1. Neglected Aspects of Information and Specificity

Discussions of information and specificity in biology are now commonplace. Philosophers have long compared the relative importance of genes to other causes and many have concentrated on the concept of specificity as fine-grained influence in relation to difference-making accounts of causal explanation (e.g., Waters Reference Waters2007). Recently, a quantitative account of causal specificity using information theory was offered as a means to arrive at more accurate comparisons of different causes (Griffiths et al. Reference Griffiths, Pocheville, Calcott, Stotz, Kim and Knight2015). However, conflicting calculations of the amount of information in DNA sequence compared to the process of alternative splicing remain (Weber Reference Weber2017). In this literature, how information and specificity are used in practice—as opposed to how they can be understood in theory—has not been scrutinized adequately. This neglect is visible in the framing of questions about these concepts. Questions such as “What is specificity?” or “What is information?” predominate (Maynard Smith Reference Maynard Smith2000; Griffiths Reference Griffiths2001; Calcott Reference Calcott2014; Griffiths et al. Reference Griffiths, Pocheville, Calcott, Stotz, Kim and Knight2015); questions such as “How do scientists measure specificity experimentally?” and “When do biologists utilize information to explain?” are largely unexplored.

Here I address these neglected aspects by focusing on the measurement of specificity in practice, specifically positional information in embryonic patterning. My thesis is that the “amount” of information in a biological system is typically measured to answer particular research questions and does not support or refute general claims about the privileging of genetic difference makers, which have been central to discussions of information in biology (Waters Reference Waters2007; Griffiths et al. Reference Griffiths, Pocheville, Calcott, Stotz, Kim and Knight2015; Weber Reference Weber2017). The aim of this article is to show that there are novel conceptual issues related to the use of information and specificity in experimental biology that are orthogonal to these standard debates but also bear on them in interesting ways.

I first recall moments from the history of information and specificity in twentieth-century biology. This helps to account for why philosophical discussion moved in the direction of questions such as “What is biological information?” and away from the roles of information in scientific practice. Next, I explore the example of positional information in developmental biology and the measurement of specificity in segment formation during Drosophila ontogeny. This example displays practices associated with measuring information in an experimental context, especially estimating the amount of information at a particular time by averaging across different embryos and managing forms of error inherent to these measurements. I argue that the central motivation for measuring information in scientific practice is to explain particular phenomena rather than compare causes, which makes salient why these practices point to distinct conceptual issues. My analysis also supports a pluralist interpretation of specificity because the examined scientific practices vindicate at least three different senses of causal specificity advanced by philosophers.

2. Moments in the History of Information and Specificity in Biology

Conrad Waddington captured succinctly the ethos of information and specificity in twentieth-century biology.

Waddington’s reconstruction grasps an important explanatory intuition shared by biologists: defining features of biological systems result from something over and above physicochemical properties, and information (resonating with an ancient appeal to “form”) is a prime candidate. However, this intuition was sharpened alongside the development of computational theory and technology in the mid-twentieth century. Information theory and linguistic imagery, which were predominant at this time, influenced the framing of concepts in molecular biological research in different ways (Kay Reference Kay2000; Kogge Reference Kogge2012). This is especially visible in many of its core metaphors (transcription, translation, messenger, code, signal, libraries, etc.). However, much of this framing was more rhetorical than substantive (Sarkar Reference Sarkar1996) and warped expectations about what we should expect to find in biological research (Kay Reference Kay2000; Griffiths Reference Griffiths2001).

One of the most trenchant critics of the dangerous inferential potential of these informational metaphors was Susan Oyama (Reference Oyama1985). She challenged the increasingly common appeal to genetic information in order to explain three different features of biological systems: change, constancy, and variability. “Information is conceived to be a special kind of cause among all the factors that may be necessary for a phenomenon, the cause that imparts order and form to matter” (3). In contrast, she held that “information” is constructed during ontogeny through complex sets of interactions among multiple factors, only some of which are genetic. These sets of interactions are reliably transmitted across generations in distinctive patterns and thereby condition the tempo and mode of evolutionary processes. The implication is that the unit of evolution is not a segment of DNA but the entire developmental system where these sets of interactions are manifested.

The perspective offered by Oyama encourages a radical reconfiguration of how we understand evolutionary processes. The gene’s-eye point of view becomes suspect; development now has substantial relevance for how evolution works; and, therefore, standard neo-Darwinian evolutionary theory needs reformulation. This helps account for the emergence of developmental systems theory (DST), which can be seen as an elaboration and generalization of Oyama’s perspective with special attention to conceptual ramifications relevant to reformulating theoretical foundations for evolutionary biology (e.g., Griffiths and Gray Reference Griffiths and Gray1994). A key debate in this context revolves around the parity thesis, which (following Oyama) holds that various factors interacting in a developmental system are all bearers of or contributors to information in organisms, although not necessarily equally in all cases. An implication of the parity thesis is that genetic difference makers have no special or general status as distinctive information bearers in living systems. Any observed privileging of genetic factors in the experimental practice of biology (e.g., preferentially selecting genetic difference makers to causally explain the properties of organisms) is interpreted as merely pragmatic. To say that the parity thesis has been controversial is an understatement (see, e.g., Moffatt Reference Moffatt2011).

Although Waddington encapsulated an ethos among researchers that sought deep connections between character specificity and information in biology, he recognized that there were inherent difficulties in making these claims. “This word [information] has been given a very definite meaning in a body of thought known as ‘Information Theory’. … Unfortunately, the definition, although precise, is also very limiting, and excludes most of what we are really interested in” (Waddington Reference Waddington1977, 140). A similar skepticism was later advanced by philosophers who shared a concern that information theory and linguistic imagery inappropriately influenced the framing of conceptual metaphors in molecular biology (Kay Reference Kay2000). “There is no clear technical notion of ‘Information’ in molecular biology. ‘Information’ is little more than a metaphor that masquerades as a technical concept and leads to a misleading picture of the conceptual structure of molecular biology” (Sarkar Reference Sarkar1996, 857). This skepticism extended to the application of information to understand embryogenesis: “on any definition of information the view that development is the expression of genetic information is misleading” (Griffiths Reference Griffiths2001, 394).

Although the historical use of information in biological practices prompted these critical discussions, philosophical questions became reoriented over time, especially following the introduction of DST. Instead of focusing on how information is used in practice (perhaps quite inappropriately), scrutiny shifted to the concept of genetic information itself and whether different conceptions were applicable to or explanatorily relevant for biological systems. Most of these discussions were theoretical in nature and abstractly structured (Is there a justification for privileging genes in investigation and explanation? Is there semantic information in DNA sequences?). The resulting landscape of options is diverse and includes information skepticism, information democratization, and genetic information privileging. Related alternatives involve shunning the concept of information in favor of specificity to account for the approach to causal selection widespread in genetics: “understanding why DNA is a distinctive cause requires invoking a concept of causal specificity as well as concepts of actual difference making” (Waters Reference Waters2007, 572). Although this encouraged some attention to scientific practices, such as methods of causal selection in experimental settings, the focus remained primarily on fine-grained influence in relation to difference-making accounts of causal investigation and explanation.

Recently, in part as a response to arguments advanced by Waters and others, a quantitative measure of causal specificity based on information theory was articulated with the aim of yielding more accurate comparisons of different causes (Griffiths et al. Reference Griffiths, Pocheville, Calcott, Stotz, Kim and Knight2015). On this account, the specificity of a causal variable is how much mutual information interventions on that causal variable carry about the effect variable. In related work, Calcott (Reference Calcott2014) moved beyond the focus on DNA-protein relationships to understand genetic information in regulatory networks using evolutionary game theory. However, in both of these cases, the focus is still on what genetic information is and what properties it can have in a biological system, not the practice of scientists. The various philosophical options on the landscape of information and specificity are primarily intended to substantiate or undermine the privileging of genetic causation (or approaches) in biological investigation and explanation: “to provide a rationale for the appeal of a certain kind of explanatory strategy that … consists in focusing explanations on biological entities that are often described as information bearing” (Weber Reference Weber2017, 574). Yet, consensus has not coalesced under the analytical conditions of formalizing the notion of specificity, leaving divergent assessments in the amount of information available in DNA sequence and the process of alternative splicing (Griffiths et al. Reference Griffiths, Pocheville, Calcott, Stotz, Kim and Knight2015; Weber Reference Weber2017). At a minimum, this impasse suggests that more effort is needed to understand practices in which scientists measure information or specificity experimentally. While this may not adjudicate these philosophical differences, it could expose new conceptual issues related to neglected aspects of information and specificity in biological practice.

3. Positional Information in Developmental Biology and the Measurement of Specificity

Positional information is the idea that cells acquire an identity in relation to their relative position in a bounded region of the embryo. Once acquired, cells differentiate into distinct types based on characteristic patterns of gene expression, which facilitates the formation of higher-level patterns, such as repeating stripes of cells that eventually yield morphological segments or other anatomical features. The concept of positional information was formulated originally by the biologist Lewis Wolpert. In reflections on the concept’s significance, Wolpert has been circumspect about the empirical support for its key aspects: “There is no good evidence for the quantitative aspects of any of the gradients and details [of] how they are setup” (Reference Wolpert2011, 359). “While there are many encouraging studies, we still do not know the molecular basis of positional information … nor do we have convincing evidence of how positional values are specified or interpreted” (Wolpert Reference Wolpert and Wassarman2016, 605). Given this situation, it is unsurprising that biologists are looking for evidence that speaks to these aspects of positional information.

During Drosophila embryogenesis, the segmental architecture of the organism is progressively established through a hierarchy of gene expression called the segment polarity network (Wolpert et al. Reference Wolpert, Tickle, Jessell, Lawrence, Meyerowitz, Robertson and Smith2010, 70–81). Throughout this process, groups of cells take on new identities as a function of their spatial position in the embryo. First, maternal effect genes determine the anterior-posterior (A/P) axis of the embryo through contrasting concentration gradients and inhibitory interactions. These maternal effect genes also induce the expression of “gap” genes. These gap genes refine this basic orientation into several broad regions, while also initiating the next stage of gene expression (“pair-rule”). Pair-rule genes establish parasegment units through patterns of differential protein accumulation and trigger a subsequent cascade of gene activity (“segment polarity”), defining boundaries for each compartment of cells that will become a segment. Finally, through regulatory interactions with gap, pair-rule, and segment polarity genes, Hox genes are activated that confer an identity on each segment, such as a thorax segment containing a pair of legs.

Dubuis and colleagues (Reference Dubuis, Tkačik, Wieschaus, Gregor and Bialek2013) measured positional information in the spatial expression of the gap genes at an early stage of Drosophila embryogenesis. They found that the amount of information is enough to specify the location of cells along the A/P axis within an error rate of 1% and corresponds closely to what is needed for each cell to have a unique identity within the regions established at this stage. This procedure is in sharp contrast to what has been under scrutiny in previous philosophical discussions, which have concentrated on DNA sequence or regulatory network information, theorized apart from research questions. However, for these researchers, the question is specific: Do the expression profiles of gap genes provide enough information to specify the exact pattern of development (cell-by-cell) along the A/P axis? And this question is responsive to the worry raised by Wolpert: “less is known quantitatively about how much information is being represented by the expression levels in individual cells” (Dubuis et al. Reference Dubuis, Tkačik, Wieschaus, Gregor and Bialek2013, 16301).

In order to answer this research question, Dubuis and colleagues developed a mathematically precise version of the statement that “the expression level g of a gene provides information about the position x of the cell” (Reference Dubuis, Tkačik, Wieschaus, Gregor and Bialek2013, 16302). With this in hand, they can generalize to information conveyed by the combined expression of multiple genes. Instead of trying to compare the amount of information carried by genes with other factors, or calculate the information carried by a gene sequence, the goal was to quantify the amount of information in a gene expression pattern throughout the embryo at a time to understand how cells differentiate and produce regionalized patterns associated with this step in the segment polarity network. The relevant measure of mutual information is not of a DNA sequence relative to a protein sequence but of an amount of gene expression relative to a location along the axis of an embryo.

An awareness of the form of mutual information is crucial because it helps the researchers concentrate on important material features of their experimental system. Gap gene protein molecules are present at low concentrations and even low absolute copy numbers. Thus, gene expression is inherently noisy, and this noise limits the transmission of information. Estimating the mutual information that one gene expression level provides about position also requires a good estimate of the conditional distribution of expression levels for cells at a position. Therefore, multiple embryos must be measured to secure an estimate of the conditional distribution of expression levels. As a consequence, variations in expression levels of one gene within an embryo compared to another embryo, as well as variation between individual cells in the same embryo, are not included in the estimate. The final calculation of a conditional distribution will be based on averages of mutual information estimates across different embryos and their cells in aggregate. Additionally, the measurement of information is time indexed (i.e., at one stage) and the conditional distribution calculated is only applicable at this stage.

Tracking time in embryos requires more than absolute chronology. There are a number of ways to conceptualize developmental time, including a sequence of events or a succession of defined intervals. One of the most common strategies for conceptualizing time in developmental biology is by embryological events, often for all of an organism’s development. A set of numbered or named stages for ‘normal’ embryonic development allows researchers in different laboratory contexts to obtain standardized results that can be compared and contrasted (Hopwood Reference Hopwood2007). In addition to whole organism staging, researchers also establish standardized sequences using developmental events related to individual organs, such as a wing. For the measurement of positional information in the gap genes of Drosophila, researchers followed this type of strategy and used the length of the cellularization membrane as a proxy for elapsed time in embryos measured.

At this point we can see a number of novel conceptual issues emerging, including choices about the conception of time and material considerations that attend tracking time precisely to make measurements of the information contained in gene expression patterns at a stage in a physical space. Additionally, there is a need to average over values gleaned from different embryos to establish a conditional distribution of expression levels for cells at a position along the A/P axis. Further issues include the fact that quantitative imaging at the poles of the embryo is more difficult to accomplish and there are additional genes associated with terminal patterning (i.e., not gap genes). Therefore, researchers confined their attention to the central 80% of an embryo’s A/P axis. Also, even though precision measurements of expression are typically done one gene at a time, researchers needed to know the joint distribution of all expression levels. This joint distribution was measured using four different colors of immunofluorescent staining, which posed a difficulty of avoiding artifacts due to spectral cross talk among the fluorescent signals. The authors devoted an entire separate paper to addressing these types of questions (Dubuis, Samanta, and Gregor Reference Dubuis, Samanta and Gregor2013).

Returning to the averaging of gene expression levels from different embryos, the standard deviation of expression across the gap genes is <10% of the maximum expression level. Thus, measurements of these levels must have a higher accuracy. Errors were minimized by attention to the orientation and age of the embryos and by comparing large numbers of embryos stained in a single batch. Standards for evaluating quantitative accuracy estimates can be formulated out of other developmental patterns, both morphological and molecular. The cephalic furrow can be observed in live embryos and has a position along the A/P axis that is reproducible with an accuracy of ∼1% of embryo length. Pair-rule gene expression, which occurs after gap gene expression, exhibits peaks and troughs of expression along the A/P axis, with positions that are reproducible to ∼1% of embryo length. The ability to achieve such high precision in measurement of expression and morphology permits an evaluation of how closely the embryo approaches optimality in its use of available mutual information from the joint distribution of all expression levels. A comparison between the observed positional information and the maximum obtainable if the embryo could adjust the distribution of nuclear positions to match the positional error perfectly fell within a few percentage points of the optimum calculated.

Overall, close attention to these conceptual issues makes it possible to address the original research question of whether expression profiles of gap genes provide enough information to specify the cell-by-cell pattern along the A/P axis. Individual genes carry twice as much information as would be expected if the expression patterns consisted only of on/off domains separated by sharp boundaries. Together, the four gap genes carry enough information to define a cell’s location with an error of ∼1% along the A/P axis of the embryo. This precision is nearly enough for each cell to have a unique identity, and the pattern of information is almost constant along the length of the embryo, which suggests optimality in the transmission of information from upstream inputs (maternal effect genes) to the output of gap gene expression.

4. Motivations for Measurement Practices

Dubuis and colleagues never compared the amount of information carried by DNA sequences with that of other factors or calculated the information carried by a gene compared with alternative splicing possibilities. Instead, they quantified the amount of information in individual and combined gene expression patterns along the A/P axis of the embryo at a specific time to understand how cells differentiate and produce morphological patterns. The relevant measure of mutual information was not that of a DNA sequence relative to a protein sequence, or that of an abstract regulatory network that passes information from one gene to another, but rather an amount of expression relative to a location along the axis of the embryo at one stage. These differences help reveal a number of conceptual issues orthogonal to standard philosophical discussions, such as conceptualizing time appropriately, containing error to achieve sufficient precision, accounting for noise due to low concentrations of molecules, and avoiding fluorescent artifacts to capture accurately the joint distribution of expression levels.

The aims of recent philosophical analyses of information and specificity include comparison of different causes, justification of privileging genetic factors, demonstration of the value of specificity to facilitate ongoing inquiry, and a vindication of the applicability of concepts of information. A comparison of different causes links tightly to discussions of parity and DST: “the aim of producing a measure of causal specificity was to use it to compare different causes of the same effect” (Griffiths et al. Reference Griffiths, Pocheville, Calcott, Stotz, Kim and Knight2015, 538). This is also true of attempted justifications for privileging genes over other factors as a consequence of their differential informational specificity: “to provide a rationale for the appeal of a certain kind of explanatory strategy that … consists in focusing explanations on biological entities that are often described as information bearing” (Weber Reference Weber2017, 574). Griffiths and colleagues express the aim of showing how a conceptually precise notion of specificity can facilitate scientific investigation: “Our motivation is the same as that of Waters and Weber: clarifying the notion of causal specificity may elucidate the notion of biological specificity and facilitate the study of specificity in actual biological systems” (Reference Griffiths, Pocheville, Calcott, Stotz, Kim and Knight2015, 531). Finally, in response to informational skeptics, there is the aim of vindicating the applicability of information concepts in biology: “positional information is one example where informational concepts in signaling systems can be clearly applied, so the signaling framework can vindicate information talk” (Calcott Reference Calcott2014, 888).

The aims of scientists such as Dubuis and colleagues are strikingly different. They are concerned with characterizing and explaining a particular biological phenomenon. “Do the broad, smooth expression profiles of the gap genes … provide enough information to specify the exact pattern of development, cell by cell, along the anterior/posterior axis? How much information does the whole embryo use in making this pattern? Is it possible … that the information transmitted through these regulatory networks is close to the physical limits set by the irreducible randomness of counting individual molecular events?” (Dubuis et al. Reference Dubuis, Tkačik, Wieschaus, Gregor and Bialek2013, 16301). There is no concern with the comparison of different causes. The well-established gene expression patterns of the segment polarity network are assumed; the issue is whether a sufficiently precise quantification is achievable: “we need to measure positional information quantitatively, in bits” (16301). One might argue that these aims are an implicit justification of privileging genetic factors in investigation and explanation, but this is implausible because the lead experimentalist has explicitly investigated the role of other causal factors in development, such as physical forces (e.g., Polyakov et al. Reference Polyakov, He, Swan, Shaevitz, Kaschube and Wieschaus2014). The focus is on quantification and measurement to characterize and explain, not a prior commitment to the privileged informational capacity of genes. Although this type of study demonstrates ostensively the value of informational concepts in actual inquiry, the focus was on whether the quantification they facilitated could provide an explanation of the embryonic patterning. This renders moot the aim of vindicating the applicability of concepts of information to biological systems.

In addition to the novel conceptual issues that arise from this investigation of the measurement of positional information in developmental biology, there are implications for traditional philosophical debates. First, against informational skeptics, we have a clear technical notion of information playing a significant explanatory role in developmental biology. (Representatives of the skeptical position can rightly observe the late date at which it appears.) Second, consideration of noise—and not just information—is extremely important; this was necessary to account for the robustness of the combined pattern of gene expression. Noise also relates to questions of measurement error that played a central role in this study and must be considered to secure estimates of information from multiple samples. Third, the calculation of an amount of information is relative to a time in embryogenesis. Almost all philosophical discussion has ignored time indexing (but see Calcott Reference Calcott2014).

Perhaps a more substantial contribution of this analysis to philosophical discussion is that it supports an unrecognized pluralist interpretation of causal specificity. At least three different senses of causal specificity exist in earlier literature. Griffiths et al. (Reference Griffiths, Pocheville, Calcott, Stotz, Kim and Knight2015) argue that specificity is evaluated by assessing the total potential variation of a cause variable with respect to its effect. This is precisely the kind of specificity that Dubuis and colleagues (Reference Dubuis, Tkačik, Wieschaus, Gregor and Bialek2013) needed to make the argument from optimality (the maximum obtainable positional information available if the embryo could adjust the distribution of nuclear positions to match the positional error). Waters (Reference Waters2007) argues for a conception of specificity in terms of the actual variation manifested in select populations. This conception of specificity is needed to make the argument for measurements of individual and joint gene expression patterns, including their inherent noisiness. Weber (Reference Weber2017) argues that specificity is not with respect to actual variation in a population or total potential variation but is restricted to a range of biologically relevant possible values. This type of specificity was used to restrict the argument for the measured value of positional information being explanatory to the normal functioning of gap gene expression during Drosophila segmentation. Thus, this example speaks directly to the legitimacy of multiple proposals regarding causal specificity.

5. Conclusion

The current analysis does not resolve ongoing debates among philosophers about information and specificity in biology. Instead, it helps to account for why the discussion took its current shape by recalling moments in its history and how this trajectory tended to neglect questions about practice. Additionally, my analysis of positional information during segment formation documents novel conceptual issues arising from the practices associated with measuring specificity in an experimental context (e.g., estimating the amount of information at a “time” by averaging across measurements from different embryos and containing error to achieve sufficient precision). These issues can be traced back to motivations for measuring information and specificity in practice, such as explaining a particular phenomenon, that differ from the aims advanced by philosophers, such as comparing different causes or justifying a privileging of genetic factors in investigation and explanation.

Although the conceptual issues identified in these scientific practices are largely orthogonal to standard philosophical debates about information in biology, they do bear on them. Informational skeptics must account for the clear technical notion of information playing a significant explanatory role in developmental biology. The presence of noise and the time indexing for measurements of information need more scrutiny. And, different senses of causal specificity may map onto different features of scientific practice and not be mutually exclusive. The measurement of information and specificity in practice does not support or refute general claims about the privileging of genetic difference makers, but it does expose neglected aspects that require more philosophical analysis, especially the aim of biologists to measure the amount of specificity in a biological system for the purpose of answering empirical research questions.