GelBox: open-source software to improve rigor and reproducibility when analyzing gels and immunoblots

Abstract

GelBox is open-source software that was developed with the goal of enhancing rigor, reproducibility, and transparency when analyzing gels and immunoblots. It combines image adjustments (cropping, rotation, brightness, and contrast), background correction, and band-fitting in a single application. Users can also associate each lane in an image with metadata (for example, sample type). GelBox data files integrate the raw data, supplied metadata, image adjustments, and band-level analyses in a single file to improve traceability. GelBox has a user-friendly interface and was developed using MATLAB. The software, installation instructions, and tutorials, are available at https://campbell-muscle-lab.github.io/GelBox/.

NEW & NOTEWORTHY GelBox is open-source software that was developed to enhance rigor, reproducibility, and transparency when analyzing gels and immunoblots. It combines image adjustments (cropping, rotation, brightness, and contrast), background correction, and band-fitting in a single application. Users can also associate each lane in an image with metadata (for example, sample type).

INTRODUCTION

Gel electrophoresis and immunoblotting are ubiquitous in many types of science. A PubMed search for “gel electrophoresis” or “immunoblots” returns 3,548 publications for 2023 alone. Many different experimental techniques can be used but they all produce an image that needs to be interpreted. A typical analysis involves cropping the image saved from the scanner to an appropriate region of interest and then identifying bands. Some analyses include linear adjustments of brightness and contrast. These operations can be performed using many different software tools that include but are not restricted to Photoshop (Adobe, San Jose, CA), ImageQuant TL (GE Healthcare Bio-Sciences, Marlborough, MA), Image Lab (Bio-Rad, Hercules, CA), ImageJ (1), Fiji (2), and IOCBIO Gel (3). Scientifically appropriate analyses can be completed with each of these tools but the workflows can be challenging to document and reproduce.

Our laboratory developed an in-house tool to help analyze our gels and blots. We wanted a system that integrated our typical workflows into a single package and allowed traceability. After discussing our approach with collaborators, we realized that our software might be helpful to other researchers. This paper details our approach and describes some of the strengths and limitations of the workflow.

METHODS

GelBox (Fig. 1) is a program developed in MATLAB App Designer (MathWorks, Natick, MA). The analysis workflow with GelBox starts with an image file and is illustrated in Fig. 2.

Figure 1.

In the GelBox user interface, green box shows the selected region of interest (ROI), and the resulting density profiles and curve fits are automatically calculated. Illustrative image shows a 1% agarose gel stained with Oriole fluorescent stain (Bio-Rad, Hercules, CA). Lane 1 shows a rat soleus sample with the most prominent band corresponding to the N2A isoform of titin. Other lanes show the N2B and N2BA isoforms of titin in samples prepared from human ventricular myocardium. AU, arbitrary unit.

Figure 2.

Analysis workflow. GBX, GelBox; ROI, region of interest.

Human Tissue Samples

All the gels and blots shown here are data generated by our laboratory using human myocardium samples collected from the explanted hearts of donors and transplant recipients. The University of Kentucky Institutional Review Board approved the collection procedure (IRB 46103). The patients gave informed written consent for the use of their samples. The collection and storage protocol of the specimens were explained elsewhere (4).

Biochemical Assays

Cardiac titin isoform, N2B, and N2BA separation was performed using a protocol from Greaser and Warren (5). Briefly, 10 mg of left ventricular tissue stored at −80°C was pulverized with a liquid nitrogen-cooled glass tissue grinder (Kimble Kontes, Vineland, New Jersey) to produce a fine powder. The pulverized tissue was tempered on ice for 20 min and then resuspended in 8 M urea buffer containing 50% (vol/vol) glycerol. Samples were mixed within the glass tissue grinder for 1 min, transferred to a test tube, and then immediately incubated at 73°C for 12 min. After incubation, samples were plunged 10 times using a syringe with a 21-gauge needle and spun at 15,000 g for 5 min. Protein (800 µg) was loaded into each lane of a 1% vertical sodium dodecyl sulfate (SDS)-agarose resolving gel containing 30% glycerol. The first lane of the gel was loaded with a rat soleus sample as a loading control since that muscle prominently expresses the N2A isoform of titin. The gel was run for 3.5 h at 30 mA in a 4°C fridge. For isoform analysis, gels were stained with Oriole fluorescent gel stain (Bio-Rad, Hercules, CA) and scanned on a ChemiDoc imaging system (Bio-Rad). The images were saved in a 16-bit raw TIF format.

For the blots probing the phosphorylation of cardiac myosin binding protein C (cMyBP-C) at Ser302, samples of human left ventricular myocardium were submerged into 4 M urea extraction buffer with phosphatase and protease inhibitors, homogenized in a bead homogenizer, and subsequently boiled for 5 min to solubilize the proteins. Each sample (10 µg) was loaded into each lane of an 8% acrylamide gel that contained 0.5% (vol/vol) tri-chloroethanol (TCE) to allow for stain-free total protein quantification (6). Each lane was loaded with samples from different patients. In addition, each gel had negative and positive control lanes containing samples treated with Lambda Protein Phosphatase (P0753S, New England BioLabs) and PKA (539576, Sigma), respectively. Gels were run for 3 h at 160 V, imaged for total protein, then transferred to a PVDF membrane, and probed with a custom phosphorylation-specific antibody targeting cMyBP-C’s Ser302 site (ProSci, 1:10,000 dilution) overnight (7). A commercially available secondary antibody [goat anti-rabbit IgG (H&L) 32460, ThermoFisher, 1:10,000 dilution] was used followed by a 1-min rinse in SignalFire ECL Reagent (6883S, Cell Signaling Technology) before imaging. Blots were scanned on a ChemiDoc Imaging System (Bio-Rad), and the images were saved in 16-bit raw TIF format.

Metadata

Gels and immunoblots have several different types of metadata. One type describes the image itself (for example, file size, image format). This is stored in the GelBox data file and can be exported to a text file when useful. Another type of metadata describes the samples (for example, species, sex, protein concentration) and must be provided by the user. GelBox allows the user to upload this lane-level information from a spreadsheet at which point it can be merged with the results of the band analyses. There is no practical limit to the amount of sample information that can be added. Users can include as many fields as the spreadsheet format will accommodate.

Image Files and Adjustments

GelBox allows users to load images in any format that MATLAB can open. This includes most of the commonly used formats in this area of science including TIF (Tagged Image File Format), PNG (Portable Network Graphics), BMP (Windows Bitmap), GIF (Graphics Interchange Format), JPEG (Joint Photographic Experts Group), JPEG 200 (Joint Photographic Experts Group 2000), and PGM (Portable Graymap). The resolution of the imported images is a crucial factor in the analysis process. Low-resolution images provide less information and may compromise the results. Users can crop their images to a desired size, rotate the image if necessary, and then adjust the brightness and contrast as needed. In addition, users can invert their image if needed. GelBox does not allow users to perform gamma correction as this adjustment can change the relative intensity of bands.

Regions of Interest

Users define a region of interest (ROI) on each lane they want to analyze by drawing a rectangular box on the image. The ROI must be at least one pixel wide and cannot exceed the size of the image, but it is otherwise unrestricted. If multiple boxes are drawn, GelBox ensures they all have the same size. If the user changes the size of a box, the other boxes resize automatically. If metadata have been provided, each ROI is linked to the appropriate information.

Analysis

Many investigators want to quantify the intensities of bands in their images. Two factors that complicate this task are background correction and potential overlap between band profiles. GelBox provides tools that handle these issues with reproducible data-driven approaches.

Background Correction

GelBox allows the user to remove trends in the background data using one of three methods: cubic smoothing spline, straight line, and rolling ball (Fig. 3). The cubic smoothing spline method marks portions at the ends of the density profile and interpolates a cubic spline between these segments. The straight-line method follows a similar approach, where the sampled points are used to fit a straight line. Lastly, in the rolling ball method, a circular element is “rolled” under the density profile. The route tracked by the center of the element is used to define the background intensity (8). GelBox has default values for cubic spline (fraction and smoothing), straight line (fraction), and rolling ball (ball size) parameters. Users can change these parameters to modify the estimated background through the user interface.

Figure 3.

Background correction methods. A: region of interest from a gel showing isoforms of cardiac titin. B: density profile and estimated background profiles with three methods. Shaded colored areas highlight the portions at the ends of the profile that are used to estimate the background with the cubic smoothing spline and linear approaches. AU, arbitrary unit.

Band Quantification

GelBox quantifies the total density associated with each band by fitting mathematical functions to the profile of the region of interest. Our laboratory has already published a variant of this method (9) and showed that it provides more accurate quantification than estimating band intensities with adjacent regions of interest when band profiles partially overlap. GelBox’s implementation extends on the prior method by allowing analysis of more than two bands simultaneously. Specifically, GelBox fits the sum of n (potentially asymmetric) Gaussians to the band profile with a mathematical function defined as:

$$S\left(X\right)=\sum _{i=1}^n{A}_i\text{exp}(-{\alpha }_i({X\prime }_i-X_i)²)and\hspace{0.17em}{X\prime }_i=\{\begin{array}{l}X+{\varphi }_i(X-X_i),X\ge X_i\\ \hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}\hspace{0.17em}X,X<X_i\end{array}$$ (1)

where x is the vertical position in the ROI, n is the number of bands, A_i and x_i are the peak amplitude and center of each band, α_i determines the width, and ϕ_i is a skew parameter.

The fit is performed by using the simplex search method (10) to minimize Eq. 2:

$$E={\displaystyle \sum _{j=1}^k\left(D(X_j)-S(X_j)\right)²}$$ (2)

where k is the height of the ROI in pixels, and D(x) and S(x) are the density and fitted profiles, respectively.

GelBox initializes the parameters for peak amplitude and its location, width, and skew at the beginning of the optimization process. These parameters are updated, as Eq. 2 is minimized. Users can review the initial and final parameters through the GelBox interface. The optimization scheme allows users to constrain or change the initial values of the parameters if required. The area associated with each band is calculated by integrating the area under each Gaussian function (11).

Workflow Traceability

GelBox provides an option to combine the raw image, the user-supplied metadata, each step in the workflow, and the band data in a single GBX (short for GelBox) file. GBX files are compressed files written in MAT format (12) and can be reopened in GelBox to revisit analyses. This is important for rigor, reproducibility, and traceability.

Data Export

Users can also export a summary of each analysis in Excel format. The first sheet in each file combines the available metadata, the position of each region of interest, calculated densities, and the background correction method. Additional sheets provide the raw and fitted density profiles for each lane. Users can perform subsequent analysis, such as comparing densities across lanes or normalization by controls, using their preferred tool with the exported spreadsheets. GelBox can also generate summary figures that show the density profiles and quantification along with the raw and adjusted image.

Availability of Software

GelBox is available at no cost to academic users from https://campbell-muscle-lab.github.io/GelBox/.

The installation instructions are provided on the website. Users can access the source code through the GitHub repository (https://github.com/Campbell-Muscle-Lab/GelBox).

RESULTS

Software Validation

GelBox was validated by comparing its output to results obtained using Image Lab (Bio-Rad, Hercules, CA) and ImageJ. The raw data were blots probing the phosphorylation of cMyBP-C at Ser302.

Image Lab analyses were performed as described in the software’s user guide (13). The analysis with ImageJ was conducted as explained by Stael et al. (14), with the uncalibrated OD option unchecked to keep the analyzed bit-depth consistent among tools. Subsequent analysis was performed using MATLAB (Natwick, MA) and SAS (SAS Institute, Cary, NC).

Each lane’s area result was normalized with the corresponding control samples in the blots (Fig. 4A). The band areas were determined differently by each tool, as illustrated in Fig. 4B. As described earlier, GelBox fits Gaussian functions to the band profiles. In contrast, Image Lab cuts off a portion of the enclosed area, while ImageJ reports the area enclosed by the profile and the manually drawn baseline. GelBox estimates were in good agreement with those obtained with Image Lab; however, they differed slightly from those obtained with ImageJ (Fig. 4C). In addition, one-way linear mixed model analysis showed that the ImageJ results differed from those obtained by Image Lab (Fig. 4D). These differences resulted from the variation in area calculation methods.

Figure 4.

Software validation. A: representative immunoblot probed for Ser302 phosphorylation. B: schematic band quantification by GelBox, Image Lab, and ImageJ. C: comparison of Ser302 phosphorylation obtained with GelBox against Image Lab (left) and ImageJ (right). D: collated data analyzed with one-way linear-mixed model. AU, arbitrary unit.

Comparison of Background Correction Methods

Background correction can impact the quantification of bands on gels and immunoblots. This was investigated by analyzing computer-generated images with known band intensities superposed on mathematically defined backgrounds. Specifically, 100 images were generated by adding a single Gaussian function to a profile defined by a third-order polynomial. The Gaussian and polynomial parameters were selected from pseudorandom distributions, yielding a wide range of qualitative image types. Speckle noise was added to each image to mimic experimental artifacts.

Each image was then analyzed using the cubic smoothing spline, linear fit, and rolling ball background correction methods (Fig. 5, A and B). The ratios of the calculated to actual band densities were determined (Fig. 5D). One-sample t tests showed that the linear and rolling ball correction methods led to underestimates of the true band density. Both approaches removed a portion of the true band during the correction procedure. These tests suggest that background correction via cubic smoothing splines may be more robust than linear or rolling ball methods when analyzing gels and immunoblots.

Figure 5.

Comparison of background correction methods. A: computer-generated pseudo image. Region of interest (ROI) is highlighted with the green rectangle. B: mean density profile and the estimated backgrounds (left) and superposed traces of corrected and the actual profile (right). C: examples of pseudo images. D: ratios of the calculated areas (A_cal) to actual areas (A_true) from 100 pseudo images. P values show the results of one-sample t tests that compare the ratios calculated by each method to unity. Linear and rolling ball methods significantly underestimated the true intensity of the simulated bands. Plots show extremes (end-caps), quartiles (box), and median (central bar).

Representative Analysis Using GelBox

The gel shown in Fig. 6A was generated by our laboratory following the aforementioned experimental protocol for samples of human left ventricular myocardium.

Figure 6.

Representative analysis. A: representative gel image showing housekeeping lanes and resolved proteins from samples of the human left ventricular myocardium. B–D, insets: region of interest (ROI) boxes 1 to 3. Shaded areas show the Gaussian profiles fitted to the extracted density profiles. AU, arbitrary unit.

Figure 6, B–D, shows the representative analysis of α-actinin (∼103 kDa), β-tropomyosin (∼36 kDa), α-tropomyosin (∼34 kDa), troponin I (∼24 kDa), and cleaved troponin I (∼23 kDa). The protein band under the α-actinin is not known to the best of our knowledge (Fig. 6B). Following the background correction, fitted Gaussian functions were used to calculate the area of each band. Troponin I had the largest area among the analyzed proteins.

DISCUSSION

GelBox is a computer program tailored to help researchers analyze gels and immunoblots. It allows the user to complete their analysis workflow in a single package and saves the raw data, user-defined metadata, and all workflow steps in a single file. This can enhance rigor, reproducibility, and traceability.

Another strength is the use of curve-fitting to determine band densities. As previously explained (9), this is particularly advantageous when the band profiles overlap. The fitting procedures reduce unintended bias and add experimental rigor. GelBox analysis was compared with Image Lab and ImageJ using a dataset collated with left ventricular samples from patients with dilated cardiomyopathy and organ donors. Further analysis with a two-way linear-mixed model showed that all the tools resulted in the same significant difference between heart failure and organ donor groups (data not shown). Although lane-to-lane comparisons slightly differed, the interpretation of the data agreed among GelBox, Image Lab, and ImageJ. Three different background correction methods are provided. The linear and rolling ball approaches are implemented in many other packages and may be familiar to many users. However, our mathematical analysis suggests that they may lead to underestimates of band intensities (Fig. 5). Cubic smoothing splines may provide a better alternative.

A limitation of GelBox is that it is written in the MATLAB language. Although the code is open source, MATLAB is a commercial product. Many universities have site licenses, so academics can use and potentially adapt the software for their own use without financial cost. Users can also install a standalone precompiled version of GelBox. This approach would allow them to use GelBox without installing the main MATLAB package, but it requires installing MATLAB runtime prerequisites. Although free, these are cumbersome to install and complicated to maintain.

The website for the software provides comprehensive tutorials that explain how to analyze images with GelBox. The site also provides a troubleshooting guide that describes how to mitigate commonly encountered issues.

GelBox simplifies the analysis of gel and immunoblot data, but it cannot compensate for a poorly conducted experiment. The quality of the raw data is critical. In addition, users need to have a clear understanding of how GelBox works to analyze their data appropriately.

DATA AVAILABILITY

Data will be made available upon reasonable request.

GRANTS

This work was supported by National Heart, Lung, and Blood Institute Grants HL149164 and HL148785.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

U.G., A.G.W.-H, and K.S.C. conceived and designed research; U.G. and A.G.W.-H performed experiments; U.G., A.G.W.-H, T.K., and K.S.C. analyzed data; U.G., A.G.W.-H, T.K., and K.S.C. interpreted results of experiments; U.G. prepared figures; U.G. drafted manuscript; U.G., A.G.W.-H, T.K., and K.S.C. edited and revised manuscript; U.G., A.G.W.-H, T.K., and K.S.C. approved final version of manuscript.