| dc.description.abstract | The Suprachiasmatic Nucleus (SCN), the region of 20,000 neurons in the mammalian
hypothalamus that controls the circadian rhythm, is a complex network of neurons, glia, and
oscillatory dynamics. When measuring the electrical or molecular activity of the cells that
compose the SCN, one will see a variety of oscillatory behaviour. One of the most notable
of these oscillations is the core transcription-translation feedback loop (TTFL), consisting
of clock genes such as period (per) and cryptochrome (CRY). By producing transgenic
animals in which the activity of these clock genes is reported via fluorescent proteins, it
becomes possible to track their expression from the intensity through long-time imaging.
In the past, this has revealed a regional specificity to the rhythms. Neurons in the dorsal
region of the SCN complete this loop with a period of slightly under 24hrs. The time of peak
expression, as measured by PER1 concentration, appears 1−3hrs earlier than the slower
ventral region, which completes the feedback loop in slightly over 24hrs. Observation of
localised phase gaps between clusters has led to many questions about the SCN network
structure that supports them and their functionality. For example, do they confer robustness
to the oscillations of the TTFL? Or perhaps they indicate different times of input and output
signalling, activity, and time-of-day information consolidation? Despite the importance of
these questions, researchers have found a conflicting number of clusters, dynamics, and
configurations. Further, the functionality and interaction structure of the network that supports
these phase gaps and clustering configuration has remained unclear.
To that end, we will study the gene expression dynamics of individual neurons in the SCN
network to elucidate some of these unknown network properties. We hypothesise that, when
categorising individual neurons into ventral and dorsal clusters by the phase of their gene
expression, certain neurons will appear to “switch” clusters throughout the time-series. That
is, given a neuron that is expressing in-phase with the ventral cluster early in the time-series,
we suspect that the SCN network can cause it to switch and start to express in-phase with
dorsal cluster later on. Equally, we suspect that neurons from dorsal cluster might also switch
to ventral in this same way. To test this hypothesis, we developed a new data analysis pipeline
for SCN gene expression imaging data that is capable of identifying, tracking, and extracting
intensity measurements from individual neurons in images over many days. Once data was extracted, we cleaned, transformed, and clustered it using a k-medoids clustering algorithm
to identify neurons with similar expression profiles. We found that, as previous research
noted, neurons had a slight preference for localisation – with phase leading neurons located
dorsally and phase lagging neurons located ventrally. Further, when clustering the neurons
in varying window sizes, we found that the size of each cluster remained static for window
sizes equal to or larger than 12hrs. Whereas in window sizes smaller than 12hrs, cluster
size appears to oscillate robustly through the duration of our study. Power spectra analysis
reveals that this oscillation in configuration size happens with a 12hr period. The dorsal
cluster peaks in size near circadian dawn and dusk (Circadian Time 0 (CT0) and CT12) and
the ventral peaks in the middle of the day and night (CT6 and CT18). We hypothesise that
this is related to time-of-day information consolidation across the SCN network and discuss
the importance of this and propose future work to test this hypothesis at the end of Chapter 3
and in the Conclusion.
Using these results, we informed a coupled phase oscillator model to describe the
switching dynamics. The Hansel, Mato, and Meunier (HMM) phase model is a second
harmonic model with mixed signs for the first and second harmonic. The original authors
found a variety of long-time dynamical behaviours, including heteroclinic switching. Through
numerical continuation and simulation of a large network of HMM phase oscillators, we study
the viability of this model to describe the switching we observed in our data. Starting with an
identical oscillator and all-to-all symmetrically coupled network, we find that the switching
dynamics occur in a bulk fashion – i.e., all constituents switch in a single instant in each
cycle. This stood in contrast to our data, where switching occurred continually throughout
the cycle. To rectify this we made a number of biophysically motivated modifications to
the HMM network (nonidentical oscillators, asymmetric network strengths, and additive
Gaussian noise) and found that breaking these symmetries creates a significant amount of
pitchfork bifurcations with branches that contain emergent oscillatory dynamics. Simulation
of cluster membership dynamics for the biologically motivated system results in a cluster
analysis that has similar features to what we report in our data. We believe that this indicates
that SCN asymmetries, such as mixed sign coupling, nonidentical oscillators, asymmetric
network strengths, and noise are vital components to supporting SCN cluster membership
dynamics. | en_GB |
