Elucidating the complex organization of neural micro-domains in the locust Schistocerca gregaria using dMRI

To understand brain function it is necessary to characterize both the underlying structural connectivity between neurons and the physiological integrity of these connections. Previous research exploring insect brain connectivity has used microscopy techniques, but this methodology is time consuming and cannot be applied to living animals and so cannot be used to understand dynamic physiological processes. The relatively large brain of the desert locust, Schistercera gregaria (Forksȧl) is ideal for exploring a novel methodology; diffusion magnetic resonance imaging (dMRI) for the characterization of neuronal connectivity in an insect brain. The diffusion-weighted imaging (DWI) data were acquired on a preclinical system using a customised multi-shell diffusion MRI scheme. Endogenous imaging contrasts from the averaged DWIs and Diffusion Kurtosis Imaging (DKI) scheme were applied to classify various anatomical features and diffusion patterns in neuropils, respectively. The application of micro-MRI and dMRI modelling to the locust brain provides a novel means of identifying anatomical regions and connectivity in an insect brain. Furthermore, quantitative imaging indices derived from the kurtosis model that include fractional anisotropy (FA), mean diffusivity (MD) and kurtosis anisotropy (KA) could, in future, be used to quantify longitudinal structural changes in neuronal connectivity due to environmental stressors or ageing.


Abstract 25 26
To understand brain function it is necessary to characterize both the underlying structural 27 connectivity between neurons and the physiological integrity of these connections. Previous 28 research exploring insect brain connectivity has used microscopy techniques, but this 29 methodology is time consuming and cannot be applied to living animals and so cannot be used 30 to understand dynamic physiological processes. The relatively large brain of the desert locust, 31 Schistercera gregaria (Forksȧl) is ideal for exploring a novel methodology; diffusion magnetic 32 resonance imaging (dMRI) for the characterization of neuronal connectivity in an insect brain. 33 The diffusion-weighted imaging (DWI) data were acquired on a preclinical system using a 34 customised multi-shell diffusion MRI scheme. Endogenous imaging contrasts from the 35 averaged DWIs and Diffusion Kurtosis Imaging (DKI) scheme were applied to classify various 36 anatomical features and diffusion patterns in neuropils, respectively. The application of micro-37 MRI and dMRI modelling to the locust brain provides a novel means of identifying anatomical 38 regions and connectivity in an insect brain. Furthermore, quantitative imaging indices derived 39 from the kurtosis model that include fractional anisotropy (FA), mean diffusivity (MD) and 40 kurtosis anisotropy (KA) could, in future, be used to quantify longitudinal structural changes 41 in neuronal connectivity due to environmental stressors or ageing. 42 43 44

Introduction: 46
A key challenge of neuroscience is understanding the emergence of behaviour from 47 neuronal activity. The idea of neuronal circuit formation sharing common design principles 48 across different species to generate behavioural equivalent outputs has had a long history since 49 the pioneering work of Ramon y Cajal and his early observations of the similarity in 50 organization between insect and human visual processing 1 . Interestingly, regardless of the 51 number of neurons that comprise the brain of an animal and morphological differences, 52 evidence suggests that the basic principles underlying neuronal connectivity are similar 2,3 . 53 The emergent field of connectomics develops the premise that the understanding of how 54 complex brain function must be related to its structural underpinnings 4 . Initial studies of 55 neuronal connectivity were in Caenorhabditis elegans, an animal with only 302 neurons with 56 simple behaviours 5 . However perhaps both the simplicity of the behaviour of C. elegans and 57 its decentralised nervous system make it an inadequate model animal to link neuronal circuitry 58 to behavioural output in a way that can be generalizable across species. Recent work in the 59 Drosophila melanogaster connectome has greater potential as insects have long provided 60 important model animals for understanding the relation between neural circuitry and behaviour 61 using electrophysiological techniques 6 . Whilst the insights gained from electrophysiology 62 have been essential in understanding how the processing in discrete neural circuits relates to 63 simple behaviours, these techniques may be limited in the insights they can provide about the 64 neuronal network as a whole. Network level understanding could be critical in understanding 65 the emergence of more complex behaviours that inevitably rely on the integration of multiple 66 discrete neuronal circuits. Linking electrophysiology and connectomics could prove to be a 67 very powerful methodology in understanding complex adaptive behaviour 7,8 . 68 The behaviour of the fruit fly Drosophila melanogasta, is far more diverse than that of C. 69 elegans and the animal has a correspondingly more complex nervous system, with an estimate 70 of 135,000 neurons. Shih et al (2015) produced a connectome of the Drosophila nervous 71 system, using 12995 images of projection neurons collected in 'FlyCircuit' 9 . The neurons were 72 found to connect 43 local processing units resulting in the formation of five distinct modules: 73 left visual, right visual, olfactory, mechano-auditory and pre-motor module. Terminology 74 based on the graph theory origins of connectomics have been used to characterize similarities 75 in neuronal network structure across phyla. For instance fly brain connectomes demonstrated 76 the presence of 'rich club' organisation in insect brains, that is 'nodes' that are extremely linked 77 and can be measured by the overall strength of connections 9 . Evidence of this and 'small world' 78 architecture in the formation of neuronal processing networks has been observed in mammalian 79 brains 10 . This form of analysis can be used to argue that whilst the mammalian brain is far 80 more complex than the insect brain, the key characteristics of its network formation are 81 essentially equivalent, thus providing further support for the concept that insect brains are 82 excellent model systems for understanding the neuronal basis of complex, adaptable behaviour 83 4,11 . Whilst these approaches from graph theory provide an overall computational approach to 84 understanding neuronal functioning they cannot be easily related to biological changes that 85 occur in fibre tracts during development, ageing or injury. 86 Thus far the construction of the complete connectome using C. elegans and Drosophila has 87 relied on the use of semi-automatic processing to produce reconstructed electron microscopy 88 images 12 . The resolution of these technique enables comprehensive synapse level connectivity 89 to be determined in an adult fly brain 11,13 . This remains a time realistic approach for these 90 animals, given their simpler nervous system, and one that has begun to characterize 91 relationships between neuronal architecture and behaviour. However, these techniques are 92 limited 14 . The reconstruction of electron microscopy images remains very time consuming 93 process even in these animals with simpler nervous systems. The process of reconstruction 94 itself is also susceptible to error. Microscopy further necessitates the animal being dead and so 95 presents a static image of neuronal connectivity 15 . This approach also limits sample size, with 96 the methodology focused on generating a standard connectome for the species rather than 97 providing the possibility of a comparative assessment of differences across animals. To truly 98 understand the link between neuronal networks and adaptable behaviours the techniques for 99 characterizing the network may need to be applied longitudinally over an animal's lifespan. 100 Further it would be highly advantageous for a visualization technique to be able to quantify 101 micro-structural changes that occur that may reflect damage, ageing or adaptive plasticity in 102 the nervous system. Such a technique would enable mechanistic questions regarding the causes 103 of any alterations in the microstructure of the brain to be addressed. 104 In parallel to the connectome research in simpler animals, the same interest in examining 105 neuronal connectivity has developed in mammals. In these studies the technique applied has 106 been magnetic resonance imaging and the modelling of the diffusion weighted imaging (DWI) 107 via diffusion tensor imaging (DTI) 16 . The body of all animals, including insects, has a high 108 water content, and in a biological tissue water molecules are in a constant state of random 109 motion. dMRI utilizes the principle that differences in water diffusion in tissues can be used to 110 indirectly infer underlying anatomy via the direction of preferential diffusion and from this 111 provide unique information on their microstructural architecture 17 . dMRI provides diffusion 112 estimates for each voxel in a series of DW images. DTI and higher order diffusion schemes can 113 then describe the estimated water self-diffusion in different dimensions by a diffusion tensor 114 or more advanced representation schemes. This tensor mathematically represents how the 115 composite architecture of the body provides structural barriers differentially restricting the 116 directions of water diffusion. The directionality of water diffusion that emerges due to these 117 biological barriers in the body is known as anisotropic or directional diffusion, in contrast to 118 isotropic diffusion where there is no directionality. In the brain the biological barriers to water 119 diffusion include axons, cellular membranes and proteins. Therefore, DTI allows the 120 visualization of fibre pathways and connectivity. Importantly, DTI does not directly visualise 121 axonal connections, but is a statistical inference of structural connectivity 18 19 . However DTI 122 is limited by assumptions intrinsic to it regarding the Gaussian distribution of water self-123 diffusion within the body, a simplification that will reduce the accuracy of the detail that the 124 method can provide about the structures. Diffusion Kurtosis Imaging (DKI) adds additional 125 dimensionality to the model of diffusion and attempts to capture the non-Gaussian distribution 126 of diffusion to give better insights into differences in the barrier for molecular diffusion and 127 from this provide a more accurate inference of the structure of the underlying tissue 20 . 128 The DTI and DKI methods, as well as inferring the underlying microarchitecture that is 129 creating the barriers to diffusion, enables the quantification of differences between different 130 regions of the brain in terms of how they impact to the diffusion of water. There is evidence 131 that measures derived from DWI via DTI are sensitive enough able to detect pathological 132 changes in axonal architecture that are associated with the osmotic swelling occurring as a 133 consequence of alterations in membrane polarization 21 . Such axonal changes are thought to be 134 indicative of Wallerian degeneration or axonal self-destruction that occurs in disease and after 135 trauma 22 . Understanding the mechanisms behind the altered architecture of axons is important 136 as this pathological process is thought to be characteristic of many neurodegenerative diseases. 137 Although there are morphological differences between insects and vertebrates, similarities at 138 the cellular and molecular level, with both neural signaling and the innate immune response 139 highly conserved from insects through to mammals 23 , have led to insects being proposed as 140 important models for understanding the biological pathways underlying axonal self-141 destruction. Fractional anisotropy (FA) is the most widely applied metric that quantifies 142 differences in water diffusion by biological barriers, and provides a measure of strength of 143 directionality of diffusion in a given region. Differences in FA have been used widely in DTI 144 studies of the human brain and have found to be of predictive value in characterizing the 145 severity of pathological processes in the CNS due to trauma or neurodegenerative diseases 16,20 . 146 DTI or DKI has not previously been implemented on insects but the application of dMRI 147 and DWI scans provides the advantage of providing a much faster means of visualising the 148 locust brain microcircuitry and neural tracts. To be able to use DTI or higher-order diffusion 149 schemes 24,25 to characterize the neural connectivity of simpler species up to more complex 150 animals will allow us to build upon knowledge in the growing field of comparative 151 connectomics 26 . A recent study applied dMRI to the squid brain, both confirming the presence 152 of connectivity previously established by microscopy techniques, and also identifying many 153 previously undescribed pathways 27 . Further the application of dMRI enables the novel 154 opportunity for quantification of indices that may reflect structural alterations that are occurring 155 due to disease, injury or plasticity. Due to the known commonalities at the cellular and 156 molecular level between insects and vertebrates, the use of these quantitative measures in an 157 insect model in combination with genetic tools 28 has potential in the future to provide insights 158 into axonal biology in vivo.

MR imaging 172
The DWI data were acquired at the bore temperature on a horizontal bore 7T Biospec micro- Where µi= ∫ ( ) , the cumulants (κi) can be described in terms of moments of 201 probability distribution (µi). The first three cumulants ( κ1-3) are equal to the first three moments 202 ( µ1-3). µ, µ2 and µ3 are the mean, variance and the skewness of the distribution, respectively. 203 The fourth cumulant is related to the kurtosis as shown in eq. 2. Diffusion Kurtosis can be 204 derived from dMR signal by using the Fourier relationship between the attenuated signal and 205 the EAP. The logarithm of the diffusion signal can be expanded as a summation of the 206 cumulants κn of P(R) 33 : 207 Under the assumption that diffusion is symmetrical (symmetric EAP), the phase of the 209 attenuated signal can be considered zero, therefore, all odd order cumulants are null, i.e., 210 For isotropic Gaussian diffusion in time Δ, the diffusion coefficient can be expressed as: 212 D=κ2/(2Δ) 24 , and by substituting it in eq. 2, the fourth cumulant can be written as: κ4=4KD 2 Δ 2 213 . 214 Under the assumption of PGSE, the diffusion weighting parameter 'b-value' is defined as: 215 Using the relations of the second and fourth cumulants and eq.5, the signal attenuation can be 217 approximated by the quadratic exponential kurtosis model after truncating eq. 4 to the second 218 term: 219 Where, b≈Δ(2πq) 2 , and when δ ≈ Δ (violation of narrow pulse approximation), the effective 221 diffusion time 'τ' (τ = Δδ/3) should be used in eq. 5 instead of Δ, i.e., 222 2 2 2 bG     values is very small (maximum b-value is too low) then the attenuation in the signal intensity 235 will be very low and the estimation of Dapp will be prone to noise. On the other hand, if the 236 range is very large (max b-value is very high), then the estimation of diffusivity will incur a 237 systematic error due to the omission of higher terms from eq. 4. 238 Diffusion MRI data processing 239 The noise profile of DWI at higher b-values is non-Gaussian, therefore, to reduce the 240 influence of noise, NLMeans filter with rician noise estimation was used for each dwi dataset 241 35 . The noise compensated datasets were then processed to correct for Gibbs ringing artefacts 242 36  tract connectivity in the locust brain to be extracted based on the diffusion direction over a 266 series of DWIs. Figure 1 shows multiple anatomical features of the inset in coronal, axial and 267 sagittal planes. For each plane, the top row shows the anatomical features using T2-W contrast, 268 whereas the bottom row shows the anatomy using the contrast generated from the averaged 269 DWIs. 270 In addition to the anatomical features, quantitative metrics can be extracted from the 271 DWIs, including FA. The use of these metrics have the potential for the visualization of the 272 locust brain to be quantified and therefore used to correlate with observed physiological 273 changes. Figure 2 shows a coronal slice with a number of contrasts derived from the acquired 274 DWIs and the Kurtosis model. In b0 or T2-W image (Figure 2-a), the highest contrast is from 275 fat bodies, whereas, in Figure 2  would be indicative of these regions having a high directionality which would be consistent 287 with the visual processing pathway, from the retina through the optic lobes towards the central 288 brain. In contrast, the optic lobe regions are hypo-intense in mean diffusivity map, which is 289 thought to relate to higher levels of structural complexity. This finding would reflect the 290 layering of synaptic networks, with each neuropil known to consist of both columnar 291 arrangement of neurons and also local interneurons 44 . 292 Using the color encoded projections of synaptic neuropils, it is possible to estimate the 293 micro-diffusion environment (Figure 3). The axonal fibres connecting the synaptic centers of 294 the optic lobes and neuronal projections towards the brain can be seen from Figure 3 (a). 295 Regions that show high contrast in Figure (Table 1). Using MK and KA, the complex organization of the underlying 308 microstructure can be quantified. For example, Lox exhibited highest kurtosis values due to its 309 complex micro-structural (diffusion) environment mapped at 78 µm 3 scale (Table 1), whereas, 310 CX, despite having high synaptic density showed lower kurtosis values as well as low fibre 311 density (Figure 3a-inset), which indicates that much higher resolution is required to better 312 classify high synaptic regions. 313 Interestingly the quantitative metrics for each region were within a similar range for 314 each animal. In future studies for these values to have utility as baseline dMRI metrics there 315 would need to be an increase in sample size, but the initial results are supportive that dMRI 316 metrics could be used to examine structural differences in relation to differing physiology.

Discussion and Conclusion 366
There is increasing evidence to support the early observations by Ramon y Cajal which 367 suggested that there were common design principles behind the formation of neuronal networks 368 to subserve complex brain function across phyla 1,3 . Therefore, the insect brain offers a simpler 369 and more we used number of contrasts and by employing a Kurtosis tensor we were able to classify the 398 diffusion patterns in neuropils (small axons, dendrites and synaptic terminals) as well as in cell 399 bodies, which are usually difficult to assess using DWI or simple diffusion models. 400 The application of the dMRI to visualize the insect brain and optic lobes is particularly 401 beneficial for understanding the formation of neuronal circuits and the potential impact of 402 physiological or environmental change on these circuits. In the insect, the visual system is 403 proportionally by far the largest region as may be expected by the necessity of vision for the 404 animal and the demanding computation it requires. There has been extensive prior 405 characterization of the visual system of insects, particularly in flies but also in locusts and bees 406 from the first observation of the similarity in the organization for insect and mammalian visual 407 systems 44 . The insect visual system is known to be energetically demanding on the animal, 408 with the circuitry highly evolved to efficiently and rapidly transmit information to the central 409 brain 54 . Interestingly there is evidence that there is experience dependent plasticity in the fly 410 optic lobe 55 . Although the visual system in the insect is already well characterized through 411 microscopy and also electrophysiology, the application of dMRI has the potential to detail 412 dynamic changes that occur in visual system wiring that is necessitated by changing demands 413 on the animal. 414 The novel application of dMRI and its signal representation schemes based 415 methodology for insect brain imaging is preferable over other possible imaging techniques for 416 many reasons. dimensionality of the diffusion model or higher field MR. However, although these problems 469 have been inherent in dMRI even as applied to mammalian brains, there is proven value in the 470 utility of dMRI metrics in predicting patient outcomes. 471 The linkage of metrics from dMRI and DKI with existing well established techniques 472 to characterize functional output of the network provides a powerful methodology to unravel 473 how neuronal structure impacts on behavioural output. This methodology will enable a 474 different and yet complementary approach to understanding the dynamics of neural circuits in 475 development, ageing and stress. Importantly the technique provides the possibility of 476 quantitative as well as qualitative imaging of neuronal structure. The locust is potentially good 477 insect model to develop these studies further as they enable questions to be addressed from a 478 cellular to systems level. 479

Data Availability 480
The raw dMRI volume data used in the study are available from the corresponding 481 author on reasonable request 482