Compartment and cell type-specific hypoxia responses in the developing Drosophila brain

Environmental factors such as the availability of oxygen are instructive cues to regulate stem cell maintenance and differentiation. We used a genetically encoded biosensor to monitor the hypoxic state of neural cells in the larval brain of Drosophila. The biosensor reveals brain compartment and cell type specific levels of hypoxia. The values correlate with differential tracheolation that is observed throughout development between the central brain and the optic lobe. Neural stem cells in both compartments show the strongest hypoxia response while intermediate progenitors, neurons and glial cells reveal weaker responses. We demonstrate that the distance between a cell and the next closest tracheole is a good predictor of the hypoxic state for that cell. Our model concludes that oxygen availability is the major factor controlling the hypoxia response in the developing Drosophila brain but cell intrinsic and cell-type specific factors contribute to modulate the response in an unexpected manner.


INTRODUCTION 44
In stem cell niches the supply of oxygen and nutrients is tightly controlled (Schofield,45 1978;; Morrison and Spradling, 2008)  As the nervous system develops in the embryo, tracheal cells invade the brain along 75 the dorsal midline and build the network of respiratory tubes, called tracheoles that 76 oxygenate the brain during larval life. In Drosophila these air tubes have a stereotyped 77 branching pattern, making it possible to draw a detailed map of the larger tracheoles 78 reaching each brain region (Pereanu et al., 2007). The present study was prompted by 79 the observation that in the developing larval brain tracheoles are not distributed as 80 brain (Pereanu et al., 2007;; Misra et al., 2017). Here, we used confocal laser 118 microscopy and transmission electron microscopy to further define this morphological 119 and functional dichotomy and investigated whether this condition prevails throughout 120 larval development. In brains immunolabelled for the synaptic marker Bruchpilot (Brp) 121 (Kittel et al., 2006;; Wagh et al., 2006) the staining co-localizes with regions of the brain 122 that are densely tracheolated ( Figure  1A, B) and that correspond almost entirely to the 123 synaptic centres (i.e. neuropils, see for example (Iyengar et al., 2006)). In contrast, 124 very little synaptic staining was found within the optic lobes ( Figure  1B). Hence, there is 125 a close topographic correlation between a dense tracheolation of the synaptic neuropil 126 in the central brain and a sparse tracheolation in the optic lobes, where there are no 127 synapses but instead undifferentiated progenitor cells ( Figure  1C). A close examination 128 of the border between these two brain regions, using transmission electron microscopy 129 disclosed the existence of a sharp interphase between two types of cell bodies (inset in 130 Figure 1C and 1D). On the side of the central brain we found glia, tracheoles and 131 neuronal cell bodies extending thick neurites into the neuropil where they formed 132 synapses (not shown). These cell bodies had a relatively large cytoplasm containing 133 abundant mitochondria, endoplasmic reticulum, ribosomes and other organelles, as 134 expected for differentiated neurons ( Figure 1E). On the opposite side ( Figure 1F), 135 within the medial region of the optic lobes, we found large numbers of smaller cells, in 136 which the nucleus was surrounded by a thin ring of cytoplasm, with fewer organelles 137 and with the typical columnar arrangement of the yet not fully differentiated neuronal 138 progeny generated in this proliferative region (Meinertzhagen and Hanson, 1993;; 139  We previously reported that at a late stage of larval development, 96 hrs after larval 141 hatching (ALH), the sparsely tracheolated optic lobe has lower hypoxia values than the 142 densely tracheolated central brain (Misra et al., 2017). To investigate if this condition is 143 specific for the end of larval life or prevails during a longer developmental interval and 144 is thus of potential relevance for brain development we extended our analysis to seven 145 time points of larval development at 12 hr intervals, from 24 to 96 hrs ALH. We found 146 that the segregation of tracheoles exists already by 24 hrs ALH ( Figure 2A) and 147 persists throughout larval life (Figure  2A-F). We confirmed that the optic lobe grows in 148 size during this time ( Figure 2G)  although not enough to compensate for optic lobe growth from 48h ALH onwards, and 154 thus the proportion of optic lobe tissue devoid of tracheoles appears to increase with 155 age ( Figure  2I).

157
Oxygen availability triggers differential hypoxia response between central brain 158 and optic lobe 159 Since our original hypothesis stated that the sparse tracheolation of the optic lobes will 160 result in a condition of chronic hypoxia relative to the central brain, we used a HIF-161 1a/Sima based hypoxia sensor (Misra et al., 2017) to monitor hypoxia in these two 162 brain compartments throughout larval life. The results were consistent with this 163 prediction because mean biosensor ratiometric values were significantly lower for the 164 optic lobe (stronger hypoxia response) as compared to the central brain at 36, 60 and 165 84 hrs ALH (Figure  3;; mean ratiometric values at 36 hrs: 0.90 for central brain, 1.23 for 166 optic lobe, n=6;; at 60 hrs ALH: 0.89 for central brain, 1.22 for optic lobe, n=6;; at 84 hrs 167 ALH, 0.90 for central brain, 1.31 for optic lobe, n=8). Our results presented here so far 168 strongly suggest that the dense tracheolation of the central brain results in higher 169 oxygenation of this compartment in comparison with the sparsely tracheolated optic 170 lobe.

172
The distance between a cell and its nearest tracheole can predict its cellular 173 hypoxia response 174 To further investigate the relationship between tracheolation and the distribution of 175 oxygen in the brain we decided to focus on the lateral optic lobe tracheole (OLTl) 176 ( Figure  4B). The study of the OLTl tracheole provided key insights into the way oxygen 188 diffuses from a given tracheole and prompted us to apply the same analysis in a 189 generalized way to the entire brain hemisphere. We produced a map that contains the 190 coordinates of every cell in the brain and measured the minimum distance between 191 each cell and neighbouring tracheoles in the entire brain. 1/Ratio values plotted as a 192 function of minimum distance to trachea show an inverse relationship that is best fitted 193 by a decaying exponential function ( Figure 4C and 4D). The result demonstrates that 194 the hypoxia response of individual cells correlates with their position in the brain in 195 relation to the tracheal system. Using the best fit exponential function, we predicted 196 ratiometric values according to cell-to-tracheole distance and depicted them with a 197 colour scale in a heat map ( Figure  4E). The image strongly resembles the distribution 198 of measured ratiometric values (compare Figure 4E to 4A), which indicates that the 199 distance to tracheoles can reliably predict the hypoxic, or conversely, the oxygenation  in relation to distance to tracheoles was reduced, which indicates oxygen saturation of 219 the brain and the loss of the difference between central brain and optic lobe ( Figure  5C;; 220 normoxia: black curve, l: 242 µm, hyperoxia;; blue line, l: 653 µm). 221 Next, we exposed larvae to ambient hypoxia by raising them in 5% oxygen from 60 hrs 222 to 84 hrs ALH in order to investigate whether the sensor is able to show decreased 223 oxygen availability in the brain. We observed an increase in ratiometric values in both 224 central brain and optic lobe compartments ( Figure 5D  The results of our brain-compartment analysis prompted us to investigate the hypoxia 242 response in different cell types found in the central brain and in the optic lobe ( Figure  243 6). For this we combined the ratiometric analysis with immunofluorescence labelling for 244 cell type specific nuclear marker proteins. We marked neuroblasts with an antibody 245 against Deadpan (Dpn), ganglion mother cells (GMCs) with an antibody against 246 Prospero (Pros), neurons with an antibody against Embryonic lethal abnormal visual 247 system (Elav) and glial cells with an antibody against Reversed polarity (Repo) ( Figure  248 6A-D). Neuroepithelial cells were segmented based on a staining with an antibody 249 against Disc large (Dlg) using TrakEM2 in Fiji ( Figure  6E). Image stacks obtained from 250 the immunostainings with these markers served to produce cell type-specific 251 segmentation masks for the ratiometric analysis ( Figure  6A´- E´). 252 In general, and as expected from our brain-compartment analysis, each cell type in the 253 central brain showed a lower hypoxic response as compared to the same cell type in 254 the optic lobe ( Figure 6F). Interestingly, however, we observed cell-type specific 255 differences in the hypoxia response between cell types, regardless of the localisation of 256 the cells in the brain. Neuroblasts (Dpn-positive cells) in the optic lobe showed the 257 strongest hypoxia response, while ganglion mother cells (Pros-positive cells) in the 258 central brain showed the weakest hypoxia response of all analysed cell types in the 259 brain (mean ratio values for optic lobe neuroblasts: 1.51 and for central brain ganglion We next measured the average minimum distance between cells of a defined cell type 268 and neighbouring tracheole ( Figure  6F). The average 1/Ratio values (oxygenation) for 269 most cell types in the central brain and the optic lobe followed a decaying exponential 270 function as described above. Of interest is that central brain and optic lobe neuroblasts 271 showed a stronger and central brain ganglion mother cells a much weaker hypoxia

Neuroepithelial cells are resilient to differential oxygen levels 280
In order to investigate the idea that certain cell types are less efficient than others in 281 sensing oxygen levels via the canonical hypoxia pathway we compared the hypoxia 282 response in neuroepithelial cells and neuroblasts under ambient hyperoxia ( Figure  6H). 283 Interestingly, while neuroblasts in the central brain and in the optic lobe were able to 284 greatly adapt their hypoxia response to different oxygen levels, neuroepithelial cells of 285 the IPC and OPC changed their response to a much lower degree. This indicates that 286 neuroepithelial cells might be less susceptible to changes in oxygen levels. 287 This finding prompted us to investigate a dataset with genome-wide information on 288 larval brain gene expression (Southall et al., 2013). In this study, cell-type specific 289 targeted DamID methods were used to compare Polymerase II occupancy between 290 neuroepithelial cells and neuroblasts in the third instar larval brain at age 96 hrs ALH. 291 In both progenitor cell types, glycolytic genes were significantly enriched, indicative of a 292 lower oxygen availability to use oxygen for oxidative respiration for cellular energy 293 production. Moreover, enrichment in hypoxia pathway genes was found in the 294 neuroblast specific gene catalogue but not in neuroepithelial cell specific gene 295 catalogue ( Table 1). The results support the notion that in neuroepithelial cells the 296 canonical hypoxia pathway might be less relevant in order to respond to differential 297 oxygen levels. We find a similar relationship between the hypoxia response and the differentiation 327 state of different brain compartments in the Drosophila brain. A large part of the central 328 brain comprises a fully differentiated nervous system, which controls larval behaviour. In correlation with the observed differences in tracheolation, we found that cells in the 344 optic lobe show a stronger hypoxic response as compared to cells in the central brain. 345 The hypoxia biosensor provides a remarkable spatial resolution and it is possible to 346 detect a differential hypoxia response in cells that are in close proximity to each other. 347 We combined the ratiometric analysis with cell type specific markers and found that 348 central brain and optic lobe neuroblasts appear to be the most hypoxic cell types in the 349 developing larval brain. It suggests as documented also by other studies that lower 350 oxygen is a condition that favours more undifferentiated and multipotent cell types 351 be an indicator of cell potency (Cipolleschi et al., 1993). A cell exposed to low levels of 358 oxygen has higher multipotency and less fate commitment than a cell exposed to high 359 oxygen. A similar relationship could be in place in the Drosophila larval optic lobe, 360 where mitotic progenitor cells will be protected from oxidative damage by being 361 segregated within a more hypoxic microenvironment, which becomes gradually more 362 oxygenated as they advance in the differentiation process. In Drosophila a previous 363 study by Homem  values. This prompts the question to what degree cell type and cell intrinsic 376 mechanisms are responsible for a differential hypoxia response. Our study leads to the 377 interpretation that the oxygen-dependent degradation of HIF-1a within a given cell 378 depends mostly on its distance to tracheoles, but to a certain degree also on cell-type 379 specific features. Among several possible explanations, we regard as probable the 380 existence of cell-type specific differences in transcriptomes, metabolisms and the 381 general capacity of given cell types to degrade the GFP-ODD of the biosensor. 382 We found that while neuroblasts adapt their hypoxia response to elevated ambient 383 oxygen levels (hyperoxia), optic lobe neuroepithelial cells seem to have a more limited 384 capacity to respond. This might be due to differential gene expression of hypoxia 385 pathway genes in these two cell types. Indeed, by re-analysing cell type specific 386 Polymerase II DamID data (Southall et al., 2013) we found that the transcriptome of 387 neuroblasts, but not that of neuroepithelial cells, is highly enriched in hypoxia response 388

genes. 389
Normally hypoxia triggers an adaptive response, which among other things upregulates to use a shorter hypoxia treatment that was still long enough to observe a change in 450 brain oxygenation with the biosensor. 451 452

Light microscopy, image processing and analysis 453
For each brain, a single hemisphere was imaged using a 60x objective on a TCS Leica 454 SPE-II laser scanning confocal microscope or with a 60x objective on an Olympus 455 Fluoview FV300. Optical sections across the entire brain hemisphere were recorded at 456 0.6 µm intervals for tracheal surface measurement, at 1 µm intervals for the temporal 457 ratiometric analysis with the hypoxia biosensor and at 2 µm for cell type specific 458 ratiometric analysis with the hypoxia biosensor. The ratiometric anaylsis was performed 459 with Fiji as described in (Misra et al., 2017). The macro and plugin can be found on

3D trachea map annotation and proximity analysis to brain cells 470
A pixel-resolution map of the tracheal system in the whole brain hemisphere was 471 produced. To obtained this map, a stack of images containing the tracheal system were 472 processed by background subtraction and smoothening. Then, a threshold was applied 473 and the x,y,z coordinates of every pixel corresponding to every tracheole was stored to 474 produce a digital map. Our ratiometric analysis provides ratiometric values for every 475 nucleus n the brain hemisphere together with their x,y,z coordinates. Combining both 476 data sets we calculated the Euclidean minimal distance from every nucleus in the brain 477 hemisphere to the tracheal system. The analysis was performed in Python.

Prediction of Ratio values based on tracheole proximity 480
The relationship between 1/Ratio values and proximity to tracheoles was best modelled 481 by a decaying exponential function of the form: 482 483 where l is the length constant. l, 0 and are fitting parameters and 0 is a constant. 485 We used the exponential function resulting from the best fit to the data to predict the 486 value of hypoxia response according to tracheole distance. We used Fiji to represent 487 these values for every nucleus in the brain with a colour scale (Figure  4 shows a single 488 section of the resulting stack image map). Fits were preformed and analyzed with Igor 489 Pro (Wavemetrics). 490 491

Cell type specific analysis 492
To analyse the hypoxia biosensor in a cell type-specific manner ratiometric values were 493 measured only for a specific cell type at a time. For this, a segmentation mask was 494 generated for each major cell type in the brain using the fluorescent intensity signal The inner proliferation center (IPC) and outer proliferation center (OPC) were 500 segmented using the TrakEM2 software in Fiji following neuroanatomical borders 501 outlined by anti-Dlg staining. 502 503

Transmission electron microscopy 504
Brain samples for transmission electron microscopy were prepared from five wild-type 505 (Oregon R) 96 hrs ALH larvae, according to the protocol detailed in (Talamillo et al., 506 2008). The brain was oriented and trimmed to obtain slightly tilted frontal views of one 507 hemisphere. Ultrathin (50-60nm) sections were observed with a JEOL JEM 1010 508 electron microscope operated at 80kV. Several grids of each brain, each containing 509 several sections, were observed. Images were taken with a Hamamatsu C4742-95 510 camera and processed with AMT Advantage and Adobe Photoshop. 511 512

Statistical analysis 513
All data analyses and graphs were done using R/Bioconductor. Scripts for graphs can 514 be found here: https://github.com/MartinBaccinoCalace. Biological replicates (n) are 515 single brain lobes of different animals. For the optic lobe and central brain ratio values 516 boxplot charts were created in such a way that each boxplot contained the normalized 517 frequency values for seven replicates in a given bin. Student t-tests were performed 518 when assumptions for parametric test were accomplished (normality using Shapiro-519 Wilk test and homoscedasticity using Levene's test). If these assumptions were not 520 achieved, nonparametric Mann-Whitney U tests were performed instead. Statistical 521 significance was set at 0.05, 0.01 and 0.001. For tracheolation analysis Mann-Whitney 522 U tests were performed and statistical significance was set at 0.05. Power analysis was 523 conducted in RStudio to estimate minimum sample size for a power of 0.8 and a level 524 of 0.05 for a two-sided student t test or Mann-Whitney U test. 525 526