Single-Cell Quantification of the Transition Temperature of Intracellular Elastin-like Polypeptides

Abstract

Elastin-like polypeptides (ELPs) are modular, stimuli-responsive materials that self-assemble into protein-rich microdomains in response to heating. By cloning ELPs to effector proteins, expressed intracellular fusions can even modulate cellular pathways. A critical step in engineering these fusions is to determine and control their intracellular phase transition temperature (T_t). To do so, this Method paper describes a simple live-cell imaging technique to estimate the T_t of non-fluorescent ELP fusion proteins by co-transfection with a fluorescent ELP marker. Intracellular microdomain formation can then be visualized in live cells through the co-assembly of the non-fluorescent and fluorescent ELP fusion proteins. If the two ELP fusions have different T_t, the intracellular ELP mixture phase separates at the temperature corresponding to the fusion with the lower T_t. In addition, co-assembled ELP microdomains often exhibit pronounced differences in size or number, compared to single transfected treatments. These features enable live-cell imaging experiments and image analysis to determine the intracellular T_t of a library of related ELP fusions. As a case study, we employ the recently reported Caveolin1-ELP library (CAV1-ELPs). In addition to providing a detailed protocol, we also report the development of a useful FIJI plugin named SIAL (Simple Image Analysis Library), which contains programs for image randomization and blinding, phenotype scoring, and ROI selection. These tasks are important parts of the protocol detailed here and are also commonly employed in other image analysis workflows.

Graphical Abstract

INTRODUCTION

Cells rely on the selective formation and disassembly of proteins to achieve desired outcomes. For example, RNA transcription, endocytosis, and signaling cascades are all accomplished by the selective assembly and disassembly of proteins in response to stimuli.^1–3 Accordingly, life scientists routinely study biological systems by inhibiting or activating protein assembly. In these assays, the goal is to selectively and precisely affect the protein of interest and any cellular pathways in which it participates.

Common tools and approaches for altering protein activity include small molecules and genetic silencing methods, such as siRNA and gene editing. Although much good work has been accomplished with these tools, small molecules often display widespread off-target effects, limited reversibility, and poor spatiotemporal resolution, owing to their passive diffusion across the cell membrane.^4–7 Genetic silencing methods may also suffer from off-target activity and, in addition, are often either irreversible or require hours to days for effect, thus allowing compensatory cellular pathways to be activated.^8–11 In both cases, the side-effects of these methods may confound experimental results.

To circumvent these limitations, control over a protein’s function can be handled more directly by creating a stimuli-responsive mutant of that protein. In the absence of external stimuli, the mutant protein would behave like its wild-type counterpart, while in response to external stimuli, the mutant version would rapidly inactivate or activate. With this rationale in mind, many investigators have designed or discovered proteins that respond to external stimuli, including chemicals, light, magnetic fields, and heat.^12–22

One such class of stimuli-responsive proteins are the elastin-like polypeptides (ELPs). These thermally responsive peptides are biocompatible and biodegradable and are constructed out of a pentameric amino acid repeat of (VPGXG)_n. In this modular design, the X residue can be selected to alter ELP hydrophobicity, and the number of repeats, n, determines the ELP molecular weight. In response to heat, ELPs can rapidly (within seconds to minutes) phase-separate and self-assemble into protein-rich domains, forming a secondary aqueous phase known as a coacervate. If the ELPs are intracellularly assembled, these coacervates appear as “microdomains”. Individual microdomains are typically several hundred nanometers to several micrometers in diameter. ELP phase separation is a thermodynamically reversible process; therefore, ELP coacervates quickly resolubilize into the cytoplasm upon cooling. The temperature of self-assembly, termed the transition temperature (T_t), is directly encoded in the ELP sequence, because it is determined largely by the hydrophobicity of the guest residue and the ELP molecular weight.^23–25 Importantly, these stimuli-responsive characteristics are retained when ELPs are fused to an effector protein, which enables them to modulate cellular pathways in which they participate. Based on this approach, we have previously reported that thermo-responsive ELP fusion of clathrin light chain can be used to reversibly inhibit clathrin-mediated endocytosis.¹⁷ More recently, we fused ELPs to epidermal growth factor receptor (EGFR-ELPs) to control epidermal growth factor signaling and also to Caveolin 1 (CAV1-ELP) to control caveolin-mediated endocytosis via the formation of microdomains^16,21

A critically important step in engineering any ELP fusion protein is the determination of its intracellular T_t. This task cannot be directly accomplished in live cells, because ELP fusions are typically not fluorescent or otherwise directly visible (unless an ELP is fused to a fluorescent reporter protein). An alternative approach could involve incubating ELP-transfected cells at select temperatures followed by fixation and indirect staining with antibodies to determine if the ELPs have formed microdomains. In practice, however, only a limited number of different temperatures can be selected before this approach becomes unfeasible. Accordingly, the precise T_t in each individual cell is unlikely to be determined with this approach. Lastly, one could clone a fluorescent reporter onto their ELP fusion protein. Although this method would permit direct visualization of intracellular ELP self-assembly, the resulting fluorescent-ELP-fusion protein would likely be several times larger than the wild-type protein. The creation of such a large genetically modified fusion protein raises concerns: (1) the behavior of this fusion protein may be very different from that of the much smaller wild-type protein (without ELP or fluorescent reporter); (2) the T_t of this fusion protein may be different from that of the non-fluorescent ELP fusion protein. Thus, we recommend that a combination of strategies be used to characterize the T_t, such as a live-cell imaging approach as well as a fixed-cell approach under controlled incubation temperatures.

An ideal method would permit visual confirmation of intracellular ELP fusion protein assembly in live cells, in real time, as they are subjected to a temperature ramp. Although direct visual confirmation of non-fluorescent ELP fusion protein self-assembly in live cells is not possible, indirect visual confirmation can be achieved by exploiting the fact that ELPs co-assemble with other ELP species. Thus, by co-transfecting cells with a non-fluorescent ELP fusion protein and a secondary fluorescent ELP fusion protein, assembly can be visualized in live cells through the co-assembly of the non-fluorescent and fluorescent ELP fusions. Importantly, if the two different ELP species exhibit differences in temperature sensitivity, the resulting mixture of ELPs will exhibit a decreased T_t that corresponds to the ELP with a lower T_t (Figure 1). In addition to changes in T_t, co-assembled ELP microdomains may also exhibit pronounced differences in size or number, compared to single transfected treatments (see Figures 4–6). These features make it possible to employ relatively simple live-cell imaging experiments and image analysis to determine the intracellular T_t of a given ELP fusion library. Moreover, this strategy reveals a range of behaviors within individual cells that form microdomains at slightly different temperatures.

Figure 1.

Intracellular T_t of non-fluorescent ELP fusion proteins can be visually determined in live cells through co-assembly with a fluorescent ELP fusion protein. CAV1 = Caveolin 1. (A) An ELP fusion protein like CAV1-V96 is not fluorescent, so its phase separation is not visible in live cells. (B) In contrast, GFP-V60 is fluorescent, and its assembly is visible in live cells. In addition, GFP-V60 phase separates at a higher temperature than CAV1-V96. (C) In cells transfected with both constructs, CAV1-V96 will phase transition and co-assemble with GFP-V60. The phase separation occurs at a lower temperature than in cells transfected with only GFP-V60. This lower temperature is the estimated intracellular T_t of CAV1-V96.

Figure 4.

Cells co-transfected with CAV1-V96 + GFP-V60 exhibit reduced transition temperatures and enlarged microdomains compared to transfection with GFP-V60 alone. Time-lapse images were taken from live cells undergoing a temperature ramp. Non-fluorescent CAV1-V96 phase separates over this temperature range, which can be visualized using co-transfected fluorescent GFP-V60.²¹ In contrast, cells transfected only with GFP-V60 phase separate at a higher temperature and produce cytosolic microdomains that are smaller and more numerous than the membrane anchored microdomains of CAV1-V96 + GFP-V60. CAV1-A96 (middle column) is insensitive to these temperatures, nor does it produce enlarged microdomains characteristic of CAV1-V96 + GFP-V60. Similarly, phase separation of GFP-V60 alone is distributed in many puncta through the cytosol. Although the only visible fluorescence is from GFP-V60, the downward shift of transition temperature combined with the dramatic differences in microdomain size, enables indirect, single-cell detection of phase separation of expressed ELP fusions. Scale bar is 10 μm.

Figure 6.

Indirect staining in fixed cells confirms that CAV1-V96 co-assembles with GFP-V60. Cells were transfected with either CAV1-V96 + GFP-V60 (top row) or GFP-V60 alone (bottom row). 72 h after transfection, cells were pre-incubated for 50 min at 4 °C and were then incubated at 37 °C for 50 min before fixation and immunostaining. In dual-transfected cells, CAV1-V96 self-assembles to form large intracellular microdomains which overlap (purple) with co-assembled GFP-V60. In contrast, GFP-V60 alone transfection assembles small numerous microdomains (green) throughout the cytosol and nucleus. CAV1-V96 was detected with an anti-myc antibody. GFP-V60 was detected with an anti-GFP antibody. Scale bar is 10 μm.

Although we have previously reported this approach to study ELP fusions of epidermal growth factor receptor and Caveolin 1 (i.e., EGFR-ELPs and CAV1-ELPs), we believe it would be useful to the community to describe the technique in more detail and provide suggestions for improvement.^16,21 As a case study, we used the recently developed CAV1-ELP library to analyze a new and unpublished dataset. Although this is a Methods paper, it reports new data regarding the behavior of temperature-insensitive CAV1-A96, which serves as an important control in studies using CAV1-ELPs. To assist with many of the common image analysis tasks that are involved in the described methods, we developed an easily installed FIJI plugin that can be used to randomize and blind imaging data, score phenotypes, and record regions of interest (ROIs) from images. These tasks are an important part of the methods described here but are also commonly employed in other image analysis workflows.

MATERIALS

Cell Culture

• HEK293T cells (#CRL-3216, ATCC, Manassas, VA)

• T75 cell culture flask (4616, Laguna Scientific, Laguna Niguel, CA)

• Trypsin-EDTA (0.05%) (25300054, Thermo Fisher Scientific, Waltham, MA)

• Dulbecco’s Modified Eagle Medium (DMEM) (11995065, Thermo Fisher Scientific, Waltham, MA)

• Fetal bovine serum (FBS) (#35-011-CV, Corning, NY)

• 1X dPBS (#25-508, Genesee, San Diego, CA)

Transfection

• Lipofectamine 3000 transfection reagent (L3000008, Life Technologies, Carlsbad, CA)

• Opti-MEM reduced serum medium (31985070, Thermo Fisher Scientific, Waltham, MA)

• Poly-d-lysine (P6407, Sigma-Aldrich, St. Louis, MO)

• 35 mm glass-bottom dishes (P35G-0-10-C, MatTek Corporation, Ashland, MA)

Live-Cell Imaging

• Live-cell imaging solution (A14291DJ, Thermo Fisher Scientific, Waltham, MA)

• LSM 880 (Carl Zeiss, Oberkochen, Germany)

• LD LCI Plan Apochromat 25×/0.8 numerical aperture objective for oil/water/glycerol/silicone immersion (420852-9871-000, Carl Zeiss, Oberkochen, Germany)

• Stage attachment Z PIEZO WSB 500 (D) (432339-9000-000, Carl Zeiss, Oberkochen, Germany)

• Stage insert Z PIEZO WSB 500 for Heating Inserts P S1/Mxx S1 (D) (432339-9050-000, Carl Zeiss, Oberkochen, Germany)

• Immersol W (444969-0000-000, Carl Zeiss, Oberkochen, Germany)

• Argon 488 laser line (Carl Zeiss, Oberkochen, Germany)

Temperature Control during Live-Cell Imaging

• Heating Insert P Lab-Tek S1 (#131-800 029, PeCon GmbH, Erbach, Germany)

• Control Sensor T S1 (#880-800242, PeCon GmbH, Erbach, Germany)

• AP07R-40 refrigerating/heating bath (89202-982, VWR, Radnor, PA)

Image Analysis

• FIJI version 2.0.0

• SIAL (a FIJI plugin for image randomization and blinding; can be installed by adding https://sites.imagej.net/D-tear/ to ImageJ Updater)

• Template Matching (an optional FIJI plugin for stabilizing lateral drift in time lapse images; can be installed by adding http://sites.imagej.net/Template_Matching/ to ImageJ Updater)

• measureStack.ijm (an optional ImageJ macro for automated measuring of multiple ROIs on an image stack; this macro can be found in the Supporting Information.)

• R and RStudio (optional; if pixel value standard deviation analysis is desired)

• batch_plot (an optional R function for analyzing pixel value standard deviation data; an R notebook of raw code and some example data are provided in the Supporting Information)

Fixed-Cell Imaging to Confirm GFP-V60 Co-localization with Non-fluorescent ELP Fusion

• LSM 800 (Carl Zeiss, Oberkochen, Germany)

• Plan Apochromat 63×/1.40 numerical aperture Oil DIC M27 objective (420782-9900-000, Carl Zeiss, Oberkochen, Germany)

• Immersol 518F (444970-9000-000, Carl Zeiss, Oberkochen, Germany)

• 640 nm (CAV1-V96), 488 nm (GFP-V60), and 405 nm (DAPI) laser lines

• 12-well cell culture plates (25–106, Genessee Scientific, San Diego, CA)

• VWR micro cover glasses, round, no. 1 (48380-046, VWR, Radnor, PA)

• Torrey Pines Echotherm IC25 fully programmable chilling/heating dry bath (IC25, Torrey Pines Scientific, Carlsbad, CA)

• 10X PBS dry pack (#MB1001, Biopioneer, San Diego, CA)

• Paraformaldehyde (PFA) (43368, Alfa Aesar, Tewksbury, MA)

• Ammonium chloride (NH₄Cl) (Alfa Aesar, Tewksbury, MA)

• Bovine serum albumin (BSA) (A9647, Sigma-Aldrich, St. Louis, MO)

• myc-tag mouse mAb (9B11 #2276S, Cell Signaling Technology, Danvers, MA) [1:1000 dilution]

• Rabbit GFP pAb antibody (ab290, Abcam, Cambridge, MA) [1:500 dilution]

• Chicken anti-mouse IgG (H+L) Cross-Adsorbed Secondary Antibody, Alexa Fluor 647 (#A-21463, ThermoFisher Scientific, Waltham, MA) [1:500 dilution]

• Donkey anti-rabbit IgG (H+L) Highly Cross-Adsorbed Secondary Antibody, Alexa Fluor 488 (#A-21206, ThermoFisher Scientific, Waltham, MA) [1:500 dilution]

• DAPI (4′,6-diamidino-2-phenylindole, dihydrochloride) (#D1306, ThermoFisher Scientific, Waltham, MA)

• ProLong glass antifade mountant (P36984, ThermoFisher Scientific, Waltham, MA)

METHODS

1. Cell Culture.

Figure 2 provides an overall schematic of the experimental procedure, beginning with cell culture and transfection. HEK293T cells were grown in T75 flasks with DMEM supplemented with 10% FBS in a humidified incubator with 5% CO₂ at 37 °C. During subculture, cells were passaged using trypsin-EDTA (0.05%) and washed with dPBS. In general, cells were maintained and subcultured under the guidelines recommended by American Type Culture Collection (ATCC).

Figure 2.

Stepwise schematic of the experimental procedure. HEK293T cells are passaged onto 35 mm glass bottom dishes. 24 h after passaging, when the cells have achieved ~70% confluency, they are transiently transfected with either GFP-V60 alone or dual transfected with GFP-V60 and another non-fluorescent ELP fusion protein. In this paper, the CAV1-ELP library is examined; however, this approach has also been applied to EGFR-ELPs.¹⁶ Theoretically, this approach can be applied to other ELP fusions. After a 72 h incubation at 30 °C and 5% CO₂, cell culture media is replaced with ice-cold live-cell imaging solution (A14291DJ, Thermo Fisher Scientific), a physiological medium optimized for short-term live-cell imaging. Cells are then immediately imaged during a temperature ramp from 15 to 60 °C to determine the temperature at which fluorescent GFP-V60 microdomains assemble. This temperature serves as the estimate of the intracellular transition temperature (T_t) of the co-transfected non-fluorescent ELP fusion protein.

2. Transfection.

2.1. Preparing MatTek Plates for Cell Culture.

To prepare the MatTek 35 mm glass-bottom dishes for adherent cell culture and subsequent transfection, a 0.1 mg/mL solution of poly-d-lysine was added to each MatTek dish. To make this solution, we followed the manufacturer’s suggested protocol and added 50 mL of sterile tissue culture grade water to 5 mg of poly-d-lysine. 2 mL of this solution was then applied, under sterile conditions, to each MatTek plate. After 5 min, the solution was removed through aspiration and gently rinsed with sterile dPBS.

Critical Points and Comments:

Poly-d-lysine improves cell attachment to the glass bottom of the MatTek dishes. An exact amount of 2 mL is not required; however, it is important that enough of the solution is applied to cover the bottom of the dish, especially the glass coverslip. When not in use, the poly-d-lysine solution should be stored at 4 °C. Also note that as long as aseptic conditions are used, the poly-d-lysine solution can be reused in future experiments with no discernible effect on cell attachment or health.

2.2. Cell Passaging onto MatTek Plates.

Before transfection, cells were subcultured from the T75 flask and resuspended in complete media at a density of 1 × 10⁶ cells/mL, and 2 mL of this solution was then placed into each poly-d-lysine-coated MatTek dish. Plated cells were then incubated for 24 h at 37 °C with 5% CO₂ before transient transfection 24 h later.

Critical Points and Comments:

Seeding densities should always be empirically determined, especially if another cell type is used. In our work, we aimed for at least 70% confluency 24 h after seeding.

2.3. Transfection.

Cells were transfected with the Lipofectamine 3000 kit. All transfection reagents and plasmid DNA were diluted in Opti-MEM Reduced Serum Medium. In preparing the lipofectamine master mix, we mimicked the suggested protocol for a single 6-well plate. The transfection mix for a single 35 mm dish involved 125 μL of the diluted DNA/P3000 mixture (containing 2.5 μg total of plasmid DNA, the P3000 Reagent at a concentration of 2 μg/μL of DNA, and Opti-MEM up to 125 μL) + 125 μL of the Lipofectamine mixture (120 μL of Opti-MEM + 5 μL of Lipofectamine 3000). Each solution was thoroughly mixed before being combined to create the 250 μL “master mix”. This master mix was then incubated for 15 min at room temperature. During this time, the cell seeded MatTek dishes were removed from the incubator, washed one time with room temperature dPBS, and were then replenished with 2 mL of Opti-MEM. After the 15 min incubation, the 250 μL master mix was pipetted drop-by-drop into the 2 mL of Opti-MEM media. Cells were then placed in a 30 °C incubator with 5% CO₂ and were incubated for 72 h before subsequent imaging. This lower temperature was chosen to help inhibit premature ELP self-assembly as the exogenous proteins are expressed during the transfection incubation period.

For dual transfected treatments (CAV1-V96 + GFP-V60 and CAV1-A96 + GFP-V60), 1.25 ug of each DNA plasmid was added to the transfection reaction, for a total of 2.5 μg DNA per MatTek dish. In the single transfection treatment, 2.5 μg of GFP-V60 plasmid DNA was added to the transfection reaction. Each MatTek dish received either CAV1-V96 + GFP-V60, CAV1-A96 + GFP-V60, or GFP-V60 alone.

Critical Points and Comments:

In our experience transfection efficiencies are approximately 10%, and transfection efficiency is typically higher in the single transfected group compared to dual transfected groups. In hard-to-transfect cells, researchers may consider viral transfection of ELP constructs.²⁶

Also note that ideally the non-fluorescent ELP and fluorescent ELP species should exhibit expected differences in temperature sensitivity of at least a few degrees Celsius. This is because large differences in temperature sensitivity will provide greater power to detect significant differences in the T_t between the single and dual transfected groups. Relative differences in ELP temperature sensitivity can be predicted before imaging by comparing the hydrophobicity of guest residues and the number of pentameric repeats. For example, if a researcher believes their non-fluorescent ELP has a similar temperature sensitivity to GFP-V60, they could use GFP-V48 as their fluorescent reporter instead of GFP-V60. This is because GFP-V48 is less temperature sensitive (i.e., has a higher T_t) than GFP-V60. While we do not propose that GFP-V60 will be able to co-localize with all future intracellular ELPs of different compositions and transition temperatures, for this study we have validated the association between the fluorescent and non-fluorescent ELP fusion (see Figure 6).

3. Live-Cell Imaging.

Live-cell imaging of intracellular ELP assembly was accomplished with an LSM880 laser scanning confocal microscope equipped with an airy scan detector and definite focus module. As the cells were immersed in aqueous live-cell imaging solution during imaging, a plain apochromat 25×/0.8 numerical aperture water immersion objective was chosen for image acquisition. Immersol W was used as the immersion media, as it exhibits the same refractive index of water but does not evaporate easily during heating.

To excite GFP-V60, the 488-argon laser line was selected, and all images were acquired using the airy scan fast mode with the definite focus module activated. The laser power selected was 17% and laser light was passed through a 488 nm main beam splitter (MBS) before illuminating the cells. Excited light emitted from the cells was then passed through a dual bandpass (495 nm/550 nm) + long pass (>570 nm) filter before reaching the airy scan detector. Other imaging parameters include a 1.84 μs pixel dwell time, with a lateral sampling of 0.149 μm, no averaging, 625 master gain, 1.00 digital gain, and a total image size of 338.29 μm × 338.29 μm at 2272 pixel × 2272 pixel. All collected images were 16-bit. Using the Zeiss Black “Time Series” module, 90 images were acquired over the course of the temperature ramp, with image acquisition every 20 s. This corresponded to a total imaging time of 30 min.

Critical Points and Comments:

In order to properly compare different treatments, the exact same imaging settings should be applied to all images in a given study during the course of that study (i.e., all images from all treatments should be imaged with the exact same acquisition parameters and all downstream analysis comparing treatments should only be applied to images which were acquired with the exact same settings.) Although imaging settings should be held constant over the course of a study, the exact imaging settings used are not necessarily critical for determining intracellular T_t. For example, in our previous work with EGFR-ELPs and CAV1-ELPs, we employed an older epifluorescent microscope equipped with lower resolution non-immersion/air objectives.^16,21 Although this imaging setup was inferior to the setup we currently employ, we were still able to detect significant differences in transition temperature and quantify differences in intracellular microdomain size and number for different ELP fusion proteins. Although this method’s results are not objectively linked to one particular imaging setting, we do recommend a few important guidelines: (1) Employ water or oil immersion objectives, ideally water immersion. The buildup of condensation during temperature ramp imaging makes work with non-immersion/air objectives extremely challenging. Water immersion objectives are ideal over oil because water (or Immersol W) will match the refractive index of the imaging solution. (2) If using a water immersion objective, use Immersol W (or a similar solution) instead of DI water as the immersion solution. Immersol W has the same refractive index as water but does not evaporate as easily. This is important because as the temperature is increased during imaging, a small droplet of water placed on the objective can easily evaporate. When evaporation happens, images lose substantial quality. (3) Balance magnification with numerical aperture strength. Although higher numerical aperture objectives provide superior resolution and light collecting ability, they are typically restricted to higher magnification objectives. Magnification power is important because temperature changes can produce focal shift in the axial and lateral planes, although in our experience drift in the lateral plane is typically much more substantial. With high magnifications/small fields of view, lateral drift means that some cells may actually drift out of the field of view during the temperature ramp. Lower magnifications circumvent this problem by tracking a larger field of view. Larger fields of view/lower magnifications also mean that more cells can be imaged at a time. In addition, given two objectives with the same numerical aperture strength but different magnifications, the objective with the lower magnification will produce brighter images. This can help avoid photobleaching, because lower laser powers can be used. (4) Balance image quality and acquisition speed with photobleaching. On a confocal microscope, once an objective is selected, laser power, pixel dwell time, and averaging are positively correlated with photobleaching. Photobleaching is especially a concern during live-cell imaging when numerous images of the same field of view are collected. However, once an appropriate laser power is selected, increased pixel dwell time and averaging can improve image quality by increasing signal-to-noise (SNR), although increasing either of these options will slow acquisition time. Rapid acquisition times are important in live-cell imaging of dynamic events, such as intracellular ELP assembly. In general, experimenters will have to empirically determine their optimal imaging settings which maximize SNR and image acquisition speed while reducing photobleaching. (5) Researcher should acquire 16-bit images. Due to cell to cell variability in differences of GFP-ELP protein expression, cells within the same culture dish may exhibit order of magnitude differences in fluorescence intensity. (In Figure 7, below, we quantify how these differences in fluorescence intensity/ELP concentration affect the intracellular T_t.) Because of this large dynamic range, it is very important that experimenters acquire images with maximal bit depth. In addition, researchers should be conscious of this large dynamic range when analyzing or simply viewing their images, as different display settings may obscure dim cells. For example, the default contrast FIJI selects when opening images often selects a contrast which obscures the dimmest cells within the image. If researchers are not careful, these dim cells may be ignored and not counted in their analysis.

Figure 7.

Intracellular transition temperature (T_t) and GFP-V60 integrated density follow a log-linear relationship. In free solution, the ELPs’ transition temperature correlates with the logarithm of concentration. To assess if a component of cell-to-cell variability T_t relates to expression level, it was plotted versus the background-corrected integrated density (BCID) of GFP-V60 signal from each cell. Each dot represents a cell from the same dataset in Figure 5. The purple diamonds are for the three CAV1-V96 + GFP-V60 cells which were filtered out based on their low microdomain size. Both CAV1-A96 + GFP-V60 (p-value = 1.92 × 10⁻⁷, R² = 0.81) and GFP-V60 alone (p-value = 3.52 × 10⁻⁸, R² = 0.68) display a significant linear trend in decreasing T_t with increasing log₁₀(BCID). In contrast, CAV1-V96 + GFP-V60 does not display a significant trend on either the unfiltered data or the filtered data (unfiltered p-value = 0.92, unfiltered R² = −0.12; filtered p-value = 0.35, filtered R² = 0.01).

4. Temperature Control during Live-Cell Imaging.

Before imaging, a single 35 mm MatTek dish, corresponding to a single transfection condition, was removed from the 30 °C incubator, washed one time with dPBS, and then replenished with 2.5 mL of 4 °C live-cell imaging solution. These ice-cold temperature steps were taken to help resolubilize any intracellular ELPs which may have phase transitioned during the 72 h incubation at 30 °C, as well as to stall any ongoing intracellular trafficking. The 35 mm dish was then placed on ice and carried to the microscope room where it was placed on a PeCon Heating Insert P Lab-Tek S1stage attached to circulating temperature bath which was held at 4 °C. To accurately measure the temperature of the live-cell imaging solution, a PeCon Control Sensor T S1 was placed directly into the live-cell imaging solution (Figure 3). Once all equipment was in position and the microscope objective was in focus, the temperature bath was set to a ramp rate of 15 to 60 °C over 20 min, and once 60 °C was achieved, the temperature bath was held at 60 °C. During the ramp, 90 images were acquired, with image acquisition every 20 s. This resulted in a total imaging duration of 30 min. The temperature of the live-cell imaging solution at each image acquisition was manually recorded. Importantly, although the temperature bath was set from 15 to 60 °C, the actual measured temperature of the live-cell imaging solution in the 35 mm dish changed from 20.2 °C (95% CI: 19.5–20.8 °C) to 42.2 °C (95% CI: 41.9–42.5 °C) during the 30 min period of image acquisition.

Figure 3.

Equipment configuration during live-cell imaging. To rapidly heat live cells during imaging, transfected cells are cultured on a 35 mm glass bottom MatTek dish which is placed on a PeCon microscope temperature stage attached to a circulating bath. To accurately measure the temperature of the imaging solution, a special lid is used that allows a temperature probe to be directly inserted into the imaging solution inside the glass bottom dish. (A) Side view of experimental setup. Note that the PeCon stage is placed inside the Zeiss piezo stage, which assists in fine-tuning automatic focus stabilization during imaging. (B) Arial view of experimental setup. The attached circulating bath hoses are seen on the back-right side of the temperature stage. (C) The circulating bath is housed on the ground underneath the antivibration table. Note the use of a bucket and the covering placed around the hoses on the table which prevent condensation from damaging sensitive electrical equipment.

Critical Points and Comments:

Accurately measuring the temperature of the live-cell imaging solution inside the 35 mm dish is one of the most challenging and important aspects of this entire method. It is important to never rely on temperature controllers alone (e.g., temperature baths, hot plates, heating stages, etc.) to estimate the temperature of the solution inside the imaging dish. Due to thermal lag and loss of heat to the ambient surroundings, the temperature of the imaging solution during temperature ramps will almost always be different than the temperature displayed on the temperature controller. This leads to two important points: (1) To obtain accurate measurements, a temperature probe must be directly submerged in the imaging solution. (2) To get the solution to a certain temperature, 37 °C for example, the temperature controller will typically have to be set to a higher point than 37 °C. We chose a temperature ramp from 15 to 60 °C as this allowed us to heat 2.5 mL of imaging solution from 20.2 °C (95% CI: 19.5–20.8 °C) to 42.2 °C (95% CI: 41.9–42.5 °C) within 30 min. However, as other laboratories may use different equipment or volumes of imaging solution, we recommend that investigators empirically determine the appropriate end points for their temperature ramps.

In recent years, the scientific field has developed methods to map intracellular temperature at the single cell level.²⁷ For the purpose of this study, the assumption was made that the large excess volume of media above the cells served as a heat sink for the cells themselves. This is because with approximately 1 million cells per well, assuming each cell might have the volume of a sphere with radius 5 μm, then it is expected that on the order of 0.02% of the volume in the dish is contained within the cells. Thus, the measured temperature of the media was used to approximate the average temperature within the cells. However, future work could explore the use of intracellular thermometers to map intracellular temperatures across the individual cells during microdomain formation.

5. Image Analysis.

5.1. Blinding and Randomizing Imaging Data.

All image analysis was performed using FIJI.²⁸ It is recommended that any image analysis requiring human intervention, whether it involves directly identifying a cell’s T_t, drawing ROIs, or selecting thresholds for subsequent microdomain analysis, be performed on blinded and randomized imaging data. To blind and randomize our datasets, we developed a Java program that copies a specified directory of images, renames each image with a random three-digit number, and then places the renamed images in a new directory, along with a key which matches each original filename with its corresponding three digit number. This program is called File Randomizer and is part of a larger plugin named Simple Image Analysis Library (or SIAL).²⁹ SIAL also contains plugins to assist in phenotype scoring and ROI recording. To download SIAL, researchers should open FIJI, go to “Help > Update…” and then update FIJI. After FIJI is finished downloading all updates, a window named “ImageJ Updater” will open. Researchers should select “Manage Update Sites > Add update site” and add this url: https://sites.imagej.net/D-tear/. To ensure that FIJI installs SIAL, researchers should click the check box next to the newly added update site. Then researchers should select “Close > Apply changes”.

FIJI will download SIAL.jar and all associated dependencies. After successfully updating, FIJI will then ask to be closed and restarted. After doing this, SIAL can be accessed via the Plugins tab in FIJI. Note that SIAL will usually be installed toward the bottom of the available FIJI plugins. To randomize a directory of images, users should select “SIAL > File Randomizer” and type in the extension of the imaging data (i.e., .tif, .czi, .ori), and then select the input directory where the images are located, and an output directory where the randomly renamed images should be placed. Links to YouTube tutorials covering the installation of SIAL and the use of its File Randomizer, PhenoScoreKeeper, and ROI Recorder programs are included in the Supporting Information. Code for SIAL can be found on GitHub: https://github.com/d-tear/SIAL.

Critical Points and Comments:

After randomly renaming images, and before beginning any analysis, researchers should ensure that the randomly renamed images are arranged by name in their chosen output directory (this is the default in the Windows and Mac operating systems we have tested). Arranging the randomly renamed images by name ensures that the images are uniformly shuffled. Once the randomized images are arranged by name, we recommend proceeding with subsequent analysis by opening up each image in the order that they are listed in the output directory. In contrast, if the randomized images are arranged by date modified or size (or any other metric), there is no guarantee that the images are uniformly shuffled after randomization. Note that SIAL’s PhenoScoreKeeper and ROI Recorder plugins automatically open up images in numeric order, so these programs ensure that the images are presented in a uniformly shuffled fashion after randomization.

5.2. Visual Determination of Intracellular T_t.

After using the File Randomizer to randomize time lapse image stacks of cells undergoing a temperature ramp, the transition temperatures of individual cells were determined through visual inspection by identifying the first frame at which that cell produced fluorescent microdomains. Examples of fluorescent microdomain formation are seen in Figure 4.

Critical Points and Comments:

As previously mentioned, due to cell to cell variability in GFP-ELP protein expression, cells across a given study will commonly exhibit large differences in fluorescence intensity. Accordingly, during visual inspection to determine intracellular T_t, experimenters will likely need to adjust image contrast to ensure that they are not neglecting cells with weaker signal from their analysis.

5.3. Microdomain Analysis.

All microdomain analysis was performed on the same blinded and randomized datasets as described in section 5.2. To analyze each transitioned cell’s microdomains and average microdomain size, we selected either the final image from the temperature ramp for that cell, or the last clear image, if the final image was out of focus or otherwise unclear. The final image for microdomain analysis was chosen for two reasons: (1) In this method, average microdomain size and number are primarily used to differentiate single from dual transfected cells. Accordingly, clear distinct images of the microdomains are required. Microdomain structure is typically clearest at the end of the temperature ramp as the microdomains fully mature and stabilize in terms of number per cell and size. In contrast, at earlier time points, including the midpoint, the intracellular microdomains often exhibit motion and rapid changes in fluorescence intensity as neighboring microdomains coalesce. These dynamic events tend to produce noisier images where distinct differences in size and number are harder to differentiate. (2) Because the individual cells exhibit a range of transition temperatures, at earlier points in the temperature ramp cells with low intracellular ELP concentrations may not display any microdomains because the temperature is not yet high enough. To measure microdomain size and number per cell, we used the Analyze Particles program within FIJI. This program can be accessed via “Analyze > Analyze Particles…”. Note that in order to use the analyze particles program, experimenters will have to threshold their image and then convert it into a binary image: “Image > Adjust > Threshold”. An appropriate threshold can then either be manually adjusted or one of the built-in thresholding algorithms can be selected. In our work we manually adjusted the threshold to identify a setting which agreed with visual inspection. However, in our experience FIJI’s default thresholding algorithm or Otsu also typically perform very well. Whichever method is selected should be consistently applied to all images to avoid misrepresenting the data, and researchers should also blind and randomize their data before selecting image processing steps. Once a threshold is selected, the image can be converted into a binary image by pressing apply.

After thresholding, microdomain measurements were obtained on the binary images for individual cells by using the ROI manager (“Analyze > Tools > ROI Manager…”) to outline individual cells. Microdomain measurements were then collected by selecting “Analyze > Analyze Particles…”. These measurements were restricted to individual cells by selecting that cell’s ROI in the ROI manager. The exact analyze particles options used for our images is shown in Figure S1.

Critical Points and Comments:

Scientific image analysis begins with good image acquisition. This is because downstream image analysis often depends on the ability to cleanly separate foreground from background. Accordingly, experimenters should optimize image acquisition to obtain the highest signal-to-noise ratio (SNR) possible without inducing phototoxicity or photobleaching. In cases where noise has contaminated a dataset, the use of filters can improve thresholding performance. Although we did not find the use of filters or preprocessing necessary in this dataset, in our experience, median filters perform well for removing “salt and pepper” noise from images. In FIJI version 2.0.0, filters can be accessed via “Process > Filters”. The addition or subtraction of filtered images from original images can also improve thresholding performance. These types of operations can be performed in FIJI via “Process > Image Calculator”. Background intensity levels can also be directly subtracted from images by using the Math tab in FIJI, accessed via “Process > Math > Subtract…”. However, it is recommended that image processing always be applied consistently across all images to ensure that the data is unbiased.

Even in datasets with high SNR, it is recommended that experimenters restrict analyses to features larger than one pixel (or the equivalent size in microns²). This size option can be directly set in the “Analyze Particles…” menu (Figure S1). In our images, we restricted analyses to particles >0.044 μm². This criterion was chosen as our images were 338.29 μm by 338.29 μm at 2272 pixel by 2272 pixel, which equals a scale of 0.02217 μm²/pixel². To ensure particle selections were conservatively restricted to more than square pixel, 0.02217 μm²/pixel² was multiplied by 2 pixel², resulting in 0.044 μm² (i.e., the smallest particles in terms of area were 0.044 μm²).

5.4. Background-Corrected Integrated Density.

All measurements of background-corrected integrated density (BCID) were performed on the same blinded and randomized datasets as described in sections 5.2 and 5.3. To collect (BCID), clear images of cells from below their transition temperature were analyzed by first collecting at least three ROIs from empty regions of that image. The average of the mean intensities of these empty regions was computed to determine the overall mean intensity of the background of that image. This value was then subtracted from the image by using “Process > Math > Subtract…”. If the overall mean intensity of the background was less than 1, nothing was subtracted from the image. After background was subtracted from the image, ROIs were drawn to outline individual cells from that image and measurements of integrated density were recorded.

Critical Comments and Suggestions:

BCID is also sometimes called corrected total cell fluorescence (CTCF), and it is a commonly applied method in quantitative fluorescent microscopy.^30–33 We previously developed an image processing pipeline to help partially automate CTCF measurements from images. The code and instructions for this pipeline are freely available.³⁴

5.5. Standard Deviation Determination of Intracellular T_t.

In addition to visually determining the temperature at which intracellular ELPs phase transitioned, we also examined an alternative technique which relies on measuring the standard deviation of the pixel values inside cells as they undergo the temperature ramp. The rationale for this technique is based on the observation that below the T_t, GFP-V60 fluorescence is homogeneous throughout the body of the cell. Accordingly, the standard deviation of the pixel values inside the cell are low. However, as temperature is increased and GFP-V60 begins to phase separate, fluorescence becomes concentrated into discrete areas of the cell (Figure 4). This produces a shift upward in the standard deviation, as some areas inside the cell become much darker while others become much brighter.

To explore this approach, time lapse image stacks were first stabilized by using the Template Matching plugin. (This plugin can be installed by adding this URL to the ImageJ updater: http://sites.imagej.net/Template_Matching/.) This step is crucial as temperature changes often induce substantial lateral drift in the focal plane as the 35 mm dish contracts and expands. After stabilizing the stacks with this plugin, ROIs were then drawn around each cell at the beginning of the temperature ramp (i.e., the first frame of the image stack). To automatically collect standard deviation measurements of each cell/ROI on each image in the stack, we wrote an ImageJ macro which reapplies these initially drawn ROIs on each subsequent image on the stack. This macro is called measureROIsonStack.ijm and it is provided in the Supporting Information. To analyze the standard deviation data and determine the temperature at which phase transition occurred, we wrote an R function named batch_plot. This function calculates the baseline standard deviation from the initial 〈n〉 images of the temperature ramp (〈n〉 is specified by the user, but the default is 20); batch_plot then determines the temperature at which the standard deviation shifts above a specified threshold beyond this baseline standard deviation. This threshold is again specified by the user. The default threshold is 3, reflecting three standard deviations away from the baseline standard deviation. Another reasonable threshold could be 2, reflecting two standard deviations away from the baseline, which approximates the 0.05 threshold commonly used to define significant p-values. Alternatively, users could select higher thresholds, 5 for example, to account for the increased probability of false positives due to testing multiple cells. The temperature/image slice at which the standard deviation shifts above the specified threshold is recorded as the intracellular T_t. In addition to recording these measurements for each cell, batch_plot also plots the standard deviation data and indicates where on the graph the threshold was passed (Figure S2). The code for batch_plot and some example data are provided in an R notebook in the Supporting Information.

The standard deviation approach is attractive because it relies on a quantifiable feature of the image. It can be applied to many cells, single cells, or selected locations within cells. Also, for datasets with large numbers of frames, this approach reduces the need for the user to discriminate between closely related frames to estimate at which point each cell undergoes phase separation. Despite this, we noted practical drawbacks to the standard deviation threshold. This method was time-consuming due to the need for the user to track the region of interest through multiple frames and can fail in frames where the focus upon the region of interest is not maintained. Therefore, we compared this strategy to the decisions of a computer-assisted user who visually scores blinded and randomized sequences of cells as described in section 5.2. Using this method, scoring an entire dataset of 56 cells could be accomplished in approximately one workday. In contrast, collecting standard deviation measurements from the same dataset took approximately three workdays. This increase in time was mainly due to the fact that hand drawn ROIs did not closely follow the cells during the time lapse imaging due to lateral drift during heating. Even after stabilization with the Template Matching plugin, it was not uncommon for several frames, out of the approximately 90 frames in a single time lapse image stack, to be shifted with respect to the rest of the frames. These shifted frames meant that the ROIs were no longer measuring the cells, but instead were measuring background, or partial background, which produced false positives as the standard deviation measurements would shift dramatically. As a workaround, the ROIs were manually redrawn at the problematic frames; however, this task was time-consuming and also introduced user error. To further combat this problem, one could imagine drawing a small ROI within the cell instead of outlining the entire cell. However, lateral drift during imaging is often large enough that even a small ROI can shift outside of the cell during image acquisition. In addition, there are limits to how small the ROI can be. This is because the ROI needs to be large enough to collect the bright and dim areas within the cell, otherwise standard deviation will remain fairly constant over the course of the temperature ramp. For the dual transfected cells especially, the bright microdomains typically encompass large portions of the cell. Accordingly, a useful ROI is nearly the size of the entire cell. In contrast, the eye of a trained human user is not fooled by lateral drift and can easily track the same cell even if it shifts around during imaging. For these reasons, at this time we recommend visual determination of intracellular T_t from blinded and randomized data. If the experimental setup or image processing could be improved to remove lateral drift in the ROI then the standard deviation threshold could be used to scale to larger datasets and improve accuracy in ROIs where visual determination is ambiguous.

6. Fixed-Cell Imaging to Confirm GFP-V60 Co-localization with Non-fluorescent ELP Fusion.

6.1. Transfection.

Fixed-cell immunofluorescence was used to confirm GFP-V60 co-localization with non-fluorescent CAV1-V96. The steps were similar to transfection for live-cell imaging, except that 12-well plates were used instead of 35 mm MatTek dishes. Before transfection, coverslips were placed individually into the wells of a 12-well plate. Coverslips were then immersed with 1 mL of poly-d-lysine, incubated for 5 min, and then washed with dPBS, as described above. A 1 mL portion of resuspended cells at a density of 0.5 × 10⁶ cells/mL was then seeded into each well before being placed back into the 37 °C, 5% CO₂ incubator for 24 h. After the 24 h incubation, cells were transfected with the Lipofectamine kit. The steps are the same as for live-cell imaging transfection, except that the volumes of transfection reagents and DNA were downscaled. The recipe for a single well involved (1) 1 μg of total plasmid DNA (i.e., either 1 μg of GFP-V60 DNA for single transfected wells, or 500 ng of GFP-V60 DNA + 500 ng of CAV1-V96 DNA for dual transfected wells), and (2) transfection mix consisting of 50 μL of diluted DNA/P300 mixture (containing 1 μg total of plasmid DNA, the P3000 Reagent at a concentration of 2 μg/μL of DNA, and Opti-MEM up to 50 μL) + 50 μL of the Lipofectamine mixture (47.5 μL of Opti-MEM + 2.5 μL of Lipofectamine 3000). Each individual mixture was thoroughly mixed before being added together to create the 100 μL master mix. This master mix was then incubated for 15 min at room temperature. During this time, the cell-seeded 12-well plates were removed from the incubator, washed one time with room temperature dPBS, and were then replenished with 900 μL of Opti-MEM. After the 15 min incubation, the 100 μL transfection “master mix” was pipetted drop-by-drop into the 900 μL of Opti-MEM media. Cells were then placed in a 30 °C incubator with 5% CO₂ and were incubated for 72 h before subsequent steps.

6.2. Temperature Incubation, Fixation, and Immunostaining.

After the 72 h incubation, the 12-well plates were removed from the incubator and placed on ice for 50 min before being placed on a heating block (IC25, Torrey Pines Scientific, Carlsbad, CA) for 50 min at 37 °C. Media was then removed from each well, and the cells were fixed at room temperature with 4% PFA for 15 min. After this incubation, cells were rinsed with 50 mM NH₄Cl (in PBS) for 5 min before being washed with 1× PBS for 3 min, three times (3 min × 3) on an orbital shaker. Note that all PBS washing steps were performed on an orbital shaker. Cells were then permeabilized with 0.1% Triton-X (in PBS) and then washed with 1× PBS (3 min × 3) before blocking with a 90 min room temperature incubation in 1% BSA (dissolved in PBS). Note that all primary and secondary antibodies were also diluted in this 1% BSA solution. After the incubation, coverslips were removed from the wells and placed face down on a 35 μL droplet of primary antibody solution, which was placed on parafilm. The primary antibodies used were myc-tag mouse mAb [1:1000 dilution] for CAV1-V96, and rabbit GFP pAb antibody [1:500 dilution] for GFP-V60. Primary antibodies were incubated overnight at 4 °C, and then coverslips were removed from the parafilm, flipped over, and placed back into the 12-well plate before being washed with 1× PBS (5 min × 5). Coverslips were then removed from the wells and placed faced down on a 35 μL droplet of secondary antibody solution, which was placed on parafilm. The secondary antibodies used were chicken anti-mouse IgG Alexa Fluor 647 [1:500 dilution] and donkey anti-rabbit IgG Alexa Fluor 488 [1:500 dilution]. Secondary antibodies were incubated at room temperature for 1 h, and then coverslips were removed from the parafilm, flipped over, and placed back into the 12-well plate before being incubated with DAPI for 5 min at room temperature on an orbital shaker. Cells were then washed with 1× PBS (5 min × 6) and then mounted on a glass slide, using ProLong Glass Antifade Mountant. Slides were cured for 48 h in the dark at room temperature before imaging.

Critical Comments and Suggestions:

“No primary-antibody controls”, consisting of labeling with secondary antibody without primary antibody incubation, are necessary to determine non-specific antibody binding and subsequent imaging settings. We also recommend the use of “bleed through controls”, consisting of only one secondary antibody in a sample, to determine non-specific bleed through of each fluorophore.

The exact immunofluorescence (IF) procedure outlined is likely not critical, as many laboratories have their own successful routines for IF staining. However, it is recommended that experimenters pre-incubate on ice to resolubilize any ELPs which may have assembled during the 72 h transfection incubation. Relatedly, it is also recommended that subsequent incubation steps, after the ice pre-incubation, be carried out by placing samples directly in contact with a heated surface. This is because direct contact with a heated surface ensures more heat transfer to the samples. Finally, researchers should ensure that samples are protected from light once the fluorescent secondary antibodies are applied.

6.3. Imaging.

Slides were imaged on an LSM800 laser scanning confocal microscope equipped with an airy scan detector. All images were acquired with the Airy scan acquisition mode using a plan Apochromat 63×/1.40 numerical aperture Oil DIC M27 objective. Immersol 518F was used as the immersion media. AF647, AF488, and DAPI channels were scanned sequentially. To excite AF647/CAV1-V96, the 640 nm laser line was used and wavelengths of 647–700 nm were collected. To excite AF488/GFP-V60, the 488 nm laser line was used and wavelengths of 487–533 nm were collected. To excite DAPI, the 405 nm laser line was used and wavelengths of 400–508 nm were collected. All images were 16-bit with a pixel dwell time of 3.14 μs, a lateral sampling of 0.029 μm, and no averaging.

VALIDATION AND EXPECTED RESULTS

Visual Determination of Intracellular Transition Temperature and Image Analysis.

For the case study presented here, live-cell imaging confirmed that CAV1-V96 + GFP-V60 dual transfection produced a substantial downward shift in intracellular T_t compared to GFP-V60 single transfection as well as CAV1-A96 + GFP-V60 dual transfection (Figures 4 and 5, Table 1). Intracellular transition temperatures for each cell in the study were determined by eye, after image stacks were blinded and randomized using the File Randomizer program within the FIJI plugin SIAL. GFP-V60 displayed a mean T_t of 35.5 °C, while CAV1-V96 + GFP-V60 displayed a mean T_t of 29.2 °C before filtering and 28.8 °C after filtering out cells with morphology consistent with expression of GFP-V60 alone. This reported T_t for CAV1-V96 + GFP-V60 is several degrees lower than our previous reported value of 35.9 °C.²¹ This difference is almost certainly due to the fact that, compared with our previous imaging equipment, our current imaging system provided superior resolution, SNR, automatic image acquisition every 20 s, and automatic focus stabilization. These improvements (especially the automated functions) allowed for more rapid and consistent acquisition of high-quality images than what was possible with the older, manually operated imaging equipment we previously employed. In contrast, dual transfection of temperature insensitive CAV1-A96 + GFP-V60 displayed a mean T_t of 37.3 °C. In these controls, the morphology of GFP-V60 microdomains appears similar to that of cells transfected with GFP-V60 alone (Figure 5B). If the CAV1-A96 induced phase separation, one might reasonably expect larger microdomains observed for CAV1-V96. Thus, the data suggests that CAV1-A96 alone may phase separate at a temperature greater than 37.3 °C, which is consistent with prior fixed-cell imaging studies.²¹

Figure 5.

Changes in intracellular transition temperature (T_t), microdomain size, and microdomain number can be quantified from live cells transfected with different ELP fusions. In each panel, a circle represents an individual cell. (A) The raw transition temperatures for all cells in the study. CAV1-V96 + GFP-V60 displays a significant difference in T_t compared to CAV1-A96 + GFP-V60 or GFP-V60 alone: p-value < 0.00001 in both comparisons. In contrast, GFP-V60 vs CAV1-A96 + GFP-V60 is not significant: p-value = 0.29. The three red dots are cells from the CAV1-V96 + GFP-V60 group, which exhibited average microdomain sizes consistent with GFP-V60 single transfection, which is quantified in panel B. (B) Different ELP fusions display differences in average microdomain size and number per cell. Although CAV1-A96 + GFP-V60 and GFP-V60 alone display numerous small microdomains, CAV1-V96 + GFP-V60 produces only a few large microdomains per cell. The black dotted line represents the maximum average microdomain size recorded among all GFP-V60 cells (1.81 μm²). This size was used as the filter to separate dual transfected CAV1-V96 + GFP-V60 cells from those which may express only GFP-V60. Note the three purple circles below the dotted line. These correspond to the three red circles in panel A and are the CAV1-V96 + GFP-V60 cells that did not pass the filter. (C) Transition temperature data after filtering out the three CAV1-V96 + GFP-V60 cells which did not pass the microdomain size filter. CAV1-V96 + GFP-V60 still displays a significant difference in T_t compared to CAV1-A96 + GFP-V60 or GFP-V60 alone: p-value < 0.00001 in both comparisons. In contrast, GFP-V60 vs CAV1-A96 + GFP-V60 is not significant (p-value was raised from 0.29 to 0.32, due to a larger residual mean squared error from removing the three CAV1-V96 + GFP-V60 cells from the dataset). All p-values were calculated from a Tukey’s post hoc test following a statistically significant one-way ANOVA.

Table 1.

Nomenclature, Amino Acid Sequence, and Phase Behavior of Expressed Proteins Inside of Cells

protein label	amino acid sequencea	MWc [kDa]	median Ttd [°C]	mean Tt (95% CI)f [°C]	amount of cells showing assemblyg [%]	intended behavior
CAV1-V96 + GFP-V60	CAV1-mycb-(VPGVG)96Y	62.5	29.1/27.3e	29.2 (27.7–30.6)/28.8 (26.7–30.9)e	100/100e	temperature-responsive
CAV1-A96 + GFP-V60	CAV1-myc-(VPGAG)96Y	59.8	37.6	37.3 (36.3–38.4)	39	temperature-insensitive
GFP-V60	GFP-(VPGVG)60Y	67.2	35.5	35.5 (34.5–36.5)	71	live-cell imaging tool

^aORF amino acid sequences are found in the Supporting Information.

^bmyc-epitope amino acid sequence: EQKLISEEDL

^cEstimated molecular weight from ORF.

^dMedian T_t obtained from live-cell imaging technique (Figure 5).

^eUnfiltered/filtered.

^fMean T_t and 95% confidence intervals obtained from live-cell imaging technique (Figure 5).

^gPercentage of cells showing a temperature-dependent assembly with T_t ≤ 37 °C. Calculated from data in Figure 5.

Image analysis with the Analyze Particles program within FIJI confirmed that CAV1-V96 + GFP-V60 produced microdomains which were larger and fewer number than those produced by GFP-V60 single transfection or CAV1-A96 + GFP-V60 (Figure 5B). The maximum microdomain size recorded for all GFP-V60 single transfected cells was 1.81 μm², and this parameter was used as the filter to separate truly dual transfected cells from those which were likely expressing only GFP-V60 (Figure 5B,C).

Measurements of background corrected integrated density (BCID), a proxy for GFP-V60 intracellular concentration, were collected to determine the effect of GFP-V60 concentration on intracellular transition temperature (see Figure 7). Linear regressions of intracellular T_t ≈ log10(BCID) for both the GFP-V60 single transfection and CAV1-A96 + GFP-V60 dual transfection groups displayed a significant linear trend of decreasing T_t with increasing log10(BCID) (i.e., increasing GFP-V60 concentration). For GFP-V60 single transfection, a 10-fold increase in BCID resulted in an average decrease in T_t of 3.89 °C (p-value = 3.52 × 10⁻⁸, R² = 0.68). For CAV1-A96 + GFP-V60, a 10-fold increase in BCID resulted in average decrease in T_t of 3.98 °C (p-value = 1.92 × 10⁻⁷, R² = 0.81). In contrast, CAV1-V96 + GFP-V60 did not display a significant linear trend with log10(BCID) of GFP-V60 signal on either the unfiltered or filtered data (unfiltered p-value = 0.92, unfiltered R² = −0.12; filtered p-value = 0.35, filtered R² = 0.01). However, the BCID measurements in Figure 7 do reveal that given the same intracellular concentration of GFP-V60 (i.e., the same BCID), CAV1-V96 + GFP-V60 cells will typically display a T_t several degrees lower than those seen in GFP-V60 or CAV1-A96 + GFP-V60 cells. As this analysis confirms, cytosolic ELPs like GFP-V60 have concentration dependent transition temperatures. Thus, a fraction of the variance present in these studies (Figure 5) may be explained by the variability in expression between cells. In future studies, it may be useful to incorporate BCID as a co-variate, which would enable comparison of the transition temperature between different groups at a standardized BCID level. Alternatively, it may be useful to derive stable transfected cell lines that display a more uniform level of expression.

Fixed-Cell Imaging to Confirm GFP-V60 Co-localization with Non-fluorescent ELP Fusion.

To confirm GFP-V60 co-localization with phase-separated CAV1-V96, transfected cells were pre-incubated on ice to resolubilize any assembled ELPs and were then incubated at 37 °C for 50 min before fixation and immunostaining with antibodies for myc (CAV1-V96) and GFP (GFP-V60). As expected, phase-transitioned CAV1-V96 assembled large microdomains which overlapped with GFP-V60 signal (Figure 6). In contrast, single transfected GFP-V60 cells displayed numerous small microdomains throughout the cytosol and nucleus. This finding suggests that live-cell imaging studies of GFP-V60 may be useful to identify the phase transition temperature of co-transfected ELP fusion proteins, above which GFP-V60 reveals the cellular distribution of non-fluorescent ELP fusion proteins such as CAV1-V96.

CONCLUSION

In this work, we presented a simple live-cell imaging technique to estimate the intracellular T_t of non-fluorescent ELP fusions. As a case study, we employed the recently developed CAV1-ELP library, although this method has also previously been employed to study ELP fusions of epidermal growth factor receptor (EGFR-ELPs).¹⁶

The intracellular T_t of a temperature-sensitive fusion, such as CAV1-V96, can be estimated by co-transfection with a less temperature-sensitive fluorescent ELP fusion, like GFP-V60. Intracellular assembly can then be visualized in live cells through the co-assembly of the non-fluorescent and fluorescent ELP fusion proteins. Critically, if the two different ELP species exhibit differences in temperature sensitivity, the resulting mixture of ELPs will exhibit a decreased T_t, which corresponds to the T_t of the more temperature-sensitive ELP. In addition to changes in T_t, co-assembled ELP microdomains may also exhibit pronounced differences in size or number, compared to single transfected treatments. These features make it possible to employ relatively simple live-cell imaging experiments and image analysis to estimate the intracellular T_t of a given ELP fusion library.

One limitation of the described live-cell imaging approach is that the expression of the non-fluorescent ELP cannot be directly confirmed. In these experiments, the co-expression of the non-fluorescent ELP is inferred from the reduced intracellular T_t and the appearance of microdomains which are often distinct from those produced in single transfected cells (Table 1 and Figures 4, 5, and 7). Perhaps the non-fluorescent ELP microdomains could be identified in live cells through DIC imaging, although low transfection efficiencies could make accurate identification difficult. Alternatively, stable cell lines expressing both fluorescent and non-fluorescent ELPs could be created, and then these cells could be used during live-cell imaging. This approach is particularly attractive because researchers could titrate out cell populations with specific intracellular concentrations of both ELPs and thus reduce transition temperature variability that is due to variability in ELP concentration (Figure 7). Future work should explore this approach.

To our knowledge, how cellular components affect the T_t is unknown. For example, differences in T_t for the same ELP among different cell types could potentially be attributed to differences in the concentrations of various cellular components, such as chaperone proteins.³⁵ Future work could address this question by examining the T_t of ELPs within cell cytoplasm mixtures with known concentrations of different components.

In addition to providing a detailed protocol to estimate the intracellular T_t of non-fluorescent ELP fusions, we also developed an easily installed FIJI plugin named SIAL.²⁹ SIAL contains programs for image randomization and blinding, phenotype scoring, and ROI selection. These tasks are important parts of the protocol detailed here but are also commonly employed in other image analysis workflows.