A few years before I wrote this proposal, a lab at Stanford found a way to break a hamster’s internal clock with a single evening of light. Siberian hamsters kept on a normal light and dark cycle were given a two-hour pulse of light partway through the night, and the next day their whole light schedule was pushed three hours later. It sounds like a mild inconvenience, roughly the equivalent of a bad night followed by a short flight, but for a large share of the animals it was permanent; their circadian rhythms simply collapsed, and their activity scattered randomly across day and night for the rest of the study.
The arrhythmic hamsters also lost much of their memory. They could no longer tell a familiar object from a new one after a day’s delay, and in a maze that normally tests spatial working memory they made only random choices.
That result alone would have been interesting. What made it strange was the follow-up. When the researchers went on to remove the suprachiasmatic nucleus (the SCN, the brain’s master clock) from hamsters that had already been made arrhythmic, their memory came back.
A disordered clock was worse for memory than no clock at all.
The lab’s explanation is elegant: the SCN is primarily a GABAergic nucleus, meaning its output is largely inhibitory, and in a healthy animal that inhibition rises and falls with time of day; an arrhythmic SCN may instead keep the hippocampus under chronically elevated inhibition, which would explain why a drug that blocks GABA-A receptors restored the hamsters’ memory without restoring their rhythms.
I think that explanation is probably right, and I don’t think it’s the whole story. When I came across these findings as an undergrad, what caught me was the shape of the result rather than the mechanism: memory didn’t just depend on having a clock, it depended on the clock being orderly, which suggests the brain isn’t simply remembering better or worse at different times of day, but that the timing itself is part of how memories get made.
My proposal was an attempt to find where that timing could be written down, and the place I landed on was the chromatin in the hippocampus. Specifically, I proposed that light, acting through the ipRGC-SCN pathway, drives the rhythmic binding of the core clock proteins BMAL1 and CLOCK to DNA, and that this binding regulates the deposition of a histone variant called H2A.Z, a known memory suppressor, in the hippocampus; memory formation would then be gated by circadian time, such that a clock that runs erratically would deposit H2A.Z erratically and leave memory unreliable, while a clock that is absent would leave it alone. I also suggested that the oscillation of these epigenetic effects may help preserve the hippocampus, and that when it’s dysregulated it could contribute to Alzheimer’s and aging-associated memory decline.
To be clear about the status of all this: the pieces are published, the connection between them is mine, and it has never been tested.
Light is the main source of information our bodies have about our surroundings in a biological sense, and some of the ways it gets into the brain were unknown until fairly recently. I’ve written before about how light reaches the brain’s master clock, so the short version is that a small population of intrinsically photosensitive retinal ganglion cells (ipRGCs) in the eye detect light and report it directly to the SCN, rather than to the visual areas that build the images we see. What became clear around 2018 is that different subtypes of these cells run to different places and do different jobs. One subtype, the Brn3b-positive ipRGCs, projects to mood-regulating regions outside the SCN; the other, the Brn3b-negative cells that project to the SCN, turned out to be the route by which light affects learning. The mood pathway has attracted a lot of clinical interest for obvious reasons (seasonal affective disorder being the most obvious), but for my purposes the learning pathway was the interesting one, because it runs through the clock.
The evidence that irregular light can damage learning directly comes from another memorable experiment. Mice were kept on a T7 cycle, three and a half hours of light followed by three and a half hours of dark, over and over, and after a couple of weeks they showed both mood and learning deficits; importantly, this happened without sleep deprivation and without their circadian rhythms falling apart, and it depended on the melanopsin-containing ipRGCs. Antidepressants restored their learning, which suggests the mood effect may come first.
So light can reach memory through the clock, and irregular light can impair memory even when sleep and the overall rhythm look intact. The question I wanted to answer was what actually happens at the molecular level in the hippocampus when that signal arrives.
Dynamic histone marks and histone variants have been a hot topic in memory and cognition research ever since the rise of neural epigenomics. DNA doesn’t sit loose in the nucleus; it’s wrapped around histone proteins, and chemical marks on those histones, along with which versions of the histones are present, help determine which genes are accessible and when. One mark in particular, trimethylation of lysine 4 on histone H3 (H3K4me3), is highly enriched at active promoters near transcription start sites and has drawn a lot of attention in hippocampal research, because it increases during memory formation; in fact H3K4 methylation appears to be required for memory formation in general, and it’s associated with cognitive disorders when it goes wrong.
Because it accumulates at transcriptionally active genes and enhancers, clock genes included, H3K4 methylation also gives rise to one of the best examples of a circadian-regulated epigenome, and the methyltransferases responsible have been found to act as coactivators for BMAL1:CLOCK. That was enough to give me a reasonable basis for how light, as an environmental factor, might act on the hippocampus through BMAL1:CLOCK and H3K4me3. It doesn’t quite explain how, or whether, memory formation actually changes as a result; it only establishes that the potential exists.
H2A.Z is what I thought might close that gap. During the active phase, the BMAL1:CLOCK heterodimer sits bound to E-box promoter sites and recruits a variety of molecules, one of them the methyltransferase that methylates H3K4. What’s easy to miss is that the same binding also promotes the deposition of H2A.Z, a variant of histone H2A that weakens the binding between DNA and the nucleosome. H2A.Z has been studied extensively as a memory suppressor, particularly in the hippocampus, and it’s associated with H3K4me3 at promoters and H3K27me3 at enhancers, with its deposition negatively correlated with gene expression (though that last finding comes from work in plants).
Its proximity to BMAL1:CLOCK is no coincidence: BMAL1:CLOCK behaves like a pioneer transcription factor, and its DNA binding promotes rhythmic chromatin opening, contributing to gene regulation and histone modification in this way. In the hippocampus, removing H2A.Z improves fear memory in mice.
If the clock is depositing a memory suppressor on a schedule, then the clock is, in effect, deciding when memories can be written.
That, I think, is where the hamsters come back in. If H2A.Z deposition follows BMAL1:CLOCK binding, then removing the SCN entirely would take the rhythm out of it, and the hippocampus would be left without a scheduled suppressor; a dysrhythmic SCN, on the other hand, would keep driving deposition, just erratically, which is a plausible recipe for unreliable memory rather than no memory. That’s consistent with the ablation result, and it sits comfortably alongside the GABA explanation rather than competing with it; there’s no reason the clock can’t act on the hippocampus through both fast inhibition and slower chromatin changes. It’s also consistent with the finding that mice with BMAL1 deleted in forebrain neurons (the hippocampus included) show memory deficits that depend on time of day, a kind of circadian gating of memory.
Not everything lines up so neatly. Sleep after learning clearly supports the consolidation of fear extinction memory, but the study I leaned on for that found the benefit held regardless of circadian phase, which is a complication for any story in which the timing of consolidation is set by the clock. I’d rather name that than pretend it isn’t there.
H2A.Z is also stranger than a simple memory suppressor, which is part of why I found it so compelling. Its absence is associated with learning-induced gene expression and its presence with steady-state transcription, so it seems to act as a kind of brake on neural plasticity. It accumulates in the hippocampus with age, becoming overrepresented in a way that affects memory formation. One of the genes that controls it is SIRT1, which can reduce H2A.Z levels by tagging it for degradation by the proteasome (shown in prostate cancer cells), and SIRT1 matters to neuroscientists independently because it regulates neurogenesis, counteracting the cell proliferation roles of H2A.Z.
Clock genes themselves also turn up in cancer research, silenced epigenetically in leukemias and other malignancies, and H2A.Z, one of its forms in particular, is implicated in cell proliferation. So there’s a dual functionality here that’s relevant to cognitive neuroscience, sitting right between neurogenesis and memory formation in the hippocampus. Since aging mammals show more and higher H2A.Z peaks that inhibit memory formation, I wondered whether the buildup might have an overall protective effect, some trade between plasticity and stability, that research hasn’t untangled yet.
The experiments I designed to test all this were straightforward on paper and, in the tradition of undergraduate proposals, entirely unfunded. The first would profile where H2A.Z sits in the hippocampus of mice on a normal 24-hour light cycle, comparing the start of their active phase with the start of their rest phase using ChIP-seq, on the reasoning that if H2A.Z is tied to BMAL1:CLOCK binding it must show some rhythmicity of its own. The second would repeat that profiling in mice with and without an intact SCN, on either the normal cycle or the T7 cycle, to see whether H2A.Z is actually regulated by the master pacemaker; I expected ablated mice to look roughly the same on either cycle and intact T7 mice to look much worse, since my whole argument was that dysregulation, not absence, is what does the damage.
The third would knock out SIRT1 and test fear memory at different times of day, on the logic that if SIRT1 clears H2A.Z away, losing it should let H2A.Z pile up and make memory worse, especially during rest; if it didn’t, H2A.Z probably isn’t the main mechanism behind circadian-dependent learning. The fourth combined everything, light cycle, time of day, H2A.Z profiles, and fear conditioning performance, in the same animals, and my written rationale for it was, in full, “We need to go deeper.” A fifth, comparing young and aged mice, appears in the proposal only as “[Imaginary funding ran out so we didn’t get to this experiment],” and the data analysis section consists of the words “[Outsourced to a collaborator more proficient in bioinformatics].”
What I expected to see was less a matter of positive or negative correlation than of robustness: stronger rhythmicity and a larger amplitude in the relationship between BMAL1:CLOCK and H2A.Z in the control conditions than in the aberrant ones. In the T7 conditions I expected the data to be relatively incomprehensible; in the normal conditions I expected more and higher H2A.Z peaks at the times BMAL1:CLOCK should be bound to E-boxes, and fewer and lower peaks when it isn’t, freeing up memory processes and improving performance on fear conditioning tasks. If the fourth experiment worked, H2A.Z peaks should be lowest exactly where recall is best.
If something like this turned out to be true, it would mean molecular memory processes and circadian memory are considerably more complex than we tend to think. It would be naive to chalk up all of memory consolidation to H2A.Z, given how many histone variants and marks epigenomics research is now investigating, but it would make a case for a background process of a circadian nature, a cyclical rhythm to memory formation that isn’t often considered, such that every day, at a set point, our brains “save data” in a time-dependent manner. I find that a genuinely intriguing thought, especially since circadian genes are highly conserved across mammals and there’s no obvious reason the mechanism would be unique to rodents (with the usual reminder that we know a great deal about mice and much less about humans).
It’s also counterintuitive in a way worth sitting with: if the brain is actively suppressing memory processing during the part of the day when it’s taking in the most information, and releasing that suppression on the way into rest, is that because whatever we still remember by the end of the day is what’s worth moving into long-term memory? Or is it for reasons that have nothing to do with evolutionary advantage at all? I don’t know, and chronotype and SIRT1 variation would be the obvious places to look next.
There’s also a methodological point hiding in here that I think matters beyond this one hypothesis. If memory-related chromatin changes over the course of the day, then the time of day an experiment is run becomes a hidden variable in behavioural neuroscience, and a better understanding of the relationship between the circadian epigenome and core clock genes might improve the reproducibility of the field. The same relationship deserves attention in aging, neurogenesis, and Alzheimer’s specifically, and in oncology, where H2A.Z has mostly been studied through the cell cycle and proliferation, both of which have close ties to chronobiology.
What I keep returning to is the hamsters. A clock that was wrong did more damage than a clock that was gone, which says that the brain cares less about whether it knows the time than about whether the time it knows is consistent.
We now live under the most erratic light any mammal has ever been exposed to, with screens at midnight, LEDs rich in exactly the blue wavelengths ipRGCs are most sensitive to, and time zones crossed in an afternoon. If memory formation really is scheduled by a clock that light sets, then erratic light may be affecting cognition more than we currently assume, and not only through lost sleep.
None of the experiments in my proposal were ever run, at least not by me. If someone has since tested any part of this, I would very much like to know whether I was right, or at least wrong in an interesting way.
References
- Baptista, T., et al. (2013). Regulation of histone H2A.Z expression is mediated by sirtuin 1 in prostate cancer. Oncotarget, 4(10), 1673-1685.
- Collins, B. E., Greer, C. B., Coleman, B. C., & Sweatt, J. D. (2019). Histone H3 lysine K4 methylation and its role in learning and memory. Epigenetics & Chromatin, 12(1), 7.
- Dai, X., et al. (2017). H2A.Z represses gene expression by modulating promoter nucleosome structure and enhancer histone modifications in Arabidopsis. Molecular Plant, 10(10), 1274-1292.
- Dunn, C. J., et al. (2017). Histone hypervariants H2A.Z.1 and H2A.Z.2 play independent and context-specific roles in neuronal activity-induced transcription of Arc/Arg3.1 and other immediate early genes. eNeuro, 4(4).
- Fernandez, D. C., et al. (2018). Light affects mood and learning through distinct retina-brain pathways. Cell, 175(1), 71-84.
- Fernandez, F., Lu, D., Ha, P., Costacurta, P., Chavez, R., Heller, H. C., & Ruby, N. F. (2014). Dysrhythmia in the suprachiasmatic nucleus inhibits memory processing. Science, 346(6211), 854-857.
- Fortress, A. M., & Frick, K. M. (2014). Epigenetic regulation of estrogen-dependent memory. Frontiers in Neuroendocrinology, 35(4), 530-549.
- Gräff, J., et al. (2012). Dynamic histone marks in the hippocampus and cortex facilitate memory consolidation. Nature Communications, 3, 991.
- Gupta, S., et al. (2010). Histone methylation regulates memory formation. Journal of Neuroscience, 30(10), 3589-3599.
- Hanoun, M., et al. (2012). Epigenetic silencing of the circadian clock gene CRY1 is associated with an indolent clinical course in chronic lymphocytic leukemia. PLoS ONE, 7(3), e34347.
- LeGates, T. A., et al. (2012). Aberrant light directly impairs mood and learning through melanopsin-expressing neurons. Nature, 491(7425), 594-598.
- Melo, I., & Ehrlich, I. (2016). Sleep supports cued fear extinction memory consolidation independent of circadian phase. Neurobiology of Learning and Memory, 132, 9-17.
- Menet, J. S., Pescatore, S., & Rosbash, M. (2014). CLOCK:BMAL1 is a pioneer-like transcription factor. Genes & Development, 28(1), 8-13.
- Papazyan, R., Zhang, Y., & Lazar, M. A. (2016). Genetic and epigenomic mechanisms of mammalian circadian transcription. Nature Structural & Molecular Biology, 23(12), 1045-1052.
- Ruby, N. F., et al. (2013). Spatial memory and long-term object recognition are impaired by circadian arrhythmia and restored by the GABAA antagonist pentylenetetrazole. PLoS ONE, 8(8), e72433.
- Saharan, S., Jhaveri, D. J., & Bartlett, P. F. (2013). SIRT1 regulates the neurogenic potential of neural precursors in the adult subventricular zone and hippocampus. Journal of Neuroscience Research, 91(5), 642-659.
- Snider, K. H., & Obrietan, K. (2018). Modulation of learning and memory by the genetic disruption of circadian oscillator populations. Physiology & Behavior, 194, 387-393.
- Stefanelli, G., et al. (2018). Learning and age-related changes in genome-wide H2A.Z binding in the mouse hippocampus. Cell Reports, 22(5), 1124-1131.
- Taniguchi, H., et al. (2009). Epigenetic inactivation of the circadian clock gene BMAL1 in hematologic malignancies. Cancer Research, 69(21), 8447-8454.
- Zovkic, I. B., et al. (2014). Histone H2A.Z subunit exchange controls consolidation of recent and remote memory. Nature, 515(7528), 582-586.