imaging. For ER labeling, cells were transfected with SNAP-Sec61β using Lipofectamine 3000 (Thermo Fisher Scientific, Inc.) with standard protocol and cultured at 37°C with 5% CO₂ for an additional 24 h. Then cells were incubated with 1 µM SiR-SNAP (New England Biolabs, Inc.) at 37°C for 1 h, and then washed three times with growth medium before imaging. The super-resolution images of MT and ER were acquired by 2D SIM from 5 dishes. In addition to the images acquired from U2OS cells using 2D SIM, we performed analysis to validate waviness characteristics and MT-associated ER formation mechanisms relying on additional published data from COS-7, U2OS, and newt lung epithelial cells using other imaging techniques, including grazing incidence structured illumination microscopy (GI-SIM) [6], and confocal [2] and wide-field microscopy [7].

A representative 2D SIM image of co-labeled ER (cyan) and MT (magenta) is shown in Fig. 1(a). Fine structures were captured in a high-contrast manner. For comparison, part of the image was replaced by wide-field signals, which revealed that SIM led to clearly resolved cellular cytoskeleton and organelle. We extracted the ER channel alone to demonstrate the quantification of orientation and waviness [Fig. 1(b)]. Pixel-wise orientation was calculated using a weighted vector summation algorithm that we have described previously [8]. To calculate the orientation of a certain pixel, vectors were defined and weighted by their length and intensity variations along each direction [Fig. 1(c), top], with the orientation of the pixel defined as the direction of summed vectors [Fig. 1(c), bottom]. A pixel-wise orientation matrix was then obtained and mapped by different colors according to orientation values [Fig. 1(e)].

 

Fig. 1. Schematic of orientation and waviness quantification for ER analysis. (a) Super-resolution image of U2OS cell acquired by 2D SIM. Scale bar, 2 μm. (b) Representative 2D SIM image of ER. Scale bar, 2 μm. (c) Brief schematic of the algorithm to calculate pixel-wise orientation. (d) Top, localized window of pixel-wise orientation (arrows representing orientation). Bottom, window with orientation difference ${\delta _n}$ at each non-central ER pixel (unit, degree). (e), (f) Color-coded maps of orientation and waviness ofER.

Download Full Size | PDF

Waviness is a metric quantifying orientation variation in a localized region. To define this metric, image thresholding was first applied to the 2D SIM image to define effective ER pixels (Supplement 1, Fig. S2) and a localized window containing orientation information for effective pixels was generated [Fig. 1(d) top, with each arrow indicating the orientation of each pixel (detailed orientation values shown in Supplement 1, Fig. S3)]. According to the flowchart of waviness calculation (Supplement 1, Fig. S4), the difference ${\delta _n}$ between the orientation of each non-central ER pixel ${\theta _n}$ and that of the central ER pixel ${\theta _c}$ was then calculated [marked in each non-central ER pixels in Fig. 1(d) bottom]. Totally, N orientation differences ${\delta _n}$ were obtained, where N is the number of non-central ER pixels [e.g., 50 for the case in Fig. 1(d)]. For axial data (i.e., fiber-like data), ${\delta _n}$ needed to be modified when it was over 90° or negative (Supplement 1, Fig. S4). Finally, waviness of the central pixel in the window was obtained as the mean of the orientation difference ${\delta _n}$ divided by 90 (Supplement 1, Fig. S4). In this manner, waviness was calculated for all the ER pixels with a normalized value ranging from 0 to 1, where a higher value corresponded to more drastic variation in orientation. A window size two to three times the diameter of the ER structure led to optimal waviness results, as detailed in Supplement 1. The proposed waviness algorithm was more suitable for fiber-like data instead of circular data. A color map was generated accordingly to demonstrate waviness level intuitively [Fig. 1(f)], where waviness of ER junctions (marked by top red dot rectangle) was higher than that of isolated straight ER tubules (marked by bottom blue dot rectangle), which indicated a higher level of orientation variation in the former area, consistent with the hues shown in the orientation map [Fig. 1(e)].

Time-dependent ER images offered an opportunity to assess the waviness change caused by ER remodeling. We found that there were some active regions with the intense ER network remodeling [Fig. 2(b)], accompanied by a violently fluctuated waviness value [Fig. 2(a) left], in contrast to inactive ones where few ER extensions were visualized [Fig. 2(a) right], which corresponded to slight changes of waviness [Fig. 2(b)]. We defined active regions as regions with at least one ER tubule branch extending or shrinking no less than 1 μm, with the occurrence of ER remodeling as the basic principle for defining an active or inactive region, while independent of waviness. We found a significantly higher level of time-dependent variation in waviness from active regions [Fig. 2(c)] as a result of the morphological alterations caused by ER rearrangements. Furthermore, these active regions typically exhibited a higher level of waviness than inactive ones [Fig. 2(d)], which made the relative waviness value a potential indicator for active regions with ER remodeling. As described above, a low waviness level often corresponded to long straight and isolated ER tubules [Fig. 2(a), right], while a high waviness level often indicated junctions within a localized region where short, curved ER tubules with different orientations crossed to form a dense reticular structure [Fig. 2(a), left], which made membrane fusion easier for ER reshaping [9]. Interestingly, we found that active regions might turn quiet when assessed over a relatively longer time scale [Fig. 3(a)], with time-dependent waviness features of the same region for two distinct time scales being greatly different [Fig. 3(b)]. Waviness was relatively higher for time scale 1 (earlier) owing to more complex junctions and varied more intensely caused by ER shrinking. For time scale 2 (later), the ER was altered to a relatively stable state, which led to a low valued and slightly changing waviness level. These results indicated that both waviness itself and its variation were sensitive to morphological alterations caused by ER remodeling.

 

Fig. 2. Waviness analysis of time-dependent ER reorganization. (a) Waviness maps from two representative regions with active or inactive ER reshaping. Scale bar, 1 μm. (b) Time-dependent waviness profiles at these two regions (top curve, active region and bottom curve, inactive region). (c) Time-dependent variation and (d) mean of waviness at active and inactive regions. Nine representative regions for each case from four U2OS and COS-7 cells are analyzed, as detailed in Supplement 1, Table S1. *, p < 0.05; ***, p < 0.001.

Download Full Size | PDF

 

Fig. 3. ER waviness assessment of the same region at distinct time scales. (a) Time-lapse waviness maps for a certain region. Scale bar, 0.5 μm. (b) Waviness profile of this representative region as a function of time.

Download Full Size | PDF

ER formation was strongly dependent on direct interactions with the MT and, by now, two main mechanisms of MT-associated ER formation were recognized including sliding and tip attachment complex (TAC). In the way of sliding (left column 1 in Fig. 4), new ER tubules were pulled out of the existing ER membrane by associating with motor proteins and then extending along MT. For the case of TAC, ER tubules branching could occur by attaching to the tips of the MT, and the TAC mechanism could be further classified as polymerizing TAC (pTAC, left column 2 in Fig. 4) and depolymerizing TAC (dTAC, left column 3 in Fig. 4) depending on the polymerizing or depolymerizing plus ends of the MT to which ER tubules were attaching [6]. Because the plus-end-binding proteins of polymerizing and depolymerizing MTs were very different [10], the dTAC mechanism might be mediated by molecules other than the STIM1 and EB1 mediators of the pTAC mechanism [11]. In addition to these three mechanisms, we discovered a fourth mechanism of ER formation relying on apparent direct interactions with the MT (Fig. 4, right). In this mechanism, one end of the growing ER tubule kept hooking to the MT, with the other end appearing as an ER branching point (Fig. 4, right). The growing ER tubule typically aligned almost perpendicular to the MT, potentially indicating that this mechanism was mediated by the force generated at the contact site, where the spastin M1 variant, with an amino-terminal extension containing one hairpin sequence, could connect the organelle and cytoskeleton [12].

 

Fig. 4. Representative time-dependent images of MT-associated ER formation mechanisms. Left side of dotted line: sliding; pTAC; and dTAC mechanisms. Right side: time-dependent intensity image; ER orientation; MT orientation; ER waviness; and MT waviness maps of the hooking mechanism. Scale bar, 1 μm.

Download Full Size | PDF

To validate that the hooking mechanism was independent of the other three, we explored the time-dependent evolution of difference between the ER and MT in both orientation and waviness for the cases of different mechanisms shown in Fig. 4. For each mechanism, only the region very close to the contact site in either the ER or MT was considered. As can be seen from the plots that describe the time-dependent MT-ER difference in waviness [DW, Fig. 5(a)] and difference in orientation [DO, Fig. 5(b)], there was a dramatic rise at later time points of the sliding mechanism in both DW and DO. Both pTAC and dTAC maintained a relatively low level in DW throughout the time assessed, while dTAC exhibited a higher level in DO than pTAC, even with a rise at later time points when the growing ER tubule fused with an existing one. In contrast to these three mechanisms, the hooking mechanism maintained at a much higher level in both DW and DO, as detailed in Supplement 1, Table S2. Based on these observations, we generated three optical biomarkers for classification of these mechanisms, including the mean of the difference in orientation [MDO, Fig. 5(c) left], mean of the difference in waviness [MDW, Fig. 5(c) middle], and variation of the difference in waviness [VDW, Fig. 5(c) right], as detailed in Supplement 1. There were obvious differences between hooking and the other mechanisms in MDW and MDO. A possible reason for this was that the contact sites were continuously moving in sliding and TAC, which in turn demanded the co-aligning between the ER and MT and made relatively slight differences between their orientation and morphology (waviness) during ER formation. However, for the hooking mechanism, the contact sites were almost stable and enabled ER to “hang” on the MT and grow (Fig. 4 right), which caused the dramatic alteration in orientation and morphology of the ER relative to the MT.

 

Fig. 5. Quantitative analysis and classification of MT-associated ER formation mechanisms. Plot showing (a) DW and (b) DO between ER and MT as a function of time for these four mechanisms. (c) Classification of these mechanisms using mean of the difference in orientation (MDO, left), mean of the difference in waviness (MDW, middle), and variation of the difference in waviness (VDW, right) between ER and MT. Totally 26 ER formation events from 15 cells are analyzed, including 8 for sliding, and 6 for pTAC, dTAC, and hooking from different cell types, as detailed in Table S3. **, p < 0.01; ***, p < 0.001. (d) 3D scatterplot using a combination of MDO, MDW, and VDW. (e) Classification accuracy results.

Download Full Size | PDF

We performed canonical linear discriminant analysis using SPSS with MDO, MDW, and VDW for four mechanisms of MT-associated ER formation, with original classification accuracy (OCA) and cross-validated classification accuracy (CVCA) acquired respectively based on the entire and the leave-one-out cross-validation data set, as detailed in Supplement 1. The OCA and CVCA values ranged between 69.2% and 88.5% when a single biomarker was used [Fig. 5(e)]. To take advantage of the informational complementarity from all of these biomarkers, a combination of them was applied, and an accuracy of 100% was achieved in classifying these four ER formation mechanisms in both OCA and CVCA [Figs. 5(d), 5(e)]. These results indicated that the hooking mechanism was an independent one that had unique properties in relative orientation and waviness between the ER and MT compared with the mechanisms recognized.

In conclusion, we demonstrate a quantitative method combining orientation and waviness to analyze the time-dependent images of co-labeled ER and MT. Owing to the capability of this technique to provide highly sensitive and automated measurement of morphological characteristics at the pixel-wise level, subtle alterations of the ER structure could be captured and quantified, enabling a better understanding of ER rearrangements. A new mechanism of MT-associated ER formation has been proposed based on distinct morphological changes resulting from MT–ER interactions, different from sliding and TAC mechanisms. Numerical differences in organization alterations of four mechanisms during ER formation are validated, which could hardly be extracted without quantitative analysis.