My Annotated HHMI Gilliam Research Plan

By Ya'el Courtney, PhD | Stanford Postdoc, Harvard PhD in Neuroscience
Awarded the HHMI Gilliam Fellowship, 2021

This is an annotated walkthrough of the Research Plan I submitted with my successful 2021 HHMI Gilliam Fellowship application. My full plan was five pages of dense neuroscience focused on choroid plexus secretion and cerebrospinal fluid signaling. In this annotated version, I'm going to share the actual text of several key sections (Background and Significance, the Overarching Hypothesis, Specific Aims, and all four parts of Aim 1A), walk through what I was trying to do in each, and then give you structural commentary on the remaining aims.

My goal is to show you the scaffolding and the substance of a strong research plan, not to have you copy any of my actual aims or language. The science here is specific to my field. The strategic choices are transferable.

Two important notes before you read:

1. The current Gilliam research plan limit is three pages. Mine was five. Everything about the structure below still applies, but you'll need to be significantly more concise than I was.

2. The science in my proposal is about a fairly niche topic (how the choroid plexus releases molecules into cerebrospinal fluid during brain development). You don't need to understand the neuroscience to see what I was doing strategically. Pay attention to the structural choices and the logic, not the specific biology.

The Overall Structure

A strong research plan for a fellowship like the Gilliam (or the NSF GRFP, or an NIH F31) follows a predictable structure:

1. Background and Significance (roughly 15-25% of the plan)

2. Overarching Hypothesis (1-2 sentences, clearly flagged)

3. Specific Aims (a short, numbered list)

4. Approach (the bulk of the plan, organized aim by aim)

5. Perspective (short closing)

6. Timeline (1-2 sentences)

Reviewers expect this structure. Deviating from it without good reason will make your plan harder to evaluate.

Background and Significance

Here's the actual opening of my plan:

Neurodevelopmental disorders can arise from a pathogenic interplay between environmental conditions and genetic programming during early brain growth. Cerebral cortical neurons are generated during embryonic development from neural progenitor cells that divide along the brain's ventricles and in contact with cerebrospinal fluid (CSF). CSF is produced largely by the choroid plexus (ChP), a tissue in each ventricle made of an epithelial cell layer surrounding a core of blood vessels. ChP epithelial cells synthesize and secrete CSF and its contents, including many important regulatory signals. My PhD adviser's early work showed that the ChP secretes growth factors required for proper cortical development, like IGF-2. The ChP also allows peripheral fetal signals reflecting the maternal environment to gain access to the CSF, with the potential for both normal regulation and dysregulation of development. Indeed, CSF aberrations are increasingly implicated in neurodevelopmental disorders for which disrupted maternal-fetal communication is believed to play a role, including autism spectrum disorder, hydrocephalus, and schizophrenia.

This first paragraph establishes the field in roughly 150 words. I defined every technical term the first time it appears (cerebrospinal fluid, choroid plexus, ChP) and oriented the reader spatially (where CSF is, what the ChP is made of). By the end of this paragraph, a non-specialist should understand what system I'm studying and why it matters for human disease.

The first sentence is deliberately framed in terms of disease relevance. "Neurodevelopmental disorders" is more compelling to a reviewer than "brain development" as an opener. This tells the committee up front why anyone should care about this project.

The middle section briefly summarizes what's been done in this area (including work from my own lab, naming my PI's contributions). Naming specific diseases at the end (autism, hydrocephalus, schizophrenia) grounds the science in human stakes.

Here's my second paragraph, which sets up my specific project:

Despite its lifelong role in titrating CSF contents, ChP secretion machinery is poorly understood, largely because this tissue is difficult to observe in vivo or maintain in ex vivo preparations. Our lab has pioneered approaches that allow both. Most interestingly, we recently discovered that the ChP displays a high-capacity release mechanism called apocrine secretion. We found that we could elicit Ca2+-dependent apocrine release by activating highly expressed G-protein coupled serotonin receptor 5HT2C. We identified apocrine structures by surface electron microscopy, validated immunohistochemical markers for tracking this process, and visualized it in real time in awake adult mice.

This paragraph does several things. It states the gap ("ChP secretion machinery is poorly understood") and explains why the gap exists (tissue is hard to study). Then it pivots to why my lab is well-positioned ("Our lab has pioneered approaches that allow both") and introduces the specific phenomenon my project will investigate (apocrine secretion). I gestured at the preliminary data (EM, immunohistochemistry, in vivo imaging) without drowning the reader in detail.

The gap statement ("poorly understood, largely because") is the most important sentence in this whole section. It tells the reader why the project needs to happen. Without a clear gap, your proposal is just "I'm going to study a thing." With a clear gap, it's "I'm going to answer a question nobody has been able to answer before."

I also included a figure (Figure 1) with five panels in this section, showing the apocrine secretion phenomenon: a schematic, electron microscopy images, immunohistochemistry, and in vivo calcium imaging. One strong figure can be worth hundreds of words in a research plan, but figures count toward your page limit. With the current three-page cap, you may only have room for one figure total.

The Overarching Hypothesis

My overarching hypothesis is that embryonic ChP reacts to cues in the CSF that herald particular developmental stages, for example the presence of serotonin (5-HT), and uses a previously uncharacterized apocrine mechanism to secrete factors that coordinate brain-wide maturation. Disruption of the magnitude or timing of this function could alter the trajectory of brain development with life-long consequences.

This sentence is doing a lot of work in a small space. It names the tissue (embryonic ChP), the process (reacts to cues), the specific cue I'm testing (serotonin), the mechanism (apocrine secretion), and the proposed function (coordinate brain-wide maturation). Every noun and verb is earning its place.

The second sentence names the stakes. If this mechanism is disrupted, brain development could be altered. That's why this research matters.

Your overarching hypothesis should be testable, specific, and consequential. If your hypothesis could be true of many systems, or if it's vague about what you'd actually expect to see, tighten it.

Specific Aims

Aim 1. Understand 5HT2C mediated apocrine secretion in ChP in terms of 1A) its time course with respect to cortical development and 1B) its Ca2+ dependence and secretory machinery.

Aim 2. Determine the contribution of apocrine secretion to CSF content. 2A) Determine the effects of blocking apocrine secretion on CSF content. 2B) Assess the effects of maternal inflammation on embryonic CSF 5-HT levels and ChP secretion in a naturalistic gain-of-function paradigm.

A few structural choices worth flagging:

• Two aims with sub-aims, not three full aims. This is standard for a fellowship-length proposal. Three independent aims usually won't fit in three pages, and it spreads your work too thin. Two aims with 2-3 sub-aims each gives you depth without sprawl.

• Aims are written as verbs. "Understand," "Determine," "Assess." Not "Studying the role of..." Aims are things you'll do, not things you'll study.

• Aims have logical flow. Aim 1 characterizes the mechanism. Aim 2 tests its function and what happens when it's disrupted. A reviewer can see immediately that these aims fit together.

• Aims are independent enough that failure of one doesn't kill the others. If Aim 1 reveals that apocrine secretion doesn't happen the way I hypothesized, Aim 2 could still produce useful results about CSF content. Reviewers look for this kind of risk distribution.

The Four-Part Structure of Each Aim

For each aim and sub-aim, I used a repeating four-part structure:

1. Rationale and Preliminary Data

2. Experimental Design

3. Anticipated Results and Interpretation

4. Limitations and Alternative Approaches

This structure is close to standard for NIH-style proposals, and reviewers expect it. Using a familiar structure reduces the cognitive load on your reviewer, which means they spend more time thinking about your science and less time figuring out where you are in the argument. Let me walk through how I executed each part for my first sub-aim.

Aim 1A: Full Walkthrough

Rationale and Preliminary Data

ChP epithelial cells in adult mice display apocrine secretion at a baseline rate that is increased by elevation of intracellular Ca2+ by application of 5HT2C agonist WAY-161503 (WAY). Apocrine release may also be an important developmental mechanism, as 5-HT is present in embryonic CSF and temporally regulated with relation to critical developmental timepoints. The ChP begins to express 5HT2C as early as E11.5, and my preliminary experiments suggest that at embryonic day 14.5 and 16.5 (E14.5, E16.5) ChP is responsive to 5HT2C activation, displaying immediate early gene transcription (c-fos) and an increase in apocrine blebs at 30' after WAY injection. To determine whether the development of ChP apocrine secretion is temporally aligned with neural development, I will use in vitro, ex vivo, and in vivo approaches to fully characterize the apocrine secretion apparatus from the time of ChP differentiation (E11.5) to birth (P0).

This section has two jobs: explain why this specific aim, and show that I have preliminary data to support it.

Notice how I built the case. I started with what we already know (adult ChP does this thing). I extended it to why it might matter in development (5-HT is present in embryonic CSF). Then I gave the reader my own preliminary evidence that this is feasible at E14.5 and E16.5 (c-fos transcription, apocrine blebs after WAY injection). By the time I got to "I will use in vitro, ex vivo, and in vivo approaches," the reader already believed the aim was worth pursuing.

The preliminary data is doing critical work. I'm not just proposing to look for something that might or might not exist. I'm proposing to extend a phenomenon I've already observed.

Experimental Design

I will use our lab's full panel of ChP imaging modalities to characterize and quantify the effect of 5HT2C stimulation on apocrine release throughout mouse development. (1) I will deliver WAY to pregnant dams by IP injection at E12.5, E14.5, E16.5, and E18.5/P0. After 30', I will microdissect LV ChP explants, fix the tissue, and map and count apocrine events using immunostaining and SEM. I will use a pilot of n=10-12 embryos at each time point to determine variance and use a power calculation to determine how many total animals I will need for significance. I will use ImageJ to quantify blebs in mice that received WAY vs. a saline vehicle control and perform Welch's t-test and other statistical measures as appropriate. (2) To control for cell health and avoid mistaking apoptotic for apocrine blebs, I will use immunostaining for cleaved caspase 3 (CC3) as well as release of nuclear DNA. (3) To determine whether results from explant approaches can be extended to intact preparations, I will image GFP-Lifeact mice with an implementation of our lab's in vivo imaging setup that allows us to visualize 4V ChP in live embryonic mice still attached to their dam.

Things to notice here:

Specificity. I named exact timepoints (E12.5, E14.5, E16.5, E18.5/P0), exact compound (WAY), exact delivery method (IP injection to pregnant dams), and exact imaging modalities (SEM, immunostaining, in vivo imaging). Compare this to the much weaker alternative: "I will treat mice at various developmental timepoints and look for apocrine secretion." Specificity signals that you know what you're doing. Vagueness signals the opposite.

Statistical rigor. I named n=10-12 for my pilot, mentioned a power calculation for final sample size, specified Welch's t-test, and named my quantification software (ImageJ). Reviewers want to know you've thought about how you'll analyze your results, not just how you'll generate them.

Built-in controls. I explicitly called out cleaved caspase 3 staining and nuclear DNA release as controls to distinguish apocrine secretion from apoptotic blebbing. This anticipates a concern a reviewer might have (how do I know these aren't dying cells?) and addresses it inside the experimental design.

Structured enumeration. I used (1), (2), (3) to organize three distinct experimental approaches within the same aim. This makes the plan easy to follow and easy to refer back to later.

Anticipated Results and Interpretation

If the ChP uses apocrine secretion to release growth factors necessary for cortical development, this mechanism would have to be functional by E12.5-E14.5 for cortical neuron differentiation and migration. Based on 5HT2C expression at E11.5 and my data suggesting 5HT2C functionality at E14.5 and E16.5, I expect to see that stimulation leads to increased apocrine secretion beginning by E12.5 both in (1) explants and (3) in utero. These observations, indicating that ChP apocrine secretion occurs contemporaneously with neuronal differentiation and migration, would support my hypothesis that the ChP provides developmentally timed growth factors in the CSF. If apocrine secretion is not functional at these embryonic timepoints, this would suggest ChP uses conventional vesicular exocytosis to release developmental factors and that apocrine release has other physiological purposes in the ChP.

This section demonstrates scientific reasoning. I told the reader what I expect to find (apocrine secretion beginning by E12.5), why I expect it (based on prior 5HT2C expression and functionality data), and what it would mean if I'm right (supports my developmental timing hypothesis).

Critically, I also named what it would mean if I'm wrong. If apocrine secretion isn't functional at these timepoints, that doesn't kill the project. It just means apocrine release serves other purposes and ChP uses conventional vesicular exocytosis for developmental factors. That's still a useful finding.

Naming alternative interpretations builds trust. It shows the reviewer that you've thought through what the data might actually look like, not just what you're hoping for.

Limitations and Alternative Approaches

(1) Dam-attached imaging is technically challenging, and we have not yet optimized this for GFP-Lifeact mice. If I cannot attain sufficient resolution to see membrane blebbing in single cells, I will use our placenta-attached setup which allows for increased stabilization, but at the expense of the embryo's natural uterine environment. (2) It is unlikely that 5-HT is the only apocrine secretagogue. If I see embryonic apocrine secretion but it is not increased upon 5HT2C activation, I will test other molecules that have been demonstrated to increase intracellular Ca2+ in ChP including bitter tastants, nicotine, and inflammatory lipid metabolites.

This is the section applicants most often skimp on, and it's one of the most important. Here I named two specific challenges and gave concrete backup plans for each.

For (1), I acknowledged a real technical difficulty (dam-attached imaging is hard and not yet optimized) and proposed a specific alternative (placenta-attached setup) with a trade-off (better stability, but less natural environment).

For (2), I acknowledged a real biological uncertainty (5-HT probably isn't the only trigger) and named specific alternatives to test (bitter tastants, nicotine, lipid metabolites).

Addressing limitations doesn't weaken your proposal. It strengthens it. Reviewers can always think of limitations themselves. The question is whether you've thought of them too, and whether you have realistic plans for dealing with them.

Aims 1B, 2A, and 2B: Structural Summary

I won't reproduce the full text of the remaining sub-aims, but they all followed the same four-part structure.

Aim 1B investigated the molecular machinery of apocrine secretion. I proposed to use 2-photon imaging of GFP-Lifeact ChP explants with pharmacological blockers of specific pathway components (PLC, PLD, ROCK, myosin II, etc.) to figure out which molecular players were necessary for 5HT2C-evoked apocrine secretion. The limitations section acknowledged that 5HT2C has intracellular pathways independent of PLC and PLD, and proposed ERK1/2 as an alternative mechanism to test if the first round of blockers didn't work.

Aim 2A asked what was actually being released in apocrine secretion events. I proposed using immunoassays and mass spectrometry to characterize the secretome, plus a 5HT2C knockout mouse line to confirm necessity. This aim also included a non-selectivity question: does apocrine release include organelles, which would suggest a fundamentally different secretion mechanism from vesicular exocytosis. The limitations section acknowledged that explants remove tissue from its physiological environment and proposed cross-validating findings with in vivo CSF collection.

Aim 2B tested what happens in a disease-relevant context. I proposed using a maternal immune activation paradigm (poly I:C injection during pregnancy) to ask whether inflammation dysregulates ChP apocrine secretion and changes the developmental factors released into CSF. The limitations section was the longest of any aim, because there were many ways this could go wrong and I needed to show I'd thought through all of them.

The reason these later aims don't need as much annotation is that they follow the same logic as Aim 1A. Once you see the four-part structure executed well once, you can replicate it for each subsequent aim.

The Perspective Section

A growing body of work supports the model that the ChP serves as a key mediator of developmental, immune, and endocrine signaling. However, the mechanisms by which the ChP introduces CSF contents remain unclear. This proposal investigates an apocrine release mechanism that enables ChP to respond to changing blood or CSF contents with high-capacity secretion events above and beyond its established vesicular exocytosis capabilities. Understanding this mechanism is necessary for untangling the complex etiology of neurodevelopmental disorders as well as someday harnessing the ChP for targeted drug delivery to the CSF and contacting surfaces.

This is one paragraph and it zooms back out to the big picture. Why does this work matter beyond the specific aims? What does it enable? I named two applications: understanding disease etiology, and enabling targeted drug delivery.

The perspective section is short but important. It leaves the reviewer with the significance of your work, not the technical details. After reading three to five pages of aims and experiments, you want to close with a reminder of why any of it matters.

The Timeline

Mine was two sentences: "All Aims will begin in Year 1. I will complete Aims 1-2A in Year 2 and Aim 2B in Year 3."

You don't need an elaborate Gantt chart. You need to show that the work is achievable within the fellowship timeline. Two sentences is enough if the math works out.

What Successful Research Plans Have in Common

Across fields, successful fellowship research plans share a few traits:

• They're accessible to non-specialists. The reviewers of your plan may be in adjacent fields, not your specific subfield. Write for someone who is scientifically literate but not an expert in your area. Define technical terms when they first appear.

• They show that the work is feasible. Preliminary data is the strongest way to show feasibility, but you can also demonstrate it through detailed experimental design, appropriate controls, and realistic scope.

• They're honest about uncertainty. The best research plans acknowledge what could go wrong and have thought through alternatives. Reviewers don't trust applicants who claim their projects will definitely work.

• They make the stakes clear. Why does this work matter? What does it enable? A research plan that only describes what you'll do, without making a case for why it matters, is incomplete.

What I'd Do Differently Now

Knowing what I know now from three years as a Gilliam Fellow and from working on other grant applications, I would tighten the writing throughout. Research plans reward density. Every sentence should be doing work. Looking back, there are phrases and transitions I could have cut to make room for more substance (or to fit within the current three-page limit).

I would also use slightly more explicit signposting. Phrases like "This paragraph does X" or "The purpose of this experiment is Y" can help a reviewer who is moving quickly. You don't want to be heavy-handed about this, but a little more structure would have helped.

Finally, I would think harder about which figure to include. I used one multi-panel figure, and it was the right choice. But I could have been more strategic about which panels to include and which to cut to save space.

Writing your own Gilliam or fellowship research plan and want feedback? I offer one-on-one fellowship coaching for applicants across STEM fields. Also see my complete guide to the HHMI Gilliam Fellowship for context on the full application.