This is the answered version of the Mechanistic Modelling practical. Every ❓ question is followed by a worked model answer (green box). All the interactive steps still work, so you can keep simulating droughts and knockouts while you read, and the self-test quiz at the end is unchanged.
In the earlier practicals you used data-driven methods: clustering and genomic selection look at a pile of measurements and let patterns emerge, without any built-in idea of why genes behave as they do. A mechanistic model works the other way around. You start from biological knowledge that already exists, such as "gene A switches gene B on" or "hormone X represses gene C", and you write those cause-and-effect rules down as equations. The computer then simulates what happens over time when you turn a knob.
Think of it as a flight simulator for a cell. Nobody is guessing from data whether the plane will stall; the simulator encodes the physics and lets you try things. That is exactly why this approach is useful:
Run virtual experiments in seconds. Apply a drought, knock out a gene, lengthen a stress, and see the whole network respond, without spending months in a greenhouse.
Test your understanding. If the rules you believe in fail to reproduce what real plants do, then your picture of the network is incomplete. The model becomes a hypothesis you can falsify.
Generate predictions. A model can point to which gene to knock out, or which experiment to run next, before you touch a single plant.
Keep in mind that a model is only as good as the rules and numbers you put in. It is a deliberate simplification of a far messier reality: useful for reasoning and prediction, but never a substitute for real experiments.
1 The network we will play with
We will use a mechanistic model inspired by
this paper
(also discussed in the Plant Science lecture). It models how tobacco responds to drought stress, and how that response changes the plant's ability to defend itself against a biotic attacker (white flies). Building models like this normally takes years of experiments and prior knowledge; here that work is already baked in, so you can just explore.
The diagram below is the gene regulatory network. Drought is an external input that feeds into the hormone ABA. From there the signal travels through the network. Green solid arrows mean one gene activates another; red dashed arrows mean one gene represses another. Drag the nodes around to untangle it.
activates represses
Hover a gene for its name. The chain ICS1 → SA → BGL2 is the plant's salicylic-acid defence pathway.
❓ Questions
Have a look at the paper. Which gene(s) are important for drought resistance, and which for white-fly resistance?
Follow the arrows: how do you think the network will respond when there is no drought stress at all?
✅ Answer
Drought resistance runs through the ABA branch: drought raises the hormone ABA, which drives ABF2 → ANAC019 and GENEC. White-fly (biotic) resistance is the salicylic-acid pathway ICS1 → SA → BGL2; BGL2 is the defence marker we read out. The clever bit the paper highlights is the link between them: the drought (ABA) branch represses the defence (ICS1) branch.
No drought: with drought = 0, ABA stays low, so ABF2, ANAC019 and GENEC stay low. Nothing is repressing ICS1, so ICS1 is expressed → SA is produced → BGL2 is high. The resting plant is fully defended.
2 A plant at rest: no drought
Let's simulate the network when nothing is stressing the plant. Each line is the expression level of one gene over time. Watch especially the defence readout BGL2 (drawn thick).
Gene expression over time with drought = 0 the whole way through.
❓ Questions
In this case, does the plant have white-fly resistance? Why does it, or why doesn't it?
What do you think will happen when we give the plant a brief pulse of drought stress?
✅ Answer
Yes, it is resistant. Because there is no drought, ABA (and therefore ANAC019 and GENEC) stay near zero, so nothing represses ICS1. ICS1 climbs to a high steady level, driving SA and then BGL2 up. High BGL2 = active salicylic-acid defence, so the plant can fend off white flies.
Prediction for a brief pulse: the ABA branch will spike briefly, nudging the repressors up for a moment, so defence may dip a little, but because the pulse is short we would expect it to recover quickly. (Confirm this in the next step.)
3 A brief drought pulse
Now we switch drought on for a single short moment (from time 10 to 11) and then off again. The grey band marks when it is on. Run it and compare with the resting plant above.
The grey band is the drought pulse. Notice which genes react, and which barely move.
❓ Questions
Can you explain this behaviour, using what you know about gene regulation and the network diagram?
Why would it make sense for a plant to have such a mechanism?
What do you expect will happen to white-fly resistance if we make the drought pulse longer? Try it in the next step.
✅ Answer
What happens: the pulse briefly raises ABA, which activates ABF2 → ANAC019 and GENEC, so you see those upstream genes spike. They momentarily repress ICS1, so ICS1/SA/BGL2 dip a little. But the pulse is over almost immediately, ABA decays, the repression lifts, and the defence pathway climbs back to its resting level. Net effect: a small, temporary dip and full recovery.
Why it makes sense: brief dry spells are normal and not life-threatening. It would be wasteful (and risky) to throw away biotic defence every time it gets dry for an hour. Requiring a sustained signal before switching off defence lets the plant ignore short, noisy fluctuations and only reallocate resources when drought is genuinely serious, which is a filter against over-reacting.
Prediction: a longer drought should suppress ICS1/SA/BGL2 more strongly and for longer, so white-fly resistance should drop noticeably. (Test it next.)
4 Make the drought last longer
Here you control the drought yourself. Slide the start and end of the drought pulse and the simulation re-runs live. Try a long drought and keep your eye on BGL2.
The grey band is the drought. Widen it and watch the defence pathway.
❓ Questions
Can you explain why the response is different when the drought lasts longer? Why would this make biological sense?
Does this model reflect the true underlying biology? In what ways does it, and in what ways doesn't it?
Which gene(s) do you think you could knock out to change the response? Test your idea in the next step.
✅ Answer
Why it differs: a sustained drought holds ABA, and therefore ANAC019 and GENEC, high for a long time. Their combined repression keeps ICS1 switched off long enough for ICS1, SA and BGL2 to actually decay away. So defence collapses for the duration of the drought, and only recovers once the drought ends. The key variable is not whether the repressors turn on, but for how long.
Why that makes biological sense: the plant has a limited budget of energy, carbon and nitrogen, and both responses are expensive. Under a real, prolonged drought the immediate threat to survival is water loss, so it makes sense to spend everything on closing stomata, adjusting osmotic balance and protecting tissue, even though the price is a temporarily lowered guard against attackers. The two branches are not independent processes that happen to interfere; the repression of ICS1 by the ABA branch looks like a deliberate switch that prioritises one stress over the other. Combine this with the previous step and you get the full logic: brief dry spells are ignored, and only a drought that persists is worth paying for. The uncomfortable consequence for agriculture is that a drought-stressed crop is a more vulnerable crop, which is exactly the trade-off breeders would like to break.
Does it reflect biology? Qualitatively yes: it captures a real, documented trade-off (drought signalling suppressing salicylic-acid defence) and the sensible logic that only prolonged stress flips the switch. But it is a heavy simplification: only a handful of genes, deterministic with no noise, fixed Hill parameters and arbitrary interaction strengths, no feedback loops, no tissue or spatial detail, and no fitness cost attached. Treat it as a caricature that is useful for reasoning, not as ground truth.
Which genes to knock out: to protect defence, target the repressors of ICS1, namely ANAC019 and/or GENEC, or cut the signal at its source by knocking out ABA. (Test in the next step.)
5 Knock out genes and redesign the response
A knockout silences a gene: its expression decays to zero and it can no longer pass its signal on. Tick a gene to knock it out and the simulation re-runs. A drought pulse (10 → 30) is applied so you can see the effect on defence.
Knocked-out genes are greyed out in the legend. Drought is on from 10 to 30.
❓ Questions
Which combination(s) of knocked-out genes give the plant the "best" phenotype according to this model, keeping its defence even under a long drought?
Would doing this be realistic biologically? What could go wrong that this model doesn't capture? Before you answer, look up what else the hormone ABA does in a plant.
What experiment would you propose next to test the model's prediction, and how would you explain your finding to a plant biologist?
✅ Answer
Best knockouts: silence the two repressors of ICS1, that is ANAC019 and GENEC together. Then even a long drought cannot switch ICS1 off, so SA and BGL2 stay high and the plant keeps its white-fly defence throughout. Knocking out ABA also works in the model, by removing the drought signal before it ever reaches the repressors, but hold that thought until the next question. Note the trap: knocking out ICS1, SA or BGL2 themselves destroys defence, the opposite of what you want.
Realistic? Not straightforwardly. Defence is metabolically costly, so a plant that can never dial it down may waste resources it needs to survive drought, hurting growth or yield, a cost this model does not include.
The ABA knockout in particular is a bad idea, and it is worth being specific about why, because in the model it looks like the cleanest solution of all. In the diagram ABA is drawn as a single node with two outgoing arrows, as though its only job were to feed this pathway. In a real plant ABA is one of the central hormones, and it is pleiotropic, meaning one molecule with many different jobs. It is what closes the stomata when water runs short, which is the plant's single most important drought response, so a plant without ABA keeps its pores wide open and dries out faster than a normal plant. It also controls seed dormancy, so seeds germinate at the wrong moment, and it feeds into cold and salt tolerance, root growth and leaf senescence. So a knockout that the model scores as a success (defence stays high) would in reality give you a plant that keeps its white-fly resistance and then dies of the drought. The general lesson: the model contains only the arrows somebody drew, and a node's real biological connections are almost always broader than the ones in the diagram. The more central and upstream a component is, the more true that is, which is why intervening on a specific downstream regulator is usually safer than intervening on a hormone.
Other things the model leaves out. Transcription factors such as ANAC019 are pleiotropic too, just less spectacularly so than ABA, so they carry the same risk on a smaller scale. Beyond that: off-target edits, redundancy between genes (another gene may quietly take over the repression), differences between tissues and developmental stages, and the fact that a greenhouse is not a field.
Next experiment and communication: create the knockout line (for example by CRISPR, or use an existing mutant), then measure both traits under a real drought: white-fly resistance and SA/BGL2 levels and drought survival, growth and yield, in controlled and ideally field conditions. Present the result to a biologist as a model-generated hypothesis with explicit caveats: "the model predicts X; here is the trade-off it ignores; here is the experiment that would confirm or refute it."
If you have time left, open this link and click Genome Editor to build your own gene regulatory network and try to grow the prettiest possible plant. While you do, notice how such "simple" networks can produce strikingly complex, emergent behaviour.