I co-taught the upper division undergraduate and graduate class “Introduction to Systems Biology” with Prof. Andrew Capaldi. The aim of the course was introducing students to how Systems Biology concepts are used in research by assigning one or two papers as pre-class reading and discussion of the papers in class.
The general scheme of classes consisted of an introduction to the topic of the class and posing of about 5-10 questions on the paper, which the students then discussed in groups of about five. Answers would then be discussed with the complete class, where an increasing student participation was required over the course of the term. I held three lectures on “-omics techniques in Systems Biology” close to the end of the course, which formed a self-contained unit.
As from my own student days I was unfamiliar with this style of student-participatory teaching, I joined four classes earlier to my section as an observer and teaching assistant. Arguably the most interesting observation I made was that, although discussions were often complex and covered a wide ground, Prof. Capaldi always succeeded in summarizing classes into a clear concept as a take-home message.
Previous parts of the course had explored how (sub-)architectures of biological networks (e.g. feedback and feed-forward loops) can cause information flow that results in remarkable signaling outputs (such as bi-stability, etc.) and engineering principles (e.g. robustness in self-adaptation). The major theme for my part of the course was how experimenters could infer the wiring of biological networks in the first place.
In the following, I will give a brief overview of my first two lectures, before discussing the third in more detail. This lecture is more fully explored as if covers a concept important for students throughout their research careers, which demanded most effort and planning in finding a way to present in a manner most useful for students.
My first class was on transcriptomics, based on classical papers by deRisi et al., 1997 1 on microarray technology and Eisen et. al, 1998 2, first introducing heatmaps. One major objective of the class consisted in addressing technical challenges associated with using microarrays, and -omics methods in general. Using clicker questions and guided student discussions, we for example explored how two-color microarrays were used to overcome limited reproducibility in sample-probe hybridization. The main focus of the class and for the remainder of the course was the question: How can we make sense of the vast amount of information from -omics experiments?
To allow students to see the big picture of this section of the course and give them a compacted take-home message, I here introduced the concept maps in fig. 1 and 2. Fig. 1 depicts which entities an experimenter can perturb and which effectors he can measure in high throughput to build biological networks. In the context of this lecture, this was represented by perturbation of external conditions (e.g. carbon source available to a yeast culture) or the genotype (via gene deletion) and measurement of RNA levels as an effector.
Fig. 1: Concept map: Points of perturbation and of measurement in -omics experiments. Components and arrows were added successively to the map over the progression of the course.
Concept map 2 (fig. 2) was similarly built up over subsequent classes and directly deals with the question of making sense of high-throughput data. In the first class, we concluded that deRisi et al. were limited to mapping data to already known biological networks, while the advent of clustering techniques not only permitted the identification of conditions affected by a treatment in a systematic manner, but also the inference of functions for previously unstudied transcripts via co-clustering.
Fig. 2: Concept map 2: Ways to infer biological networks from -omics data.
In the second lecture, we discussed phosphoproteomics technologies, using papers by Ptacek, et al. 3 and Olsen, et al. 4 to contrast a protein array vs. mass spectrometry based approach. We explored the limitations of a pure in vitro approach, but how protein arrays are nonetheless useful tools in different contexts. We also covered data acquisition by mass spectrometry in some depth and what are the associated limitations of the resulting data.
On our path towards network inference we highlighted the value of extracting direct kinase-substrate relationships from protein arrays. The phosphoproteomics paper in contrast, provided time course data, allowing temporal clustering and inference of connections from the order in which phosphorylation changes are observed (fig. 2).
I will finally discuss the third of my lectures in more depth. The topic of the class was “Functional Genomics” and during the class preparation I realized that this class could not be taught without a firm understanding of epistasis, one of the most important concepts students need to master when embarking on biology research and reading scientific literature. As the composition of the class was heterogenous, it could be assumed that that some of the students would have some, while others would not have had previous exposure to the concept.
To give adequate focus to discussing epistasis, I removed one of the papers discussed in the class in previous years and retained Jonikas, 2009 as the only paper to discuss in the class 5. Given its importance, I also made this class the subject of my lesson plan for the ‘Introduction to Evidence-Based STEM Teaching’ course.
A manuscript in the teaching literature by Knight, et al. 2013 was immensely helpful in preparing the class, as it clearly pointed out the differences in epistasis in metabolic vs. signaling pathways 6. As I could not dedicate the full class to discussing the concept, I only partially followed their lesson plan and adapted some of their clicker questions.
An outline of the class is shown in fig. 3. At the heart of the class paper is the so-called “Double-mutant plot”, a scatter plot where each point represents an experimental readout obtained from a single-mutant yeast strain (x-axis) vs. from the same mutation in combination with mutation of a second gene, “YFG” (y-axis) (fig. 5). Different flavors of the plot occurred multiple times in the class paper and a solid understanding of epistasis is required for its interpretation. The plot forms the gateway to advanced analysis in the paper and in functional genomics generally. It therefore constituted the focal point of the lecture.
Fig. 3: Lesson plan for class 3.
I started the class with a short Youtube video providing a motivation for studying the biological question addressed in the paper (Infolded Protein Response).
We next continued straight to some warm-up clicker questions (using the free version of Socrative), adapted from and inspired by Knight, 2013, asking students to predict the outcome of some epistasis scenarios. Students then were given the opportunity to discuss the questions in small groups afterwards and I provided further explanation on points that remained unclear, before asking students to re-take the same questions. Examples of representative questions are shown in fig. 4. I summarized the main take-away from the clicker questions, an oversimplified version of which reads: “In signaling pathways mutation of the farthest downstream gene dominates, while in metabolic pathways mutation of the farthest upstream gene dominates.”
Fig. 4: Examples of representative clicker questions.
We proceeded by introducing the basic experimental approach taken in the class paper and how this functional genomics approach differs from transcriptomics covered in class 1 (“Transcriptomics informs on which genes are affected by a biological response, while functional genomics informs which genes contribute to the response.”). Both were based on questions posed to the whole class and students` answers, which I further expanded on. We then worked out the limitations of the basic experimental approach and how epistatic analysis would be helpful in gaining further insights.
Proceeding to the core of the lecture, I built up a double mutant plot on the board in a stepwise manner by means of an example. To help the reader follow along the discussion, the main points of the plot are:
The readout from a strain where mutation of gene xxx is without effect will be at 0 on the x-axis. If the gene represses the readout, i.e. its deletion increases the readout, the strain is represented by a point to the right of 0 on the x-axis.
If we were to simply plot the readout of ∆xxx strains on both the x- and the y-axis, all points would by definition be on the main diagonal. If plotting a single mutant ∆xxx vs. double mutant ∆yfg ∆xxx, a point on the main diagonal indicates that YFG deletion is without effect.
A additive contribution of YFG to the biological function will shift a point from the main diagonal in the y-direction. (The shift is downwards if YFG activates the biological function and upwards if it represses the function.)
As the contribution of YFG is additive for most strains, the distance of the diagonal formed by the points from the main diagonal corresponds to the effect of YFG-deletion. Points that appear at this value on the y-axis with a smaller x-axis value correspond to strains in which the effect of XXX-deletion is masked by the effect of YFG-deletion. As the example is taken from a signaling pathway, YFG is downstream of XXX in these cases.
Points on the main diagonal, where the effect of XXX-deletion masks effects of YFG-deletion represent strains where XXX is downstream of YFG.
Points that exhibit a shift on in-the y-direction that surpass the additive effect of YFG-deletion (blue) indicate a cooperative effect between XXX- and YFG-deletion.
Fig. 5. Double-mutant plot from Jonikas, 2009.
To make students think in depth about the concept of epistasis and demonstrate their learning, I asked them to gather into groups, drew three signaling scenarios (e.g. fig. 6) on the board and asked them where a strain with mutation of X in the scenario would be located on the plot. I then asked groups to send one representative each to draw the location of the points on the plot on the board. We repeated the exercise in two rounds of two additional scenarios each. After each round, we discussed the answers with the whole class.
Fig. 6: Example scenario for students to determine location of mutant in X on the double mutant plot. UPR (Unfolded Protein Response) is the measured output.
Overall, it was difficult to get students to go to the board and to commit where to draw the points. Interestingly, one student excelled, whom I had not particularly noticed for his participation in previous classes (I suspect, because he did not feel sufficient challenged). En large, however, it did not appear that students had arrived at a complete understanding of the admittedly difficult concepts. I believe this could only be rectified by the students immersing themselves more deeply into the matter as preparation for the class. A more complete reflection of the outcome of the class is part of “Section 4: Reflections”.
I then returned to Powerpoint slides to introduce some additional nomenclature frequently found in research papers on the topic and to illustrate how the authors of today`s research paper synthesized the functional relationship of gene A with all other genes and gene B with all other genes to tell about similarities in function of A and B. Although this was a major point of the paper, I reduced the emphasis on it compared to previous classes in favor of the above conceptual discussion.
I finally briefly discussed how the same concepts are utilized in modern functional genomics studies, which employ different ways of perturbing gene function and measuring outputs of the perturbation. I concluded the class by pointing out which additional arrow was introduced by today`s content in concept map 1 and giving a brief preview which arrows would be added by the final class given by Prof. Capaldi, however without lifting the suspense in how this would be accomplished.
1. DeRisi, J. L., Iyer, V. R. & Brown, P. O. Exploring the metabolic and genetic control of gene expression on a genomic scale. Science 278, 680–686 (1997).
2. Eisen, M. B., Spellman, P. T., Brown, P. O. & Botstein, D. Cluster analysis and display of genome-wide expression patterns. Proc. Natl. Acad. Sci. 95, 14863–14868 (1998).
3. Ptacek, J. et al. Global analysis of protein phosphorylation in yeast. Nature 438, 679–684 (2005).
4. Olsen, J. V. et al. Global, In Vivo, and Site-Specific Phosphorylation Dynamics in Signaling Networks. Cell 127, 635–648 (2006).
5. Jonikas, M. C. et al. Comprehensive characterization of genes required for protein folding in the endoplasmic reticulum. Science 323, 1693–1697 (2009).
6. Knight, J. K., Wood, W. B. & Smith, M. K. What’s Downstream? A Set of Classroom Exercises to Help Students Understand Recessive Epistasis †. J. Microbiol. Biol. Educ. JMBE 14, 197–205 (2013).