Drawing a line: Setting guidelines for digital image processing in scientific journal articles

Guidelines for authors

Guidelines relating to digital image manipulation can be accessed from the journal websites, as part of the ‘Information for authors’ section. Both JCB and the Nature journals distinguish guidelines relating to digital image alteration from other guidelines relating to ‘figure preparation’. Whereas ‘figure preparation’ guidelines include information about accepted file formats, resolutions and labelling conventions for images, image manipulation is treated as part of ‘editorial policies’. For example, the image-processing guidelines for the Nature family of journals are provided on a webpage with the title ‘Image integrity and standards’. ⁶ At least rhetorically, this separation of content serves to distinguish image processing from more ‘technical’ aspects of manuscript preparation, instead linking the guidelines to standards of behaviour and ethics.

The guidelines essentially comprise written lists of digital image manipulations that are deemed to be essential, desirable, acceptable or unacceptable. They do not provide detailed protocols or step-by-step instructions for how to process images; rather, a stated aim is ‘to clarify boundaries of acceptability in preparing images for publication’ (Editorial, 2006b: 237). JCB (published by Rockefeller University Press) has been one of the more proactive journals in establishing image-processing guidelines, and their summary recommendations have been used as a starting point for several other journals:

No specific feature within an image may be enhanced, obscured, moved, removed, or introduced. The grouping of images from different parts of the same gel, or from different gels, fields, or exposures must be made explicit by the arrangement of the figure (e.g., using dividing lines) and in the text of the figure legend. Adjustments of brightness, contrast, or color balance are acceptable if they are applied to the whole image and as long as they do not obscure or eliminate any information present in the original. Nonlinear adjustments (e.g., changes to gamma settings) must be disclosed in the figure legend. (Rossner and Yamada, 2004: 12)

Across the journals examined here, most of the guidelines that have been developed touch on four main points regarding image processing. The first relates to the scope of manipulations. Any adjustments made using digital processing software must be applied to the whole image, not selectively to discrete parts of the image. Furthermore, no global adjustment should be undertaken if it hides or removes information present in the original image. Practices such as adjusting contrast levels on images of gels or blots in order to eliminate faint (and, perhaps according to the author, spurious or irrelevant) bands are therefore not acceptable (see Figure 1). Nor, according to this rule, is it acceptable to remove discrete artefacts from an image – for example, a speck of dust, or a routine cosmetic defect arising from a known imperfection with the scientific equipment being used (for examples of such adjustments, see Figure 2; Lynch and Edgerton, 1988: 205–209).

OPEN IN VIEWER

Figure 1.

Example of inappropriate brightness and contrast adjustment of blots (reported in Rossner and Yamada, 2004: 13). The original image contains the following caption: ‘Manipulation of blots: brightness and contrast adjustments. (A) Adjusting the intensity of a single band (arrow). (B) Adjustments of contrast. Images 1, 2, and 3 show sequentially more severe adjustments of contrast. Although the adjustment from 1 to 2 is acceptable because it does not obscure any of the bands, the adjustment from 2 to 3 is unacceptable because several bands are eliminated. Cutting out a strip of a blot with the contrast adjusted provides the false impression of a very clean result (image 4 was derived from a heavily adjusted version of the left lane of image 1).’ ©2004, Rockefeller University Press. Originally published in The Journal of Cell Biology 166:11-15 doi:10.1083/jcb.200406019.

OPEN IN VIEWER

Figure 2.

Example of selective adjustment of images (reported in Rossner and Yamada, 2004: 14). The original image contains the following caption: ‘Misrepresentation of immunogold data. The gold particles, which were actually present in the original (left), have been enhanced in the manipulated image (right). Note also that the background dot in the original data has been removed in the manipulated image.’ ©2004, Rockefeller University Press. Originally published in The Journal of Cell Biology 166:11–15 doi:10.1083/jcb.200406019.

The second point relates to cutting and pasting, or manipulations involving the rearrangement or grouping of images. According to journal guidelines, producing composite images by cutting and pasting together (selected portions of) images can be acceptable, provided that the various sub-images are explicitly delineated on the figure and the composite nature of the figure is described in the accompanying figure legend. An example deemed to have involved inappropriate image manipulation in this regard relates to two articles on cell signalling published in Nature Cell Biology in 2003 (Sawada et al., 2003a,b), in which the first author – one of the laboratory’s postdoctoral researchers – used Photoshop to make composite western blot images by cutting and pasting together bands from several different experiments (Pearson, 2005: 953). A formal investigation by the journal editors concluded that, although the interpretation of the research findings was not affected by the image manipulation, ‘the frequency and severity of the manipulations’ undertaken necessitated full retraction of the papers (Editorial, 2007b: 355; Sawada et al., 2007).

The third point emphasized in the journal guidelines is transparency in process. Just as details regarding the arrangement of composite images should be provided in the relevant figure caption (see above), so too should other aspects of image preparation be detailed. For example, the guidelines produced by the Nature journals stipulate that ‘authors should list all image acquisition tools and image processing software packages used’, as well as ‘document key image-gathering settings and processing manipulations’. ⁷ For images obtained using microscopy, this information includes the make and model of the microscope and lens used, together with a list of the instrument settings used for image capture, a description of the experimental sample, and details of any post-acquisition adjustments. As well as providing greater context for image interpretation and promoting increased transparency in the reporting of findings, the documenting of such information is described as a useful mechanism for raising awareness among scientists about processes of image preparation. ⁸

The fourth key issue mentioned in most of the journal guidelines is a requirement to keep all original data relating to images. This stipulation applies less to the manipulation of digital images themselves than to practices of data storage and record-keeping by scientists. The consequences of failure to comply with this requirement are potentially serious; for example, JCB editorial policies state that the journal may request to see original data ‘for comparison to the prepared figures’. Regardless of whether inappropriate image manipulation has occurred, ‘if the original data cannot be produced, the acceptance of the manuscript may be revoked’. ⁹

In-house policies and procedures

In parallel with establishing normative guidelines for digital image manipulation by authors, journals are also developing in-house policies and procedures for handling and evaluating digital images, and for guiding action in cases of suspected misconduct. The Council of Science Editors has also become involved with summarising recent developments concerning image integrity, and suggesting best-practice strategies for journals (Council of Science Editors, 2009: 53–55). A number of editorials mention the reluctance of their journal staff to act as ‘data police’ (see Editorial, 2007a: 215; Editorial, 2007b: 355), but they nonetheless identify a need for procedures designed to detect and manage cases of misconduct in image manipulation. Several journals have thus implemented procedures for screening images once a manuscript has been provisionally accepted for publication. ¹⁰ For example, JCB instituted a comprehensive screening programme in 2002, which employs an in-house ‘forensic expert’ to systematically screen all digital images in manuscripts that have been accepted for publication (Pearson, 2005: 952; Rossner, 2002: 1151). Since then, Science and the Nature family of journals have also instituted systems for ‘spot-checking’ digital images in papers that have been provisionally accepted – these systems variously involve choosing papers at random, or selecting articles thought to be at ‘high risk’ of image manipulation. ¹¹

Editorial staff at JCB estimate that 25 percent of the manuscripts submitted to their journal contain what they deem to be inappropriately manipulated images. ¹² In most cases, the editors determine that the manipulations do not affect data interpretation; if authors are able to supply new figures that comply with the journal guidelines, the manuscript is published. However, 1 percent of manuscripts contain image manipulations that are deemed to have ‘crossed the line’ and affect data interpretation – in such cases, the manuscript is rejected, and the case may be formally reported to bodies concerned with scientific integrity.

The line between benchtop and desktop: Skill and practice in image production

Journal editors describe the use of digital image acquisition and processing tools as facilitating the production of ‘clean’ and aesthetically pleasing images, and indeed as having ‘removed the physical impairments to perfect images’ (Editorial, 2006a: 892). The increasingly widespread availability of desktop image-processing software is identified as problematic, because it has the potential to disrupt a long-held supposition: that the quality of an image tells us something about the skill of the scientist who made it. Pretty pictures are highly valued in scientific work; for example, having one’s image published as the cover artwork for a high-profile scientific journal is a coveted honour. Furthermore, a high-quality image, suggest editors of journals including JCB, Nature and Nature Immunology should reflect ‘effort’, ‘skill’, ‘expertise’ and even ‘technical mastery’ on the part of scientists (Editorial, 2006a: 892; Editorial, 2007a: 215; Rossner and Yamada, 2004: 11). The advent of software such as Photoshop threatens to sever this relationship, as care and skill at the laboratory bench are no longer necessary to generate an image of publishable quality. Instead, to cite an extreme example, it becomes possible ‘to transform a featureless black microscope snap into a starry vista littered with labelled proteins’ while sitting in front of a desktop computer (Pearson, 2005: 952).

This quotation introduces a second and related concern on the part of journal editors. Technologies such as Photoshop do not simply offer a shortcut or a substitute for careful work at the laboratory benchtop, but furthermore can result in the generation of images that do not ‘accurately’ represent the experimental data obtained. Again, the editors suggest, this development represents a departure from longstanding practice, in which ‘data acquired at the bench were almost identical to the data published, blemishes and all’ (Editorial, 2007a: 215). Journal editors frown upon the use of image-processing software to remove such blemishes, or to otherwise tidy, clean up or prettify images in preparation for publication. Nature in particular has adopted a strong stance in this regard, stating that ‘beautification is a form of misrepresentation. Slightly dirty images reflect the real world’ (Editorial, 2006a: 892). Digital image processing is thus identified as problematic not only because it can yield images that provide a false impression of how skilled an individual scientist is, but because it might result in misrepresentation of the original data, and thus mislead the readers as to what the phenomenon under investigation ‘looks like’. The possible misrepresentation of data through image processing will be discussed further below. The point I wish to make here is that statements such as ‘beautification is a form of misrepresentation’ effectively serve to draw a line between ‘scientific’ and ‘aesthetic’ practices in image production. According to journal editors, the production of beautiful images for publication should rely on scientific skill as opposed to aesthetic dexterity; practices guided by aesthetic considerations have the potential to misrepresent data, and to mislead reviewers and readers alike.

The guidelines for image preparation being developed by journals might thus be read as an intervention designed to limit actions in the name of beauty or aesthetics, and to safeguard the scientific skill involved in image production. In practice, these guidelines define a boundary between the scientific and the aesthetic by drawing a line between practices of image-making at the laboratory benchtop, and adjustments made at the computer terminal or desktop. The focus is predominantly on limiting the use of software for processing images after they have been captured in a digital format. This separation of image acquisition from image processing, and of ‘scientific’ from ‘aesthetic’ practices, is arguably a pragmatic move on the part of journals. However, this move simultaneously exposes ambiguities that have been recognized through science studies research on visual representation. Two key points might be singled out in this regard. First, detailed ethnographic studies such as those carried out by Latour and Woolgar (1979), Lynch (1985a,b), Knorr-Cetina and Amann (1990) and Myers (2008) reveal difficulties in distinguishing clearly between scientific and aesthetic practices in image production; indeed, image-making is often portrayed as something akin to a skilled craft or a design process (Knorr-Cetina and Amann, 1990: 280). In their study of digital image-processing in astronomy, Lynch and Edgerton identify aesthetic practices and judgments as deeply embedded within practices of image production. Such aesthetic considerations are not discussed or deployed by scientists in the name of creativity or beauty, ¹⁴ but are typically directed towards achieving a certain ‘representational realism’ (Lynch and Edgerton, 1988: 200), of ‘composing visible coherences, discriminating differences, consolidating entities, and establishing evident relations’ (p. 212). Framed this way, the incorporation of aesthetic judgment might be considered necessary for the production of meaningful scientific images, with an absolute separation of the scientific and aesthetic being difficult to maintain in practice. Indeed, Carusi (2008: 248) suggests that a ‘compelling’ scientific image has both epistemic and aesthetic qualities.

Second, the recent journal guidelines concentrate on digital image-processing for the purpose of publication. Perhaps as a consequence of this pragmatic focus, images are discussed primarily as products or outputs of scientific research, and as representational devices for communicating findings. Again, ethnographic studies reveal possible difficulties in constraining the scope of journal guidelines to this stage of image-making. Images do not come into being at the point of preparing a manuscript for publication (however, images destined for publication are typically singled out for particular scrutiny and preparation – see Knorr-Cetina and Amann (1990) and Lynch (1985a: 94–98)). Rather, as the scientific community is well aware, visual representations are key and constitutive parts of the knowledge production process, and are actively constructed, transformed and rendered throughout scientific practice. For natural phenomena operating at scales not accessible to direct, unmediated observation with the human eye, such transformations may be necessary to make these phenomena ‘visible’ in the first place: ‘Researchers cannot directly observe living brain cells, ribosomes, strands of DNA, or bird migration routes without making use of complex procedures for technically visualizing these phenomena as picturable, graphable, mappable, or measurable configurations’ (Lynch, 1991: 208). ¹⁵ In such instances, reality cannot be distinguished from visual representation: ‘there is no way to compare a representation of a … phenomenon to the “real” thing, since the thing becomes coherently visible only as a function of representational work’ (p. 208). The reliability of the tools and work practices underpinning the entire observation process thus becomes crucial to the credibility and representational accuracy of knowledge claims being made. This point has particular salience for digital imaging, as new, increasingly intricate and interactive configurations of equipment and practice (requiring increasingly specialized knowledge and judgment) are used to render phenomena visible and meaningful (see, for example, Burri and Dumit, 2008: 303; Cambrosio and Keating, 2000; Editorial, 2005; Myers, 2008).

Although focusing primarily on images as illustrations or end-products, journal editors do acknowledge that the generation and acquisition of images are important parts of scientific practice, and some of their recommendations extend into the realm of benchtop practice. For example, image-processing guidelines for the Nature family of journals (arguably the most detailed of all the journal guidelines analysed here) state that ‘positive and negative controls, as well as molecular size markers, should be included on each gel and blot’. ¹⁶ Such requirements have consequences for the design of individual experiments. For the most part, however, statements in the guidelines regarding benchtop practice remain quite general, such as ‘authors must … take care to exercise prudence during data acquisition, where misrepresentation must equally be avoided’. ¹⁷ On the whole, alterations to the form and composition of images are deemed preferable if performed at the laboratory bench as opposed to with post-acquisition image-processing software. This rhetorical line drawn between benchtop and desktop results in some seemingly contradictory guidelines. For example, when discussing the preparation of figures containing blots and gels, the editors of JCB would prefer authors to ‘perform multiple exposures to get the bands at the density you want, without having to overadjust digitally the brightness and contrast of the scanned image’ (Rossner and Yamada, 2004: 13). In their study of image preparation in a molecular genetics laboratory, Knorr-Cetina and Amann (1990: 279–280) detail seven manipulations of an autoradiograph blot that were considered by scientists when they were preparing what they deemed to be an image of publishable quality: these included cutting the image so as to retain only those data of interest, changing the exposure time of the x-ray film (to highlight or reduce the intensity of particular bands), and running the experiment again under different conditions designed to yield an appropriate image for publication. According to recent image-processing guidelines, each of these possibilities would be deemed preferable to adjusting the contrast levels on a digital image of the autoradiograph. Such advice may seem contradictory in terms of curbing the possibility for intervention in the final, published image, but it becomes a necessary consequence of ‘drawing a line’ between image acquisition and image processing. ¹⁸

The line between acceptable and unacceptable manipulation

When developing image-processing guidelines, journal editors also attempt to define a boundary between acceptable and fraudulent image alteration (Council of Science Editors, 2009: 53). Again, drawing a clear line here is difficult. ¹⁹ On the one hand, editors acknowledge that image-processing software can be used for the complete fabrication of results. On the other hand (and related to the discussion above), they acknowledge that banning all digital image manipulation is not only unnecessary and undesirable, it also would be impossible. As suggested in a news feature from Nature:

No one wants to ban image manipulation outright. In cell-biology experiments, for example, researchers often have to adjust the relative intensities of red, green and blue fluorescent markers in order to show all three in a single image. Even drastic changes are sometimes considered tolerable if scientists spell out exactly what they did. But it is tough to draw a precise line between acceptable and unacceptable image manipulation. (Pearson, 2005: 953)

Their guidelines aim not to outlaw the use of digital image-processing software, but to limit it to instances where it might be determined ‘essential’, with the general stipulation that ‘the final image must correctly represent the original data and conform to community standards’. ²⁰ Journal editors identify a spectrum of problematic image manipulations, ranging from honest mistakes or ‘innocent embellishment’, through to ‘scientific misconduct’ and deliberate fraud (Editorial, 2006b: 237; see also Council of Science Editors, 2009: 53). Despite the use of terms such as ‘innocent’ or ‘deliberate’, editors do not distinguish the acceptability of specific image adjustments on the basis of intent alone. They do not treat a lack of understanding or training in how to use image-processing software as an excuse for image beautification – as an editorial in Nature Methods states, ‘good intent does not make all practices acceptable’ (Editorial, 2006b: 237). Rather, journal editors distinguish between acceptable, inappropriate and fraudulent image adjustments largely on the basis of whether they affect the interpretation of data (see Council of Science Editors, 2009: 54). However, this criterion is itself far from straightforward; one might ask whether it is possible to determine a single, ‘correct’ interpretation of an image against which manipulations should be compared, given that, in principle, the interpretation of data is always context-dependent and open to re-evaluation (Lynch, 1991: 201). ²¹ The reference point according to the journal guidelines seems to be the interpretation that the authors offer in their manuscript to account for the presented data. Authors are thus granted authority to present a written interpretation of the image they have prepared for publication, but simultaneously bear the responsibility for ensuring that the interpretation offered is appropriate to the image manipulations performed (and relatively independent of any image adjustments made for primarily aesthetic purposes).

Although clear and unambiguous images are held up as an ideal, journal editors acknowledge the context-dependence of data, and furthermore emphasize that interpretations of images may change over time: ‘some observations that do not appear to make sense in the context of the current body of knowledge may turn out to be logical once the biology of the system is understood’ (Editorial, 2006b: 237). They warn authors against unnecessary image alterations, as ‘removing … peripheral information from images today will lead to contradictions tomorrow’ (p. 237). Their editorials associate extensive alteration of images with misrepresentation, and in turn link misrepresentation with possible misinterpretation of images by readers. For example:

innocent efforts to smarten or prettify images [can] end up with unintended consequences. At the very least, biologists risk erasing potentially valuable information, such as low levels of fluorescently labelled protein swilling around a cell’s cytoplasm. At worst, such manipulations can lead researchers to the wrong scientific conclusions. (Pearson, 2005: 952–953)

Altering selective parts of images – to remove artefacts or ‘unnecessary clutter’ – should be avoided for similar reasons: ‘it is not acceptable … to remove ugly, unexplainable or confusing areas of gels/blots for cosmetic reasons. Not only can it mislead the editor, referee, and ultimately the reader, but it can also hide important information pointing to real biological insight’ (Editorial, 2004: 275). ²²

Sensitive to this complexity, in practice the guidelines treat image manipulations as ‘acceptable’ if they conform to the published guidelines and do not affect data interpretation. ‘Inappropriate’ adjustments are those that violate the guidelines, but are not seen to affect the particular interpretation or conclusions that the authors propose. Most of the problematic adjustments identified through editorial scrutiny of submitted images are described as belonging to this category. Such alterations can be quite extensive, as long as the resulting images do not alter the conclusions of the paper. For example, they can include ‘adjustments of brightness/contrast to a gel image that completely eliminate the background … or that obscure background smears or faint background bands’, and ‘the splicing of images from different microscope fields into a single image that appears to be a single field’ (Council of Science Editors, 2009: 54). Image manipulations that fall into the ‘fraudulent’ category are those that do affect data interpretation, for example by ‘deleting a band from a gel to “fix” a negative control that did not work or adding a band to a gel to indicate the presence of a product that was not really there’ (Council of Science Editors, 2009: 54).

Cases of outright fraud are discussed as being fairly rare, and in any case are difficult to safeguard against through the use of guidelines alone. ²³ Rather, the guidelines are presented with the aim of promoting integrity in ‘normal’ or routine scientific practice. Journal editors suggest that much inappropriate image manipulation is done unknowingly, that scientists are often ‘unaware that their efforts to achieve the cleanest images for publication have crossed the line of acceptability’ (Pearson, 2005: 952). The introduction of guidelines for image preparation shines a spotlight on this process, and in this way may start to open up issues surrounding practices of image-making that have hitherto not received explicit attention. Furthermore, the guidelines go beyond awareness-raising by detailing a number of criteria that publishable images must comply with. They avow that the need for guidance to shape best practice in biological sciences (as opposed to other scientific sub-disciplines) is particularly acute, owing to a perceived lack of training for young researchers with respect to good practice in image preparation: ‘graduate school curricula typically do not offer systematic instruction in microscopy or image formation, with the result that most biology graduate students rely on ad-hoc training by more senior students or postdocs’ (Peterson, 2005: 881). ²⁴ Image-processing programmes such as Photoshop offer simple interfaces for adjusting images, and journal editors are concerned that biologists can be tempted or even ‘duped’ into inappropriate image manipulation owing to the availability and ease of using such software (Editorial, 2006c: 101). ²⁵

The development of journal guidelines serves as a basis from which to evaluate existing circumstances and to develop new standards or community norms. Consistent with the training function of guidelines, new briefing materials, training courses and online tools are being developed in direct response to the image-processing guidelines. ²⁶ By encouraging training and discussion regarding image-making practices, the guidelines aim to promote norms of transparency and integrity in scientific practice. Procedural requirements such as noting down instrument settings, detailing any image adjustments performed and retaining all original images are seen as practices that should help to make visible, and render more accountable, the process of image preparation.

To summarize, the image-processing guidelines being developed by journals must grapple with the complicated issue of what counts as appropriate or inappropriate image adjustment. Definitions of research misconduct necessarily enter this discussion, and journal editors have drawn a line at the point where image alteration affects data interpretation (in itself an ambiguous point). However, they are concerned primarily with shaping and guiding normal scientific practice in light of technological advances in image-processing capabilities. This ambition is consistent with a broader shift in the 1990s from initiatives designed to prevent research misconduct to those promoting research integrity (see Montgomery and Oliver, 2009). The production of image-processing guidelines is an example of the role that journals can have as sites or institutions active in continually defining and refining norms of research integrity, both with and for the scientific community.

The relationship between author and reader

The publication of guidelines for digital image processing raises questions about the relationship between the author and the reader of a research paper. In general terms, such papers might be understood as a means of advancing knowledge claims to the scientific community for the purpose of evaluation. Authors try to convince readers of the validity and importance of the claims being made by demonstrating logically sound reasoning, skilled execution of work, and a clear display of the phenomenon under investigation. In a practical sense, images are treated as central components of scientific manuscripts; it is through images that readers may see for themselves what has been achieved, and judge whether the links made between the author’s observations and knowledge claims seem reasonable. By introducing guidelines that require authors to disclose details of the experimental setup for capturing images and to list any image manipulations performed, journals are promoting an ethic of transparency focused on tracing the histories of presented images, and providing clearer correspondence between image and experiment. ²⁷

In addition to promoting transparency, some of the guidelines attempt to impose broad limits on author intervention in image production; for example, the first sentence of the Nature journals’ guidelines states that ‘images submitted with a manuscript for review should be minimally processed’. ²⁸ By using guidelines to shape author practices, journals are arguably attempting to safeguard a degree of power and autonomy for the reader, protecting his or her ability to witness from afar. Although, in principle, readers are granted authority to judge for themselves the meaning of a scientific image, it is the author who creates and selects the image(s) to present. The process of selecting, refining and framing data is central to the presentation of scientific knowledge claims, but the availability of image-processing software such as Photoshop potentially grants the author unprecedented control over the appearance of images. Authors routinely identify growing pressure to publish beautiful images, not least linked to the fierce competition for publication in high-impact-factor journals (Franzen et al., 2007). ²⁹ For a paper to catch the eye of discerning editors and reviewers, the belief is that it must contain positive and ‘significant’ findings, and provide a compelling visual demonstration of the phenomenon under question. Journal editors acknowledge the resulting pressure on authors, and state a desire to ‘end the fetish of the perfect image’ (Editorial, 2006a: 892). In practice, though, images are often chosen to represent the ‘best’ data and experiments out of many possible examples in order to persuade editors and readers of the importance and validity of a particular knowledge claim. These data might not reflect the average of all of the experiments performed. ³⁰ The selection of representative images or averaging data for the purpose of publication clearly differs from the ideal of performing an experiment directly in front of observers or witnesses, where there is no opportunity to select the most appropriate results to present.

Particularly for highly interdisciplinary journals such as Science and Nature (which appeal to a broad scientific audience), one might ask whether images in contemporary scientific papers are deployed predominantly as tools for ostension or for witnessing. ³¹ If readers are assumed to unquestioningly accept knowledge claims made in publications, a pedagogical or idealized rendering of images may be acceptable. If, however, a publication is understood to be a basis for evaluating or testing a particular set of claims, then representational accuracy and completeness become more salient. ³² The journal guidelines seem to advocate the latter understanding, asserting expertise and authority on the part of the reader for validating knowledge claims.

Importantly, despite being able to control the selection and presentation of images, ultimately authors cannot determine how these images are understood or used by readers – not least because ‘an image usually carries information beyond the specific point being made’ (Rossner and Yamada, 2004: 11). A recent initiative pioneered by JCB editors affords readers increased control over data access and image interpretation: in December 2008, the journal launched JCB DataViewer, an application that allows readers of an article published in JCB to download the original image data and its associated metadata in order to perform their own image analysis. The JCB DataViewer is designed particularly for microscopy image datasets and time-lapse videos, but can also display other images (including gels and blots) captured in a variety of file formats. JCB editors emphasize that ‘the JCB DataViewer interface makes any published image as accessible to readers as if they had acquired it’ themselves, and importantly, that it grants readers access to ‘a maximum amount of information from published images, far more than can be gleaned from a single, two-dimensional optical slice’ (Hill, 2008: 969). Through the JCB DataViewer, the interactive, digital format of captured images is maintained for readers. With this tool, readers are granted the ability to probe beneath the ‘face value’ of a printed image (Coopmans, 2011), and through this, are thought able to assume more of a ‘witnessing’ and evaluative role than previously possible. Indeed, this development might be seen to blur the relationship between authors and readers of manuscripts.

The JCB DataViewer is also presented as a tool for ensuring ‘transparency’ and ‘data integrity’ in published articles (Hill, 2008: 970). Both authors and readers are suggested to benefit – authors because they can ‘better showcase their data and … better substantiate the conclusions drawn’, and readers because they have ‘all of the information necessary to evaluate authors’ interpretations’ (p. 970). ³³ In requiring original data files, the JCB DataViewer repository indirectly serves as a mechanism to limit the degree of processing that authors undertake in preparing their images for publication: ‘We are not just asking to see a larger section of a microscope field or a larger piece of a blot, but the actual data files acquired by authors (that have not had a whiff of Photoshop!)’ (p. 970). At present, uploading of image files to the JCB DataViewer is optional; JCB editors note that as of September 2010, just over 20 percent of published papers were accompanied by images in this interactive archive, which is now being accessed by about 15,000 users each month (deCathelineau et al., 2010).

In summary, at least two aspects of recent journal interventions address the relationship between authors and readers of scientific images. By requiring increased methodological transparency associated with individual images (through written disclosure of image acquisition and processing steps), and by developing software tools that allow researchers (authors and readers) to access ‘raw’ image data and metadata, the guidelines try to limit the scope for image processing and interpretation by the author, and assert a greater role for the reader in evaluating the claims being put forward. These actions effectively serve as an attempt to reduce or constrain what might be seen as an imbalance between author and reader (one that is potentially exacerbated in the face of digital image-processing technologies), by trying to ensure comparable access to digital image data and thus equal authority on the part of authors and readers to propose and challenge interpretations of observed phenomena.

Defining objectivity in the digital age

Broadly speaking, the image-processing guidelines being advanced by journals can be read as attempts to define what counts as visual evidence and good representational practice in contemporary molecular biosciences (and nominally science more generally). This promotion of guidelines might be seen as part of a broader movement to assert a form of procedural objectivity in science, and to promote trust through ‘apparatuses of surveillance, control, and institutional discipline’ (Shapin, 2008: xvii). This push to articulate guidelines highlights a fundamental tension regarding the role of images in scientific publications: on the one hand, images should convey information that convincingly supports a particular knowledge claim, and on the other they should provide an objective representation of some aspect of the natural world. These ideals are not mutually exclusive, but the tension between them has ramifications for scientific practice, influencing the degree to which authors can acceptably intervene in framing and shaping the content of published images. By attempting to define (or refine) the position of this line, journal editors inevitably address the nature of objectivity in scientific practice. Indeed, the very development of written guidelines and conventions for image preparation might be thought of as a symptom or a reflection of a new moment in the trajectory of objectivity as a guiding value in science. ³⁴ The growing ease with which images can be digitally produced, transformed and manipulated re-poses questions about the epistemic virtues, practices and ethics of image production (Daston and Galison, 1992, 2007). What, then, might objectivity mean in the digital age, and how might it be best upheld?

The journals’ push to limit author intervention in image preparation is not limited to concerns regarding a role for the reader in evaluating evidence. The rhetoric of journal editors treats the authentic or unadulterated image as somehow a ‘better’ or more accurate reflection of the natural world than a digitally manipulated one: ‘slightly dirty images reflect the real world’ (Editorial, 2006a: 892). The current guidelines appeal to an understanding of objectivity similar to what Daston and Galison refer to as mechanical objectivity in their historical account of objectivity as performed through atlas images. They situate the emergence of mechanical objectivity in the 19th century, and stress its emphasis on scientific and ethical self-control; mechanical objectivity is about resisting ‘the temptations of aesthetics … the desire to schematize, beautify, simplify’ (Daston and Galison, 2007: 120). This form of objectivity is manifest through ‘the insistent drive to repress the willful intervention of the artist–author, and to put in its stead a set of procedures that would, as it were, move nature to the page through a strict protocol, if not automatically’ (Daston and Galison, 2007: 121). The development of journal guidelines, together with the emphasis in their accompanying editorials on distinguishing between ‘aesthetic’ and ‘scientific’ practices in image preparation, promotes a view of objectivity consistent with this description. This view plays out in the details of the guidelines too, for example through the stipulation that authors should not selectively remove known artefacts from images. ³⁵ In this respect, the guidelines treat the authenticity of an image as key to its objectivity, and as critical for engendering trust in the image. This link between trust and abstaining from intervention is not an inevitable one; for example, Lynch and Edgerton’s (1988: 206–208) study of digital image processing in astronomy identifies the ‘cleaning up’ and removal of artefacts from images as a possible (although not universally accepted) strategy for demonstrating competence among scientific colleagues. A key difference between these two examples may lie in the perceived expertise of the author in using image-processing tools – with Photoshop in particular being singled out as problematic in encouraging authors to ‘black-box’ the underlying mathematics of digital image manipulation (Greene, 2005: 143). Questions of whether it is better to leave in or to remove noise from an image, or about whether data manipulation enhances or detracts from objectivity, are intimately bound up with particular disciplinary tools, conventions and judgments about what counts as good scientific practice. ³⁶

Daston and Galison’s account of mechanical objectivity in the 19th century describes an increasing reliance on machines (such as the camera) to uphold objectivity where the scientific self might falter. Contemporary debates about image preparation are unfolding in the context of a pervasive (and yet fairly recent) digital medium for image capture and processing, and in an era of highly specialized scientific and medical research that relies on increasingly complicated configurations of machines and software for making natural phenomena visible in the first place. Imaging software and practices can be used simultaneously as tools for controlling apparatus, analysing data and representing phenomena. Cambrosio and Keating (2000) suggest that the scope of these functions effectively collapses the distinction between representing and intervening – a distinction that is actively promoted through the ideal of mechanical objectivity. Although the assertion of mechanical objectivity is a common response when expertise is challenged and distrust is high (Porter, 1995), one might ask whether reinforcing this ideal is realistic (or even desirable) in light of the pervasive switch to digital media in scientific practice.