Missed connections: Exploring features of undergraduate biology students’ knowledge networks relating gene regulation, cell–cell communication, and phenotypic expression

INTRODUCTION

Examining biology from a systems perspective has become the prevailing paradigm in recent decades, which is in contrast with historical investigations using reductionist methodology to study pieces of a larger system. While an important and relevant approach, reductionism cannot predict nor explain system level properties, and thus, examining systems as a whole is critical to fully understanding a biological system (van Regenmortel, 2004). Biologists at the 2009 Nobel Symposium similarly agreed that to completely understand biology, one must “embrace a ‘fifth great idea’ that can be summarized as follows: multi-scale dynamic complex systems formed by interacting macromolecules and metabolites, cells, organs, and organisms underlie most biological processes” (Vidal, 2009, p. 3891). In other words, understanding biological phenomena requires seeing how interactions of system components at different levels of organization form a complex system.

The systems perspective to biological inquiry has made its way to guidelines for undergraduate biology curricula (NRC, 2009; AAAS, 2011). Vision and Change recognizes ‘Systems’ as one of five core concepts for biological literacy in undergraduate education and emphasizes the need to study dynamic interactions of system components at all levels of organization (e.g., molecules, ecosystems, and social systems) and identify how interactions at lower levels of organization result in the functional properties of entities at higher levels of organization (AAAS, 2011). Therefore, to understand biology and the concept of systems, undergraduate biology students are tasked with integrating concepts across scientific domains, thinking about entities interacting at different levels of organization, and dissecting complex systems (AAAS, 2011). For example, to completely explain how organ systems function, a student must understand the context of the system at large and know about processes happening at the lower levels of organization.

An understanding of biological systems would enable students to engage in systems thinking which is comprised of skills and ways of thinking that allow for an understanding of the structure and functioning of a system (reviewed in Momsen, et al., 2022). Recently, a four-level hierarchical framework of skills needed for biology systems thinking (BST) was proposed which contains four levels each with three associated skills. The authors contend that this framework will allow for targeted research and assessments to reveal student competence and guide instruction (Momsen et al., 2022). While our work does not aim to examine systems thinking specifically, it is related to, and could have implications for instruction about, some of the skills in the BST framework.

Biological Context of the Current Work

Across biological fields of study and subdisciplines, organisms sense and receive environmental information, input these cues into feedback processes, and produce a response appropriate to their changing environment (Brownell et al., 2014). For example, consider the process of bacterial chemotaxis (van Mil et al., 2013). Bacterial cells sense chemo attractants and chemo repellents in their environment via receptors embedded in the cell’s membrane. This sensory input translates into a signaling pathway that activates intracellular proteins, which modulate levels of phosphorylated CheY proteins. When the concentration of phosphorylated CheY proteins is high, CheY proteins affect the motors of the flagella, causing a change in a bacterium’s movement pattern and orientation. Put simply, the bacteria sense the changing environment, translate the sensory information into a signal that affects intracellular proteins, and those proteins’ activity results in a response. Describing how cells and organisms sense, signal, and respond is a useful framework for explaining how biological phenomena occur in a system. In the context of our study, we have defined three subsystems relevant to sensing, signaling, and response processes: 1) gene regulation, 2) cell–cell communication, and 3) phenotypic expression. We consider these subsystems to be inclusive of processes and concepts operating within larger biological systems and organisms. Gene regulation is involved in signaling and response processes, cell–cell communication is relevant to sensing and signaling processes, and phenotypic expression is involved in response and sensing processes. We operationalize each of these three subsystems, with examples and select citations from basic science and education literature, in Table 1. Accordingly, within our study context, a biological system would be bounded by genes at the small scale and the immediate local environment of a cell or an organism at the largest scale, depending on how students thought about the subsystems.

The biology education community has demonstrated the importance and relevance of these three subsystems in understanding and explaining biological phenomena (Brownell et al., 2014; Couch et al. , 2015; Summers et al., 2018; Couch et al., 2019; Semsar et al., 2019). Previous work has explored students’ conceptual models of the relationships between genes and phenotypes and genes to evolution in the context of biological systems (Dauer et al., 2013; Bray Speth et al., 2014; Reinagel and Bray Speth, 2016). However, the literature has yet to deeply explore the ways in which students describe these three subsystems and the connections between them.

Research Focus

To help students develop competence in explaining biology in an integrated and mechanistic manner, we aimed to better understand the ways in which they think about and connect different fundamental subsystems in biology: gene regulation, cell–cell communication, and phenotypic expression. These three subsystems are relevant across biology subdisciplines and paramount to understanding biology. Using the theory of knowledge integration and mechanistic reasoning to guide our research study design and analysis, we sought to answer the following research questions:

1. How do undergraduate biology students define three foundational subsystems in biology (i.e., gene regulation, cell–cell communication, and phenotypic expression)?

2. What are the features of undergraduate biology students’ knowledge networks regarding the three foundational subsystems in biology?

Theoretical Underpinnings and Conceptual Frameworks

To support student understanding of biology and biological systems from an integrated and mechanistic perspective, we aimed to characterize the ways in which students define and connect the three subsystems (i.e., gene regulation, cell–cell communication, and phenotypic expression). Specifically, we sought to examine their functional definitions of the subsystems, what relationships students form between the subsystems, and the qualitative nature of those relationships. To accomplish this, we draw from both the theory of knowledge integration and from work in the field of mechanistic reasoning in the design and analysis of our research.

Learning in the sciences requires grappling with conflicting ideas and knowledge gained from instruction or experience in order to understand complex phenomena and new contexts (Clark and Linn, 2003). This dynamic process of knowledge integration involves “linking, connecting, distinguishing, organizing, and structuring ideas about scientific phenomena” and culminates in a network of knowledge (Clark and Linn, 2003, p. 452). We define a knowledge network as the organizing body of knowledge pieces, composed of nodes representing concepts and linkages between nodes. This is similar to knowledge systems (Vosniadou, 2013, p. 22), but for the purposes of our work, we are using ‘network’ to build from previous literature (Southard et al., 2016) and to avoid confusion with the biological systems and subsystems in this study. As a student learns more about a concept, they must reorganize their knowledge to integrate new nodes and connections.

Engaging students in the process of knowledge integration prepares them to more deeply learn science, because students with an integrated understanding of scientific concepts can readily process new ideas into their knowledge networks. Indeed, it has been argued that integrative frameworks are required to understand molecular and cellular biology (Southard et al., 2016) and that evolution should be used as a core idea of students’ knowledge structures to understand biology in general (Tibell and Harms, 2017). Additionally, building an integrated knowledge network to deeply learn biology is aligned with the ideas in Vision and Change, which call for integrated understanding of biology concepts (AAAS, 2011). Ideas that are not integrated into the knowledge network become isolated and lost over time. Thus, it is important to assist students in resolving conflicts that prevent the formation of normative connections to new ideas. These conflicts may arise if students hold nonnormative ideas about certain concepts or have previous knowledge or experience that do not agree with new information. Providing opportunities for students to confront their conflicts and integrate scientific concepts into their knowledge networks allows for the knowledge to persist across time and, therefore, make students lifelong learners and advocates of science (Clark and Linn, 2003).

Knowledge integration serves as an appropriate framework in the context of this research study, because it allows us to elicit and analyze student knowledge of the three subsystems and their knowledge networks. It has been previously used in other work to uncover students’ knowledge networks of DNA replication, transcription, and translation (Southard et al., 2016). The theory of knowledge integration posits that students must process and sort their ideas when integrating new knowledge into their knowledge networks. Guided by this tenet, we first activate students’ prior knowledge (Hammer et al., 2005) of the three subsystems by asking how and why they occur. In addition, for ideas to be properly connected to other knowledge elements and persist in knowledge networks, the ideas and connections must be normative (Clark and Linn, 2003). Thus, we analyze students’ conceptual ideas of the three subsystems for correctness. We then reveal students’ knowledge networks by explicitly asking them to describe how the three subsystems are related. Using their explanations, we are able to study the ways in which students integrate the three subsystems in their knowledge networks (Table 2).

For knowledge networks to be useful tools for understanding and reasoning about biological phenomena across disciplinary contexts, connections between concepts must be specified and well-supported. Embedded into these knowledge networks are how entities interact to give rise to processes in one part of a system and dynamically affect other processes in that complex system. Building connections that are mechanistic in nature can support deep integration of knowledge and help students reason through processes occurring in a system. Mechanisms are defined by Machamer and colleagues as “entities and activities organized such that they are productive of regular changes from start or setup to finish or termination conditions” (Machamer et al., 2000, p. 3). By this definition, in order to describe a mechanism, one must identify the activities that produce change, the things or entities which engage in those activities, and the conditions of time and space. Inclusive in the conditions of space is knowing where entities and activities in a mechanism are spatially organized or located in relation to other structures (Russ et al., 2008; van Mil et al., 2013; Lira and Gardner, 2020). It has been argued that molecular mechanisms must also include entities and activities occurring at different levels of organization (van Mil et al., 2013), which can range anywhere from the molecular and submolecular levels to higher levels such as organismal and population levels.

We leverage mechanistic reasoning in the methods to prompt for mechanisms and in the data analysis to investigate the features of their explanations (Table 2). During knowledge activation, we intentionally framed questions as “how”, in an effort to elicit student knowledge in the form of a mechanism. This is because previous work has identified that how answers are mechanistic in nature (Mayr, 1961; Machamer et al., 2000; Abrams and Southerland, 2001). Thus, we analyzed students’ answers for features of mechanisms (e.g., entities, activities, and causal sequence of events). When revealing students’ knowledge networks, we similarly pose the relationship question in the form of how to elicit mechanistic explanations that connect the three subsystems. We use ideas derived from the work of mechanistic reasoning to investigate the qualitative nature of the connections students specify between the three subsystems. Drawing from previous research, we sought to evaluate how students justified connections between subsystems, which can range from pure association (Southard et al., 2016) to causal linkages (Becker et al., 2016) to mechanistic connections (Machamer et al., 2000). We also evaluated the nature of the content within each of the connections to describe what level of organization the entities reside and what localization the processes occur (Russ et al., 2008; van Mil et al., 2013; Lira Gardner, 2020).

Thus, we use the theory of knowledge integration and mechanistic reasoning in combination to activate, reveal, and characterize students’ knowledge networks. From these data, we can investigate how students connect knowledge elements and whether these connections are supported by mechanistic features (Table 2).

RESEARCH METHODS

Student Population

We recruited student participants for our interviews via an email invitation sent out to all undergraduate biology majors at a large public midwestern university with very high research activity during Fall 2019. Out of the nine students interviewed, three were sophomores, three were juniors, and three were seniors. At this institution, students can specialize within the biology major, and the distribution in our participant sample was: one Genetics; one Ecology, Evolution, and Environmental Biology; three Neurobiology and Physiology; and four General Biology. Interviews lasted between 30 to 90 min, and all participants were compensated with a $20.00 Amazon gift card for their time.

Our university’s Institutional Review Board evaluated and approved our project methods for recruitment and data collection before carrying out our research (IRB #1806020745).

Interview Materials Development

We leveraged the theory of knowledge integration (Clark and Linn, 2003) and mechanistic reasoning in the design of the interview protocol to elicit and characterize student knowledge and the features of their knowledge networks. Specifically, we sought to reveal what knowledge students have about the three subsystems and how students integrate the three subsystems in their knowledge networks. Previous work using knowledge integration and mechanistic reasoning as frameworks asked students to describe what they know about each process, including “what is physically happening, who are the players, and where does it happen” (Southard et al., 2016, supplement). We were interested in participants’ spontaneous responses and, therefore, opted for less structure in our interview protocol by plainly asking students to explain how each of the subsystems occur. These questions were intentionally framed as how in an effort to elicit student knowledge in the form of a mechanism, because how answers are mechanistic in nature (Mayr, 1961; Machamer et al., 2000; Abrams and Southerland, 2001). Following the work by Southard and colleagues (2016), we also asked students why each of the subsystems occur to help activate prior ideas about the relationships between subsystems and students’ broader understanding of their role(s) in biology. This is because why answers account for the origins, purposes, or rationale for the phenomenon (Mayr, 1961; Abrams and Southerland, 2001). Thus, our how probes revealed student knowledge about how the subsystems function and our why probes prepared students to reveal the ways in which the subsystems are integrated.

Before collecting data for this study, four experts (three STEM education researchers and one biology educator) not directly involved in this research study reviewed the interview questions. Additionally, we piloted the interview protocol with seven undergraduate biology major students and revised questions for clarity based on feedback.

Data Collection

We conducted semistructured, think-aloud interviews, asking participants to narrate their thought processes while performing tasks and answering questions (Ericsson and Simon, 1980). We performed think-aloud interviews for the purpose of eliciting the ways in which the participants approached the task. While this method may reveal the process by which participants came to their answers, it is limited in that there is no guarantee that the participant narrated all their thoughts. We held student interviews face-to-face in a private office, and sessions were audio-recorded. The data reported here are from a larger study aimed at revealing students’ knowledge networks and mechanistic reasoning in both an open and specific biological context. In this article, we describe data collected from the open context portion of the interview.

One researcher performed all interviews using a standardized interview protocol. The interview consisted of two main phases aimed at activating and revealing knowledge.

Activating Prior Knowledge.

We first asked participants to explain how the three biological subsystems occur and then describe why cells undergo these subsystems (adapted from Southard et al., 2016). The purpose was to activate knowledge and reasoning about the subsystems and for the interviewer to get a sense of the participants’ prior understanding (Hammer et al., 2005). Afterward, we then provided our participants with textbook-derived definitions of the three subsystems and encouraged participants to reflect on the definitions and identify areas of similarity and potential gaps in their knowledge. The purpose was to provide participants with at least basic, common knowledge about the subsystems from which to build for the interview. The specific questions used in this phase are: 1) Explain to me how the process of gene regulation occurs, 2) Why does a cell undergo the process of gene regulation?, 3) Explain to me how the process of cell–cell communication occurs, 4) Why does a cell undergo the process of cell–cell communication?, 5) Explain to me how phenotype is regulated, 6) Why is it important that the genotype and phenotype are linked?, and 7–9) Please take a moment to read this description of __(insert subsystem name)__. What do you notice in the definition that is similar or different from what you said?

Networks and Relationships.

In the second phase, we then asked participants the question “How are these three processes related to each other?” The purpose of this question was to prompt participants to integrate their previous thoughts of the subsystems with any new ideas cued from the definitions and describe how the concepts are connected. This question, thus, explicitly elicited the relationships in their knowledge networks of the three subsystems, which we were interested in characterizing.

Analysis

We analyzed verbatim audio transcripts using deductive and inductive coding (Creswell, 2013). For data from the Activating Prior Knowledge phase, we originally intended to analyze how responses for features of mechanistic reasoning and why responses for correctness of connections to biological concepts from students’ spontaneous explanations (i.e., without additional prompting). However, students’ how responses were rarely in the form of mechanisms and the richness of their answers warranted another approach for analysis. We first used iterative, inductive coding to develop content codes describing ideas and concepts the participants used to explain the how and why of the subsystems. Simultaneous coding (Saldaña, 2013) was then employed to characterize each content code by two additional dimensions: the form of the response and the correctness of the response. For response form, we categorized each content code a participant provided as either a what, how, or why response to help reveal the nature of student answers, if not by a fully articulated mechanism (Mayr, 1961; Machamer et al., 2000; Abrams and Southerland, 2001; Hmelo-Silver and Pfeffer, 2004). For example, one content code for how gene regulation occurs was the effect of external factors on a cell’s gene regulation. A student may provide a what response that this effect exists, a how response describing the mechanism of a factor affecting gene regulation, or a why response describing the functional purpose of external factors on a cell’s gene regulation. It is important to note that we coded each content code holistically for response form. Describing components or players (the what) is inherent in explaining a mechanism (the how) or the purpose (the why). For a participant to receive a what code, they only described the things related to the content code and did not provide a mechanistic or narrative-based explanation. Because normative ideas are necessary to make stable connections across knowledge elements in knowledge networks (Clark and Linn, 2003), we evaluated each content code for scientific normativity for later analysis with the knowledge networks. Thus, for each unique statement uttered by a participant, we assigned a content code, the form of the response (i.e., what, how, or why), and its correctness.

Based on responses to the relationship question in the Networks and Relationships phase, we generated models of the participants’ knowledge networks of their connections between gene regulation, cell–cell communication, and phenotypic expression. To create the models, we first analyzed responses for direct naming of or descriptions of the three subsystems. We then identified words, phrases, or descriptions that linked one subsystem to another (e.g., cell–cell communication affects gene regulation). A directional arrow was then inserted between the subsystem names to represent the descriptive relationship the student uttered (e.g., cell–cell communication → gene regulation). The end result is a pictorial representation of the three subsystem names and arrows between them as described by the participant. By probing for student knowledge of the subsystems in the Activating Prior Knowledge phase using prompts guided by knowledge integration, students were primed to discuss their knowledge networks in the Networks and Relationships phase. It is important to note that the knowledge network models we produced are a pictorial translation of the participants’ verbal response. Thus, these models may not fully represent the participants’ knowledge network. Member checking did not occur as these representations were made after interview completion. However, our methods of generating these models are similar to other work where they coded interviews for evidence of specific concepts and identified relationships between structures and their associated functions and behaviors (Hmelo-Silver and Pfeffer, 2004; Hmelo-Silver et al., 2007).

We then characterized the participants’ knowledge networks in four ways: 1) model structure, 2) nature of connections, 3) level of organization of the players, and 4) localization of the processes. For 1) model structure, we analyzed the ways in which participants connected the subsystems, and then we binned models into categories based on similarities in the overall structure. (Please note we use the word ‘phenotype’ in the model names and depictions for brevity.) For 2) nature of connections, we characterized the nature of each arrow relationship in the knowledge networks to describe how the participant justified the connection. Guided by previous work, we analyzed arrow connections for evidence of either a Mechanistic (Machamer et al., 2000), Specified Causal (Becker et al., 2016) Unspecified Causal, or Associative connection (Southard et al., 2016; Table 3). To examine 3) the level of organization of players (van Mil et al., 2013), we scanned students’ explanations for things (i.e., nouns or objects such as cells, genes, and signaling molecules) and then coded its appropriate level of organization. We purposefully use the term ‘players’ as opposed to ‘entities’ to describe these data. The canonical definition of entities are things that have associated activities that produce change in a mechanism (Machamer et al., 2000), and we found that participants did not often describe entities, but rather things with no activity and/or without properties (e.g., spatial relations, orientations, structural properties, and state). In our data set, we identified the following levels of organization: Undefined, Environmental, Organismal, Cellular, Macromolecular, and Molecular (Table 3). Additionally, we found that participants often used processes as players in their explanations. We chose not to include these instances in our level of organization coding, because processes are comprised of multiple different players interacting often on varying levels of organization. Lastly, as the subsystems occur in and across different spatial locations within natural systems, we coded the 4) localization of the processes (Russ et al., 2008; van Mil et al., 2013; Lira Gardner, 2020) occurring in the participants’ explanations. We analyzed localization based on the descriptor text surrounding the verbs or phrases explaining how the process is occurring. For our data, we identified our localization areas as Unspecified, Outside a Cell, At the Cellular Membrane, and Inside a Cell (Table 3). While some participants were explicit about where the process was happening (e.g., in the extracellular fluid), some of the localization coding had to be inferred from contextual clues in their explanation. However, usage of process names did not automatically denote a specific localization code. For example, saying that “cell–cell communication goes to gene regulation” would not qualify for any specific codes (e.g., Inside a Cell even though gene regulation occurs intracellularly) because no information was given for where or how the participant was conceptualizing the explanation. Additionally, we found instances of students describing localizations that did not fit into our four categories, but these were very infrequent and diverse and, thus, not included in the results.

Normative Knowledge

Per the framework of knowledge integration, we needed to analyze students’ answers for correctness. Thus, we derived the definitions for the three subsystems presented to participants at the end of the Activating Prior Knowledge phase based on glossary definitions in Molecular Biology of the Cell (Alberts et al., 2008) and Life: The Science of Biology (Purves et al., 2004; see Supplemental Material for definitions). While sufficient to establish basic definitions, consulting these sources did not offer nuanced, content knowledge for functional definitions nor clear, normative knowledge for the relationship question in the Networks and Relationships phase. In an effort to identify features of normative answers for our student data, we recruited six biology instructors for interviews from the same research university during Summer 2020, under the same approved protocol as student participants (#1806020745). We personally invited specific biology faculty who currently teach or had previously taught foundational and introductory courses to biology major students. While their areas of research expertise varied (e.g., molecular, cellular, and developmental biology), they all indicated that they teach all three subsystems in their classes. However, two instructors noted they emphasize one subsystem to a lesser extent compared with the other two subsystems (i.e., cell–cell communication for one instructor and phenotypic expression for the other). Interviews lasted between 45 to 75 min, and all participants were compensated with a $20.00 Amazon gift card for their time.

Instructors participated in the Activating Prior Knowledge and Networks and Relationships phases of the interview protocol. Due to the COVID-19 pandemic, we used slightly altered methods in our interviews with instructors. All instructor interviews were conducted virtually using an online video conferencing platform to host interviews and capture audio and video recording. In addition, instructors used PowerPoint Online as an aid during the interviews (e.g., to view text that would normally be delivered on paper) and to assist their explanations of connections during the Networks and Relationships phase via drawing. The nine students in our participant pool were not asked to draw during the Networks and Relationships phase. However, before implementing the drawing task in our interviews, we conducted a pilot interview with a senior biology student. The pilot student’s explanation and drawing shared many features with this study's sample of nine nondrawing student models. The instructor data were transcribed and analyzed the same as the student data. We coded instructors’ responses in the Activating Prior Knowledge phase for content and response form and categorized all responses as correct, serving the normative basis for the functional definition questions (unpublished data). Using the instructors’ drawings as a foundation, we additionally created knowledge network models based on responses in the Networks and Relationships phase (unpublished data) to aid in identifying normative connections between the three subsystems.

It is important to note that instructor interviews were conducted to establish normative knowledge for our analysis and not for a comparison within an expert-novice framework.

Trustworthiness

We utilize multiple different strategies that follow Guba’s (1981) model of relevant criteria to establish trustworthiness in qualitative research (Krefting, 1991). For truth value, we present dense descriptions of our findings in the Results section and relate our findings to previous work in the Discussion section. For consistency, we provide dense descriptions of our methods and describe our coding process in the following paragraph. For neutrality, members of the research team independently coded the data and arrived at comparable conclusions (i.e., peer examination). Our data analysis is also enriched by the perspectives each researcher brought to the project. Our team consists of one graduate student, two undergraduate students, and one associate professor. All members have varied years of teaching experience and research experience and are of different disciplinary backgrounds. The first author (S.F.) is a biology education graduate student with an undergraduate major in micro­biology, the second author (K.H.H.) is a junior-year student majoring in general biology and creative writing, the third author (G.K.R.) is a junior-year student majoring in health and disease, and the senior author (S.M.G.) is a physiologist by training. The design, data collection, and data analysis were guided and mediated by S.M.G., who has many years of experience in qualitative research methodology. Our approach of coding, recoding, and reflexivity led us to understand the data more deeply and establish higher confidence in our interpretation of the data. As a qualitative research study with a small sample and constrained to one research site, we do not claim our work to be generalizable. Thus, we do not fulfill the criterion for applicability, but share rich descriptions of our findings to allow other readers to decide how applicable our context is to theirs. Our methods of establishing trustworthiness in the data are similar to approaches taken by other research groups (e.g.,Cooper et al., 2018 and Stanton et al., 2015).

For coding of data from the Activating Prior Knowledge phase, one author (S.F.) independently evaluated all the responses and developed a preliminary codebook. The senior author (S.M.G.) coded ∼20% of the data using the codebook. The two authors then met to discuss coding and modify the codebook. Following this initial round of coding, the first author (S.F.) then iteratively revised the codebook with the third author (G.K.R.). S.F. had previously trained G.K.R. on inductive and deductive coding and codebook development on a separate research project before working on this project. Both authors met after each round of coding to discuss and modify the codebook, consulting with S.M.G. when needed to resolve disagreements or provide input. After multiple rounds of discussion and modification, the authors settled on a final codebook. S.F. and G.K.R. independently coded all the data and had over 80% agreement before discussion. All coding was discussed to complete agreement. The process for coding data from the Networks and Relationships phase was the same, except that subsequent coding refinement and analysis was performed by the first author (S.F.) and the second author (K.H.H.). S.F. had also previously trained K.H.H. on inductive and deductive coding and codebook development on another research project before working on this project, and all the final coding was also over 80% agreement before discussion to consensus.

RESULTS

The alignment between the research questions, data sources, and analysis is shown in Figure 1. The results from this research project are presented in the order of the research questions, and each data analysis section is labeled with its relevant data source. As mentioned in the Methods section, the purpose of interviewing the instructors was to identify correct content knowledge of definitions and gain insights about knowledge networks to aid in coding and analysis of student data and not to serve for expert-novice comparisons. Thus, the instructor data will not be formally presented.

Research Question 1

Activating Prior Knowledge – Functional Definitions of the Three Subsystems.

To answer our first research question, we analyzed students’ responses in three ways: 1) the relationship between the type of question and the type of answer, 2) what the participant said, and 3) the normative nature of their answer. For 1), we binned answers as what, how, or why responses (Figure 2). What answers are descriptive explanations that account for the ‘things’ or entities of a system (Hmelo-Silver and Pfeffer, 2004). How answers are mechanistic explanations that account for the interactions between the entities that cause the phenomenon (Mayr, 1961; Machamer et al., 2000; Abrams and Southerland, 2001). Why answers are narrative-based explanations that account for the origins, purposes, or rationale for the phenomenon (Mayr, 1961; Abrams and Southerland, 2001). One code (what, how, or why) was assigned to each content code to capture the nature of the entire answer. For example, while a participant may describe relevant entities (the what), the participant would receive a how code if they used those entities to provide a mechanistic explanation. In our sample, students did not explicitly situate their answers within specific biological phenomena but rather described less defined contexts and processes. As such, we do not refer to their explanations as descriptions of phenomena. Inductive and deductive coding analysis was performed for 2) to categorize responses by knowledge used to answer the question, and 3) correctness of the knowledge was based on deductive codes from our normative data sources (i.e., textbook definitions and instructor interviews).

Responses to How the Subsystems Occur.

Overall, most answers to the questions of how gene regulation, cell–cell communication, and phenotypic expression occur were in the form of a what response (Figure 3). Students chose to describe generally what the subsystem is or what things are involved in each of the subsystems instead of describing a mechanism. The majority of responses were coded as normative for cell–cell communication and phenotypic expression. However, for gene regulation, the responses are almost split equally with nonnormative answers being slightly more common. Across the three subsystems, all how responses were described using normative information.

Gene Regulation.

Students provided what, how, and why answers to how gene regulation occurs and, overall, used nonnormative knowledge (Figure 3). Normative knowledge included discussing concepts of gene regulation such as transcription factors regulating gene activity, the modulation of operons, and the effect of external factors on gene regulation. Less than half of the students (4/9) discussed relevant mechanics of gene regulation (Supplemental Table S1) with only two of those students describing how those mechanics function. A third of the students correctly mentioned that external factors affect gene regulation (Supplemental Table S1), but only Student one provided a how explanation, even if not fully specified:

Most common nonnormative what responses described gene regulation as ensuring that the cell’s genes are kept “healthy” or as governing the process of cell division (e.g., accurately duplicating and dividing the DNA to daughter cells). The following quote from Student two is an example of a nonnormative what answer for gene regulation in which they describe that gene regulation keeps genes “healthy”:

We found two instances of students providing why answers, with both answers being nonnormative to the function of gene regulation. For example, Student two describes later in their answer that gene regulation is involved in cell division and explains why this occurs:

Lastly, one student was unsure and did not provide an answer to the question. In total, less than half of the students (4/9) discussed normative ideas regarding gene regulation and only 2/9 of the students explained how gene regulation occurs using a how response.

Cell–Cell Communication.

In explaining how cell–cell communication occurs, students described answers using what and how responses and overwhelmingly provided normative descriptions of cell–cell communication (Figure 3). Normative concepts for this subsystem included describing ideas such as the sending and receiving of signals, signaling transduction pathways connect to gene regulation changes, and alteration of targets’ properties and providing examples of cell–cell communication types (e.g., paracrine, autocrine, and endocrine), types of signaling molecules (e.g., hormones), and types of receptors involved in communication (e.g., transmembrane receptors). Almost all of the students (8/9) described relevant concepts or processes of cell–cell communication, and five of these students explained their answer using a how response (Supplemental Table S2). Among these how responses, two students explained the process of cell–cell communication using language akin to mechanistic explanations (i.e., including much of the following: naming entities, their activities, and interactions over space and time). Student one exemplifies this code well with their description of:

As demonstrated in the last quote, this student also described how the process of cell–cell communication can affect gene regulation in a cell. Among our student participants, this was the only student to make this connection to gene regulation during this question.

The what category contained descriptive accounts of cell–cell communication concepts (6/9 students) and listed examples of cell–cell communication types, molecules, and/or junctions (4/9 students; Supplemental Table S2). The following quote from Student five is an example of listing types of cell–cell communication without elaborating on what they do or how they work:

The only nonnormative answer was coded in the ‘Other’ category and described how certain parts of cells have different regulatory properties depending on “the area of the body or the area of the gene,” which affect what genes, ions, or molecules can enter or leave the cell to affect communication.

To review, we found that all of the students we interviewed were able to provide at least one normative response by describing relevant concepts or naming examples of cell–cell communication types, molecules, and/or junctions. However, answers were primarily framed as what responses with only 5/9 students describing cell–cell communication with a how explanation.

Phenotypic Expression.

Students responded to the question regarding how phenotype is regulated with what and how answers and mainly described normative ideas (Figure 3). For this question, normative ideas included (but were not limited to) describing that phenotype is informed by the genotype, phenotype is regulated based on environmental factors, and genotypes code for proteins and RNA and their actions give rise to phenotypes. We also accepted definitions of phenotype or its defining attributes as normative knowledge. Only 2/9 students explained how phenotype is regulated using a how response (Supplemental Table S3). One student correctly described how phenotype can change based on interactions with the environment, and the second student provided normative ‘Other’ ideas about how phenotypic regulation occurs. For example, Student two describes how phenotype can be regulated using modern tools:

Most of the students (7/9) commented that phenotype results from the genotype via a what response (Supplemental Table S3). Many students also described that genotypes code for physical characteristics such as height, eye color, and hair color as opposed to coding for proteins or RNA. The next most common what response was definitions or descriptions of phenotype (5/9 students; Supplemental Table S3). Only a third of students discussed the effect of environment on an organism’s phenotype (e.g., phenotypic plasticity) with two of these students mentioning this effect through a what response. While all of these what responses were considered normative, the remaining what answers were nonnormative. Specifically, 5/9 students expressed that phenotype results from gene heredity (typically Mendelian ideas) or cell division mechanics (e.g., mitosis, meiosis, and crossing over events; Supplemental Table S3). The following passage from Student four illustrates the cell division mechanics code:

In summary, while over half of students (7/9) stated that phenotype depends on the genotype and/or environment, we found that less than half of students’ whole answer (4/9) included only normative ideas about phenotypic regulation. Only two students answered the question using how framing, and those how answers were normative.

Responses to Why the Subsystems Occur.

Overall, most answers to the questions of why the three subsystems occur were in the form of a why response, except for phenotypic expression in which what responses dominated (Figure 4). Aligned with the prompt, the majority of students chose to describe why the subsystems of gene regulation and cell–cell communication occur or their purpose/importance to the cell or organism. Most answers were coded as nonnormative for gene regulation, slightly more normative answers for cell–cell communication, and normative for phenotypic expression. This trend is also consistent when examining only why responses for gene regulation and phenotypic expression. However, the majority of why responses for cell–cell communication were nonnormative.

Gene Regulation.

In our data set, we found that students only described why answers to the question why gene regulation occurs (Figure 4). Normative answers to this question included ideas such as not all genes are expressed all the time or are differentially expressed for cell specialization, to conserve energy and/or resources, to respond to the environment, and to tightly regulate cell processes. Less than half of the students (4/9) described ideas represented in the normative knowledge data sources: not all genes are expressed all the time (1/9), to respond to the environment (2/9), to efficiently use energy/ resources (1/9), and from an evolutionary perspective (1/9) (Supplemental Table S4).

Over half of students (5/9) emphasized nonnormative ideas that gene regulation occurs to ensure correct cell division or to keep the organism healthy (Supplemental Table S4). As an example, Student eight describes in this quote that gene regulation helps maintain organismal health:

One student suggested the nonnormative ‘Other’ reason that gene regulation serves as a process to permanently alter the DNA for the cell’s needs (Supplemental Table S4).

To summarize, we found that only 4/9 students described ideas relevant to our normative knowledge data sources concerning why gene regulation occurs, and all answers provided were framed as a why response.

Cell–Cell Communication.

Responses for why cell–cell communication occurs were divided between what and why responses, and answers were slightly more normative overall (Figure 4). Our normative knowledge for this question included ideas such as cell–cell communication allows for cells to monitor/ react to their environment, to achieve specific goals (e.g., coordination of cells to grow toward certain nutrients or signaling insulin secretion during digestion), and for maintaining health such as immunological signaling during infection or providing nutrient or structural support to neighboring cells. Almost all students (8/9) gave vague why statements that cells need to communicate in order for the whole to function (Supplemental Table S5). We coded these statements as nonnormative, because the idea that cells communicate ‘because they must’ is not consistent with our normative knowledge and is teleological in nature. The following passage from the interview with Student seven typifies this code:

The remaining why answers were normative. Only a few students (2/9) mentioned that cell–cell communication allows for cells to react to the environment, many of the students (5/9) discussed cell–cell communication occurring for health-related reasons such as monitoring cell health to prevent cancer, and 1 student gave specific reasons aligned to their described examples (Supplemental Table S5).

A few students (3/9) chose to answer this question with a what response by describing normative features or examples of cell–cell communication with varying levels of specificity to why those features occur (Supplemental Table S5). For instance, one student simply stated that cells can react to the environment, but did not explicitly describe that one purpose of cell–cell communication is to be able to sense and react to the environment.

In total, over half of the students (7/9) provided a response aligned with our normative knowledge regarding why cell–cell communication occurs, and from this subset, six students framed their answer in a why response.

Phenotypic Expression.

Students provided normative what and why responses when asked why it is important that genotype and phenotype are linked (Figure 4). Normative knowledge for this question included answers such as the scientific community’s current understanding of phenotype relies on its definitional link to genotype, the link helps determine what genes correspond to what phenotypes, the link allows for favorable phenotypes to be selected for in the environment, and the link facilitates alteration of an organism’s phenotype in response to its environment. Most students had limited responses regarding why the link between genotype and phenotype is important. Only three students provided a why response, each describing a different, normative reason: phenotype manifests through its link to the genotype, for deducing which genes cause what phenotypes, and for selection of phenotypes to be heritable through the genes (Supplemental Table S6). To illustrate the last reason, Student one explains:

All other student answers were provided in the form of a what response. The most common what answer (4/9 students) was descriptions that without the link between genotype and phenotype, scientists would not be able to understand or explain phenotypes (Supplemental Table S6). Many students seemed uncertain in general and stated that they were not sure how genetics, genotypes, or phenotypes would work otherwise. The following quotes highlight this code:

In a similar vein to the definitional link code, some students (3/9) described the practical limitations of being unable to deduce the function or role of genes without a corresponding phenotype (Supplemental Table S6). Lastly, one student did not know how to answer the question.

Taken all together, while 8/9 students described relevant ideas regarding why it is important that genotype and phenotype are linked, only 3/9 students expressed their answer in the form of a why response.

Research Question 2

Networks and Relationships – Knowledge Networks: Structures.

To further understand participant thinking about the three subsystems and the ways in which they are connected within biological systems, we took the verbal transcript and created models of their knowledge networks by forming arrow relationships to represent the linkages they described. From our analysis, we identified four total types of knowledge networks in regard to the three biological subsystems: 1) Multiple Subsystems Link to Phenotype, 2) Converging End Point, 3) Unidirectional Loop, and 4) Bidirectional Loop (Figure 5). The Multiple Subsystems Link to Phenotype model is defined by two unidirectional arrows, linking one subsystem to a second and the second to the third. The Converging End Point is defined by two subsystems leading to a singular subsystem (the end point) without any feedback arrows to either of the two subsystems. The Unidirectional Loop is defined by three unidirectional arrows that form a cyclical relationship (i.e., one subsystem is linked to a second, the second to the third, and the third back to the first). The Bidirectional Loop is defined as three bidirectional arrows between all three subsystems.

Out of the four models expressed by students, Multiple Subsystems Link to Phenotype was the most frequent knowledge network (Figure 5A). The majority of students built linear knowledge networks (6/9) and expressed unidirectional relationships (6/9). The nine students described a total of 26 relationships and various relationship types with uneven frequency. For example, the relationship that cell–cell communication affects gene regulation was described 8/9 times, but the relationship that phenotype affects gene regulation was only described once.

From our instructor interviews, we intended to determine which relationships are normative and should be represented in all knowledge networks describing how gene regulation, cell–cell communication, and phenotypic expression are related. However, we are unable to make a judgement on which relationships are normative and which are nonnormative (see Limitations and Future Directions). Thus, we cannot report on which students expressed normative or nonnormative knowledge network structures. At this time, we can only analyze whether the content used to justify the relationship was normative. We found that few students (3/9) used incorrect or incomplete ideas when describing linkages between the subsystems. For instance, Student four describes how gene regulation affects phenotype by saying, “…gene regulation like creates a DNA that codes for the specific cells.”

Networks and Relationships – Knowledge Networks: Nature of Connections.

For every relationship that a participant described, we evaluated the language used to explain and justify the connection. We identified four different connection categories through our inductive and deductive coding analysis: 1) Mechanistic, 2) Specified Causal, 3) Unspecified Causal, and 4) Associative connections (see Table 3 for code definitions). Briefly, a Mechanistic connection is a detailed, causal sequence of events driven by entities and activities (Machamer et al., 2000), a Specific Causal connection contains at least one causal factor to contextualize the relationship (Becker et al., 2016), an Unspecified Causal connection is a simple causal relationship without players, and an Associative connection is a relationship based purely on superficial features without any causal reasoning (Southard et al., 2016). Of the total 24 connections described by our nine participants, we found that 9 connections were Specified Causal, 14 connections were Unspecified Causal, and 1 connection was Associative. The majority of students described Unspecified Causal connections in their explanations, and none of them used a Mechanistic explanation. We also disaggregated the data to see whether any relationships existed between specific relationship types and nature of connection category (Supplemental Figure S1). We found that students tended to describe all relationships using Specified Causal and Unspecified Causal connections somewhat equally. The only exception appears to be for relationships between cell–cell communication and phenotypic expression in which Associative or Unspecified Causal connections were described.

Networks and Relationships – Knowledge Networks: Physical Scale-level of Players and Localization of Processes.

To examine the level of organization of players (van Mil et al., 2013), we scanned their explanations for things (i.e., nouns or objects such as cells, genes, and signaling molecules) and then coded its appropriate level of organization. In our data set, we identified the following levels of organization: Undefined, Environmental, Organismal, Cellular, Macromolecular, and Molecular (see Table 3 for code definitions). All six level of organization categories were represented in the student data (Figure 6). The majority of students favored Undefined, Cellular, and Macromolecular levels of organization, and very few students described Environmental, Organismal, and Molecular as levels of organization in their answers.

We coded localization (Russ et al., 2008; van Mil et al., 2013; Lira Gardner, 2020) based on the descriptor text surrounding the verbs or phrases explaining how the process is occurring. For our data, we identified our localization areas as Unspecified, Outside a Cell, At the Cellular Membrane, Inside a Cell, and ‘Other’ (see Table 3 for code definitions). Three out of the four localization categories were represented in the students’ explanations (Figure 7). The majority of students favored Unspecified and Intracellular localizations. Very few students (2/9) described processes happening Outside a Cell, and no student talked about processes occurring At the Cellular Membrane.

DISCUSSION

Students in our sample expressed a variety of ideas when defining how the three subsystems occur and their functional purposes (Activating Prior Knowledge). Overall, these ideas were nonnormative for gene regulation, normative for cell–cell communication, and had varied alignment for phenotypic expression. Additionally, most students did not spontaneously provide a mechanistic explanation when explaining how the subsystems occur. Students often discussed various ideas and components relevant to the subsystem (i.e., what), but did not use these pieces together to compose a mechanism. When explaining how the three subsystems are related (Networks and Relationships), most descriptions were simple and linear, unspecified, and favored particular levels of organization and localizations. Importantly, we did not identify any Mechanistic connections in students’ knowledge networks.

Bridging these data analyses, we explored in our sample how students’ functional definitions of the three subsystems relate to features of their knowledge networks. To facilitate this, participants were given a score based on the number of normative definition codes out of their total codes. We then arranged participants based on their normative score value and analyzed the data for patterns. We found that broad model type (i.e., linear or nonlinear) was seen across the range of student knowledge. Interestingly, students in our sample who expressed strong understanding of the normative knowledge of the definitions also used more normative knowledge when describing connections and expressed more Specified Causal connections. We also explored whether alignment of normative responses to question type (e.g., a how answer with a how question) was associated with features of their knowledge networks. We found that the amount of normative aligned responses did not relate to model type. However, we observed that most students with at least one correct how-aligned response also provided normative Specified Causal connections in their knowledge networks.

Our data indicate the importance of both general normative knowledge of the subsystems and how the knowledge is applied (i.e., alignment of their response type to question type) to well-specified and correct knowledge networks. These results are supported by literature on knowledge integration (Clark and Linn, 2003). Students who demonstrated a nonnormative and nonmechanistic understanding of the subsystems often translated to knowledge networks built with incorrect knowledge and with unspecified connections. In contrast, students who demonstrated more normative and more mechanistic understandings of the subsystems also expressed knowledge networks built with correct knowledge and with specified connections. These knowledge networks are essential for student learning of science, transfer of knowledge to other disciplines, and lifelong understanding. Thusly, instructors need to ensure that students’ conceptions of various biological concepts are normative and that students have opportunities to develop their knowledge networks.

Student Conceptions of Gene Regulation

Many undergraduate students in our study referred to gene regulation ‘regulating’ DNA health or correct cell division instead of describing the modulation of gene expression. Previous work has demonstrated that students have varied conceptions of genes and gene expression. A study by Newman et al. (2021) found that when explaining the concept of gene expression, only half of students described its relevant molecular processes (e.g., transcription and translation) and about 70% of students discussed gene expression being involved with trait outcomes, often referring to Punnett squares. In our data set, similarly, only 4/9 students described relevant molecular processes involved in gene regulation (e.g., activation of transcription and RNA processing). While none of our students linked Punnett squares to gene regulation, one student mentioned that gene inheritance is involved in gene regulation. Interestingly, our work appears to be the first to report students describing gene regulation being involved in keeping DNA “good” or governing correct cell division. We posit that this new finding may be due to students in our sample conflating gene regulation with passing the checkpoints of the cell cycle. Indeed, two instructors in their interviews described an intimate connection between gene regulation and regulation of the cell cycle. Students in our sample seem to be missing the broader definition of gene regulation, which is that it controls the expression of genes.

Our normative data sources described that gene regulation serves to selectively express certain genes at certain times or differentially across a multicellular organism to efficiently use resources and for specialization of cell types. However, we saw that many undergraduates mentioned that gene regulation functions to maintain organismal health or correct cell division. Without knowing the normative definition of gene regulation or how it occurs, it is understandable that many students could not correctly describe why it occurs. The function of genes and gene regulation are critical to understanding information flow, a core concept in biology education (Brownell et al., 2014). Thus, more work is needed to identify the ways in which students think about gene regulation and guide them toward a normative understanding.

Student Conceptions of Cell–Cell Communication

As cell–cell communication is a broad term, students mentioned a variety of topics such as chemical, electrical, neuronal, and contact-dependent signaling. Though encouraging that all students were generally familiar with cell–cell communication and discussed normative concepts pertaining to how it occurs, only one student described intracellular signal transduction pathways. Previous work has noted the challenging nature of teaching signal transduction in the undergraduate classroom (Kramer and Thomas, 2006; MacDonald et al., 2019). It seems that most students broadly know that cells communicate using various methods which have different characteristics, but their understanding of communication ends at the membrane. For example, one student described in great detail how chemicals in one neuron are packaged into vesicles, the vesicles fuse with the outer membrane, the molecules are released into the gap, the molecules travel across the gap to then bind to receptors on the receiving neuron, and then “they continue traveling along.” The student did not specify what the end of the signaling pathway looks like once the message arrives at its final destination. This appears to be a common black box for students in our interviews.

Almost all students provided the teleological reason that cells “have to communicate” in order “to get things done” or “for the whole to function.” Few students mentioned the importance of cell–cell communication as an avenue to recognize and respond to environmental stimuli. The environment plays a key role in cell–cell communication, whether signaling cues come from within the same multicellular organism or the external surroundings. Our work shows, however, that recognizing the environment’s role in cell–cell communication remains a challenge for students. Previous work has found that students utilize “sensing-responding” models less often to explain how organisms interact with the environment to result in phenotypic plasticity (Haskel-Ittah et al., 2020). As cell signaling is critical to living systems’ functioning and to different subdisciplines in biology (Wachira et al., 2019), helping students understand how cell–cell communication occurs and why it occurs should be an important objective in undergraduate biology education.

Student Conceptions of Phenotypic Expression

Most undergraduate students described phenotype being regulated by what genotype is received from cell division or inheritance instead of how expression of the genotype in combination with environmental factors results in molecular or macromolecular products whose activity causes the phenotype. Several of our findings are consistent with previous work on students’ conceptions of phenotypes and proteins. Newman et al. (2021) found that students rarely described genes as coding for molecular products, but frequently described that genes code for traits such as eye or hair color. We similarly found that none of our students described how phenotype results from molecular products made during gene expression but generally said that the genotype codes for characteristics like height, eye color, and hair color. This disconnect between genes, proteins, and traits has been reported in many research studies (Marbach-Ad, 2001; Duncan and Reiser, 2007; Jalmo and Suwandi, 2018; Read and Ward, 2018) and may be related to improper bridging of classical genetics and molecular biology (Reinagel and Bray Speth, 2016). Similarly, in the context of our study, students appeared to focus on ideas of cell division and inheritance—important concepts related to classical genetics—and did not discuss molecular biology nor the role of proteins. However, studies have shown that instruction of gene-to-evolution models can help students understand how variations in the genetic code on the molecular level translate to changes in phenotypic variation within a population (Dauer et al., 2013; Bray Speth et al., 2014).

It is unsurprising that less than half of students (4/9) correctly noted the influence of environment on the phenotypic outcome. Phenotypic plasticity is an important biological concept that demonstrates this interplay between environment and genotype, but university students are unfamiliar with the concept (Batzli et al., 2014) and cannot fully explain phenotypic plasticity (Haskel-Ittah et al., 2020). Additionally, a study on how five common biology textbooks describe the relationship between genotype and phenotype found that while 4/5 books included the interaction of environment in the definitions of phenotype, there were minimal explanations or examples of this concept (Puig and Jiménez-Aleixandre, 2011). It is hugely important that the influence of environment on the expression of genes is stressed in the classroom to avoid views of determinism, which can lead to social discrimination and racism (Donovan, 2014, 2017). Our work reemphasizes the need to help students understand the interactions between gene expression and the environment, an area in genetics education research that has not been deeply explored (Haskel-Ittah et al., 2020).

Features of Students’ Knowledge Networks

In our analysis of students’ knowledge networks using the theory of knowledge integration (Clark and Linn, 2003), we characterized several aspects of students’ explanations incorporating the three subsystems. Most of the students’ knowledge networks were linear, unidirectional, and favored certain relationship types over others. From the theory of knowledge integration, we would expect knowledge networks to be highly interconnected, but the most common student knowledge network in our data was Multiple Processes Link to Phenotype which features only two arrow connections. These data suggest that our students are still developing integrative knowledge networks. Further, while identifying the structures and relationships between them is thought to be a lower-level skill (Momsen et al., 2022), our students faced challenges in the context of our study.

The finding of few integrative knowledge networks synergizes with our nature of connections data in which we found a majority of Unspecified Causal connections and no Mechanistic connections expressed spontaneously by students. Scholars have suggested that integrated knowledge networks need to be built with mechanistic ideas and causal drivers (Southard et al., 2016; Haskel-Ittah and Yarden, 2018). We might expect that students who can articulate mechanistic ideas when defining how processes occur and relate to each other would have a more complex and integrated understanding of biological concepts. While we did not find a relationship between model type and amount of normative aligned responses (e.g., more mechanistic definitions), our data suggest that normative aligned responses may relate to providing more specified connections within knowledge networks. Additionally, the skill of characterizing qualitative relationships between structures and processes is more difficult than identifying and describing a system (Hmelo-Silver, 2007; Momsen et al., 2022). However, there are a variety of possible reasons why we may not have identified more Mechanistic connections (see Limitations and Future Directions).

Interestingly, our results from analyzing the level of organization of the players further support our nature of connection data. Unpacking factors at lower scalar levels and connecting their interactions to the higher level of organization is essential in explaining scientific phenomena (Krist et al., 2019) and is a tenet of mechanistic reasoning (Russ et al., 2008; van Mil et al., 2013). Thus, describing players at many different levels of organization is required to adequately explain the appearance of an organism’s phenotype within a biological system. Our data showed that the Molecular level of organization was often missing in students’ knowledge networks. Gene regulation is a Macromolecular and Molecular process; cell–cell communication encompasses players at many levels of organization including Cellular, Macromolecular, and Molecular; and phenotypic expression may involve Molecular players. In the context of our three subsystems, not describing Molecular players can hinder production of a fully mechanistic explanation. Accordingly, we did not find Mechanistic connections in our student knowledge networks based on students’ spontaneous descriptions. Our data emphasize that in order to generate more Mechanistic connections, students may require support identifying players and activities at different biological levels of organization and how those interactions impact other players and processes to result in the mechanism.

CONCLUSIONS

Limitations and Future Directions

One of our primary findings was the small number of mechanistic explanations in the overall data set: our inductive coding revealed few mechanistic ideas, not all students answered our how questions with a how response, and this was reflected in a lack of Mechanistic connections in knowledge networks. There are some possible reasons why this occurred. First, participants were not overly pressed or provided with scaffolds for constructing mechanistic explanations as we were interested in their spontaneous, unstructured responses. Previous work has shown that undergraduate students can produce explanations using mechanistic reasoning, and introductory and upper division students are capable of describing connections in concept maps with functional, causal, mechanistic, or action-based ideas (Southard et al. 2016). However, in their study, Southard and colleagues (2016) specifically ask students to think about components of mechanisms in their interview question: what is happening (activities), what players are involved (entities), and where it is happening (spatial arrangement). Therefore, it is possible that our participants had the capacity to provide mechanisms but did not realize that mechanistic explanations were being asked of them or may have required additional prompting or structure. We found that participants generally provided what answers that were descriptive of ideas related to the subsystem or identified players. While important for students to know the ‘what’ or lay out ideas first, most students ultimately did not describe the ‘how.’ Previous work has shown that students often answer how questions with why explanations (Abrams and Southerland, 2001). Further, we contend that it is important that biology students in our study do not describe mechanisms as a way of describing biological systems and points to recommendations for instruction and assessment. In addition, it may be that building a mechanistic explanation is challenging (Abrams and Southerland, 2001; Southard et al., 2017; Scott et al., 2018). It may also be that describing a mechanism without a context is awkward or challenging. Indeed, one instructor expressed difficulties in describing their knowledge network acontextually and the desire to embed tangible examples into the knowledge network. Nevertheless, students need to develop a broad conceptual understanding of important biological processes and relationships that are portable and that they can instantiate with examples. Given that our data suggests students seem to need to ground their ideas in specific examples, future iterations of this research will ask participants to contextualize their knowledge network models with an example of their choosing.

This recommendation was additionally guided by findings from our instructor interviews in which we sought to deduce the normative knowledge network structure to analyze the correctness of the students’ knowledge networks. Knowledge network structures for instructors were invoked at different frequencies, and instructors emphasized the importance of context in framing the knowledge networks (e.g., how subsystems are operationalized, the biological context of the task, and background of the individual; unpublished data). Situated cognition (Brown et al., 1989) explains the diverse knowledge network structures captured in our data. Learning and cognition are inherently situated in the contexts, activities, and cultures in which it was learned (Brown et al., 1989). Therefore, the ways in which an individual understands, defines, and uses concepts will also vary. Previous work has also demonstrated how the contextual features of a task impact type of reasoning and explanations of phenomena (Duncan, 2007; Nehm and Ha, 2011; Shea et al., 2014; Haskel-Ittah et al., 2020). As all our participants were operating from differing perspectives and under different contexts, it is arbitrary to derive one normative knowledge network and thus, we were unable to analyze the correctness of our students’ knowledge networks (see Flowers and Gardner, 2022 for a detailed description of the instructor data). Guided by these findings, our future research will additionally explore how the participants’ chosen context affects the construction of mechanistic explanations and features of their knowledge networks.

Providing our participants with textbook-derived definitions of the subsystems during the Activating Prior Knowledge phase may have implications for a student’s ability to perform on the cognitive tasks in our interview. Reading the definitions may have either reinforced existing conceptions (affirming understanding) or introduced new ideas (illuminating gaps or inconsistencies) and could have significant implications for a student’s ability to perform on the cognitive tasks. It is possible that students who provided normative responses were lent confidence from the definitions and students who provided nonnormative responses or no answer may have felt uncertain or required more time and energy to integrate these ideas into their knowledge. Thus, students who already had normative ideas may have been further advantaged in their response to the relationship question merely by confirming their ideas.

Lastly, we describe other modifications and desires for our future interviews to attend to other potential limitations. We plan to modify the wording of questions regarding the phenotypic expression subsystem to address minor confusion expressed by participants. We would like to interview more students and instructors to see whether there are other diverse functional definitions and knowledge networks not captured in our data set. Additionally, a larger sample size may be necessary to more deeply explore how knowledge of the three subsystems affects the features of their knowledge networks. We would also like to interview students and instructors from varying subdisciplines to more deeply investigate how disciplinary background and knowledge affects their answers. From these new insights, we will characterize knowledge networks using the theory of knowledge integration, mechanistic reasoning, and situated cognition as analytical lenses.

Instructional Implications

From our data, we recommend the following instructional practices. First, provide explicit instruction and review of how gene regulation, cell–cell communication, and phenotypic expression occur and their functional purposes, including how they are related in terms of mechanistic explanations. As participants did not always describe where the processes were occurring, discussions should also include where the subsystems are localized and how they traverse varying parts of the cell and the system (e.g., cell membrane, cytoplasm, and extracellular space). Each biology course and even instructors may hold varying definitions of these subsystems. For example, while the mechanics of gene regulation are the same regardless of subject, where gene regulation starts and ends may differ across classrooms. As foundational subsystems, students learn new concepts based on or around these subsystems. To aid in student learning, we recommend reviewing what these subsystems mean in the context of the classroom and not assuming that all students have normative ideas or the same contextualized ideas as the instructor.

Additionally, our data suggest a need for instructors to consider systems thinking and to build up students’ competence across the hierarchy of system thinking skills to promote students’ abilities to explain and reason about systems. Thus, we suggest incorporating more opportunities for students to practice describing global connections across biological concepts within larger networks and systems. Educational reform calls in undergraduate biology advocate for cultivating a broad, conceptual understanding of biology and viewing biology as an interconnected system (AAAS, 2011). Promoting mechanistic reasoning and systems thinking skills (Momsen et al., 2022), which underscore that how and why concepts are connected and interconnected within different systems, will encourage students to engage in the process of knowledge integration. In addition, discussing biological concepts outside of contexts they are typically taught in, an area that has not been greatly emphasized in biology teaching (Haskel-Ittah et al., 2020), may help students bridge concepts across different disciplinary contexts and build an interconnected understanding of biology. For example, instruction can highlight how sensing and responding to environmental cues is important for body systems and cells not only in the context of homeostasis, but also in genetics (Haskel-Ittah et al., 2020). This is not only important to develop critical reasoning skills in the next generation of scientists but also to encourage lifelong scientific knowledge and literacy for the larger public because building incorrect connections can lead to wrong knowledge about concepts and, in the long term, loss of connectivity between ideas (Clark and Linn, 2003). These opportunities will also help students practice thinking across levels of organization and consider how multi-scalar interactions generate biological phenomena.

ACKNOWLEDGMENTS

REFERENCES