Western Blots and the Challenge of Image Integrity

The following article is an opinion piece written by Nikolas Chmiel, PhD. The views and opinions expressed in this article are those of the author and do not necessarily reflect the official position of Technology Networks.

The scrutiny that has followed a series of retractions attributed to image irregularities has fallen recently on western blotting figures, and the reason is not hard to identify. Affordability, specificity, and accessibility have made the western blot a fixture of protein biology literature, and so it has also become one of the techniques in which image irregularities are most often found.

Given the prominence of improper image handling, journals, including the Journal of Biological Chemistry, Cell, and Nature, have introduced explicit submission requirements covering image handling and normalization. What those requirements expose, more often than not, is not misconduct, but the absence of any clear guidelines telling well-intentioned researchers where reasonable image processing ends and misrepresentation begins.

Distinguishing legitimate adjustment from manipulation

A useful working distinction rests on the scope of the edit, rather than the motivation behind it. Adjustments that are applied uniformly to the entire image, thereby altering its appearance without disturbing relationships within the underlying data, are legitimate; manipulation arises where an edit is local or selective, or where it alters what a reader would conclude from the experiment.

Cloning and healing tools, which are common in general-purpose image processing software and act on the image at a local level, could reasonably be considered manipulation, as could the splicing of images to remove or relocate lanes without delineation showing what has been altered.

The routine practices that may compromise well-intentioned work

Some routine practices, such as contrast and brightness handling, cause data integrity issues more often than others, precisely because there are no set rules governing usage, and the amount applied is left to the author's discretion, making it easy to overdo in a way that changes the reader's interpretation of the data. Best practice is to apply such adjustments globally and sparingly, avoiding the aggressive enhancements that produce blown-out bands or sterile white backgrounds—both of which can conceal real data—and keeping the image histogram off the edges wherever possible.

Background subtraction is similarly part-art and part-science. Since aggressive subtraction can eliminate real data, while too little muddies the image and obscures genuine differences in signal, the key is to apply it reasonably across all lanes, and never to the extent that it hides or intensifies variation. Cropping blots and removing empty lanes, though usually undertaken to clarify a result, can equally lead to misinterpretation where the alterations are not clearly described, which is why the sites of cropping and splicing should be clearly delineated, and the raw images included with the publication.

Multiple exposures of a single blot represent a reasonable way of showing both low and high intensity bands, but quantitation performed across more than one image will be inaccurate, and quantitation should therefore always be carried out on a single image. Physically cutting the membrane raises comparable concerns, and while the practice maximizes the utilization of precious samples by allowing each slice to be incubated with a different antibody, a molecular weight marker should accompany every membrane fragment. Alternatively, multiplex fluorescence detection allows the membrane to be kept intact.

Underpinning many of these decisions is the variability that a finished figure never shows. Uneven transfer, edge effects, and patchy background are all non-uniform in character, which means that any technique applied to correct one region of an image risks overcorrecting elsewhere, and local corrections can easily cross into the realm of manipulation. Researchers face two options: to live with the imperfection, or to address the problem before image acquisition by optimizing the transfer and immunodetection protocols so that this issue does not arise.

The limits of what hardware and software can enforce

Modern imaging systems carry most of the tools required to safeguard data integrity, beginning with high dynamic range sensors that allow bright and faint objects to be imaged simultaneously, and software capable of calculating the optimal exposure time for a given sample. Built-in safeguards can restrict adjustments to only those that are applied across the whole image. Metadata relating to acquisition and postprocessing parameters—such as exposure times, lens corrections, dark subtraction and flat fielding—can also be attached to the file, acting as an additional safeguard. Similarly, lossless export features and audit trails help ensure that all captured data is preserved and recorded to help verify what was done to an image after acquisition. For laboratories working under 21 CFR Part 11 compliance, selecting imaging and analysis software that supports compliance directly, through administrative controls, secure file export, document signing and write-once storage, is the most efficient route to meeting those obligations without sacrificing ease of use.

An imaging platform can, on this basis, enforce provenance whereby the raw file is intact, the metadata is captured, and the edits are traceable. The boundary lies in what the user does to the image once it has been acquired; an instrument cannot determine whether a background subtraction is correct or a normalization is accurate, though the platform can offer tools that help the researcher assess which parameters are optimal for each image.

Normalization as a question of integrity

Normalization is where that residual judgement carries the greatest consequence, because it establishes whether a difference in band intensity reflects a difference in biology, rather than a difference in sample loading or gel transfer. Poor normalization can cause misleading conclusions to be drawn from otherwise sound data, which is why the choice of method has come to be treated as a matter of integrity rather than preference.

Normalizing to a single housekeeping protein presents two long-standing difficulties. Housekeeping proteins are frequently expressed at levels far higher than the protein of interest, so that the exposure required to detect the target can drive the control outside the linear range of quantitation. Their expression also varies with cell cycle, density, type, age, and treatment, such that normalization against a poorly characterized control invites erroneous conclusions.

Total protein normalization navigates both issues, since stain-free chemistries and reversible stains (such as Ponceau S) are linear across a wide sample loading range; the total signal is considerably less likely to change with biology than the abundance of any single housekeeping protein. Where the total sample load falls below that linear range, a housekeeping protein may be sufficient, provided it has been well characterized under the specific experimental conditions in question. Otherwise, total protein normalization remains the safer approach to maintaining experimental integrity, and the method selected should always be reported alongside a justification grounded in the biology of the experiment.

Where the tools and standards are heading

Requirements for providing raw, high-resolution images together with their associated metadata are likely to become standard over the next few years. Similarly, forensic tools used to identify manipulated images are likely to become commonplace within editorial workflows. AI adds a further layer of complexity to this picture, since most journals already ban the use of AI to generate or modify images, yet the explosive growth in the sophistication of generative models means that alterations will become increasingly difficult to detect, and progress on the editorial side will depend on detection tools keeping pace with those advances.

From the researcher's side, I suspect meaningful progress will look less like additional publishing obligations and more like convenience, in the form of automated total protein analysis of a blot, and source files and audit trails that travel with each image by default. If we want integrity to be the norm, the tools that satisfy the requirements need to be easier to use than the shortcuts used to undermine them.