Neuroblast-specific chromatin landscapes allow integration of spatial and temporal cues to generate neuronal diversity in Drosophila

During early neurogenesis in flies and mice, spatial and temporal cues interact to specify neuronal diversity, yet in no organism is it known how spatial and temporal cues are integrated. We used Targeted DamID (TaDa) to identify the genomic binding sites of the temporal transcription factor Hunchback in two adjacent Drosophila neuroblasts (NB5-6 and NB7-4). Hunchback targets were different in each neuroblast. Profiling chromatin accessibility showed that each neuroblast had a distinct chromatin landscape: Hunchback-bound loci in NB5-6 were in open chromatin, but the same loci in NB7-4 were in closed chromatin. Moreover, binding of the spatial factor Gsb/Pax3, essential for NB5-6 specification, was correlated with open chromatin and Hunchback-enriched loci in NB5-6, but not NB7-4. We propose early-acting spatial factors establish a unique chromatin landscape in each neuroblast, thereby restricting temporal factor binding to different loci in each neuroblast, resulting in different neurons in each neuroblast lineage. Impact statement Integration of spatial and temporal identity during Drosophila neurogenesis is due to spatial factors generating neuroblast-specific chromatin thereby biasing subsequent temporal transcription factor binding and producing neuroblast-specific neurons.

NBs, in the independent specification model, TTF binding will be identical in all neuroblasts 140 whereas in the sequential specification model, TTF binding will occur at different loci in each 141 neuroblast.

142
To discriminate between these models, we sought to determine Hb genomic targets in NB5-143 6 versus NB7-4. If independent specification is used, we expect to find similar Hb occupancy in 144 each neuroblast ( Figure 1B), whereas if sequential specification is used, we expect to find 145 different Hb genomic binding in each neuroblast ( Figure 1C). Our goal was to identify Hb 146 occupancy within the early NB5-6 and NB7-4 lineages during the Hb competence window, when 147 Hb retains the ability to generate ectopic early-born neuronal identities, and thus presumably can 148 still bind its normal genomic targets. To identify Hb occupancy in these two neuroblast lineages, 149 we adapted the previously described TaDa method (Marshall et al., 2016;Southall et al., 2013).  Figure 1E) or Dam can be fused to a transcription factor such as Hb, which 155 provides a read-out of Hb genomic occupancy ( Figure 1F). 156 Here we characterize two Gal4 lines that are specific for NB5-6 and NB7-4 lineages in the Here we characterize two Gal4 lines that label either the NB5-6 or the NB7-4 lineages, which 174 is a prerequisite for profiling neuroblast-specific Hb binding sites. NB5-6 forms in the Gsb 175 domain, whereas NB7-4 forms in the Engrailed domain ( Figure 2A). To label NB5-6 and its 176 lineage we used ladybird early (lbe)-Gal4, which is reported to specifically label NB5-6 and its 177 progeny (Baumgardt et al., 2009;Urbach and Technau, 2003). We confirmed that lbe-Gal4 178 expression was highly specific to the NB5-6 and its lineage from stage 10 through stage 12, the   , 1997). Therefore, we used NB5-6-Gal4 to generate MultiColorFlipOut (MCFO; Nern et al.,198 2015) single neuron labelling among NB5-6 progeny. We repeatedly (n=31) identified a Hb + 199 neuron that had a characteristic ipsilateral ascending projection, which we name the Chaise 200 Lounge neuron due to its distinctive morphology; two segmentally repeated Chaise Lounge 201 neurons are shown in Figure 2H; inset shows a Chaise Lounge neuron expressing Hb. We 202 searched the EM reconstruction (Ohyama et al., 2015) and identified an identical Chaise Lounge 203 neuron ( Figure 2I). Thus, NB5-6 makes a distinctive ipsilateral neuron during its Hb expression 204 6 window. Similarly, we used NB7-4-Gal4 to generate MCFO single cell labelling, but could not 205 directly identify a Hb+ neuron either due to loss of Hb from early-born neurons prior neuronal 206 differentiation, or due to lack of gal4 expression in these neurons. Instead, we used multiple 207 criteria to identify a putative early-born neuron, the G neuron, using MARCM clones ( Figure 2J), 208 and EM reconstruction ( Figure 2K). Our criteria for assigning this neuron as early-born include (i) 209 presence of the neuron in full NB7-4 clones ( Figure 2J) but not in the NB7-4-Gal4 pattern ( Figure   210 2, supplement 1), which misses early-born neurons; (ii) cell body position next to the neuropil, 211 where most Hb+ neurons are located (Kambadur et al., 1998)    The fact that Dam:Hb can induce early-born neuronal identity suggests that it can bind the 241 same genomic targets as Hb, but we wanted to determine this important point experimentally.     Table 1). In addition, there were 309 2,860 loci occupied by Dam:Hb in both neuroblast lineages (Supplemental Table 1). Importantly, 310 while the read densities at individual loci are similar between replicates, they are strikingly 311 different between the two neuroblast lineages.

312
Next we represented the differentially bound loci using a volcano plot, where the magenta 313 dots highlight the most significantly differential loci with more than 2-fold change and an FDR of    open chromatin peaks in that lineage. A Monte Carlo analysis showed these overlaps to be 383 highly significant, detecting 5.23% overlap with a set of random peaks in NB5-6 and 6.75% in 384 NB 7-4 (100 iterations, p-value <1 e -300 for NB 5-6 and 8.9 e -133 for NB 7-4, see methods). As a 385 control, we assayed loci bound by Dam:Hb in both neuroblast lineages and found that there was 386 no difference between lineages in open chromatin at these sites ( Figure 6C). We confirmed these 387 findings at the top five differentially bound Dam:Hb loci in the two neuroblast lineages. All but 388 two of these differentially bound loci were also identified in the differential chromatin analysis; 389 even the two that were not picked up in the analysis (sqz and mspo) were qualitatively different 390 between the two neuroblast lineages ( Figure 6D,E). We conclude that neuroblast-specific    Carlo analysis found this enrichment to be highly significant (average real NB5-6/NB7-4 fold 422 change = 2.219, average simulated NB5-6/NB7-4 fold change = 1.088, 100 random iterations, p-423 value = 5.1039 e -9 ; see methods). We conclude that loci differentially bound by Hb in NB5-6 are 424 enriched for Gsb occupancy, although we note that occupancy may occur at different times (Gsb 425 earlier, Hb later).

426
Taken together, these data support the sequential specification model, where a transiently 427 expressed STF (e.g. Gsb) sculpts a lineage-specific chromatin landscape in NB lineages (eg.     Dam:hb flies were crossed to about 3,000 Lbe-K-Gal4 or 19B03 [AD] /18F07 [DBD] flies. Embryos were 568 collected every two hours and aged for 7.5 hours at 25°C, and similarly treated until sufficient 569 material was collected -for each replicate, 4 X 1.5 µL tubes of 50 mg of control and experimental 570 embryos.

571
The TaDa experimental pipeline was followed according to (Marshall et al., 2016), with a few Quality control. Each file was assessed for quality using FastQC (Andrews, 2010). Reads with 589 quality score less than 30 were discarded. Any contaminants were removed using BBsplit of the 590 BBmap suite (https://sourceforge.net/projects/bbmap/ ).

592
The damidseq_pipeline was used to generate log2 ratio files (Dam:hb/Dam) in GATC resolution 593 as described previously (Marshall and Brand, 2015). Briefly, the pipeline uses Bowtie2 594 (Langmead and Salzberg, 2012) to align reads to dm6, the reads are extended to 300bp (or to the 595 closest GATC, whichever is first) and this .bam output is used to generate the ratio file    slop' was used to extend the 5-6 and 7-4 peaksets to 4kb (2kb on either side of the peak center).

639
An equal number of random peaks were generated for 5-6 and 7-4 as in the actual data 640 (respecting distribution of peaks on the chromosomes). 'bedtools shuffle' was used to generate 641 these random peaks. The Gsb data obtained from Florence Maschat was converted from wig to 642 bedgraph using 'wig2bed' from bedops, then dm3->dm6 using CrossMap, and finally from 643 bedgraph to bigwig using 'bedGraphToBigWig' from kentUtils (https://github.com/ENCODE-644 DCC/kentUtils). 'bigWigAverageOverBed' from kentUtils was used to generate the average Gsb 645 signal at each peak. The average signal for each iteration was generated using awk. The 646 difference in average Gsb signal between (randomly generated) NB5-6 and (randomly generated) 647 NB7-4 was calculated for a 100 such iterations. The difference between average Gsb signal for 648 the real data (i.e. 5-6 enriched Hb loci minus 7-4 enriched Hb loci) was similarly calculated. Z 649 scores and p-values were calculated based on these 100 simulations and real differences in Gsb 650 signal. A bash script was written to automate the above steps (available upon request). Similar  Figure 6, supplement 1. In all cases, signal files (of ChIP or TaDa data) were supplied as 656 bigwig files, and peaks regions were supplied as bed files. Figure 3F peak file was the narrow 657 peaks generated by MACS2 in the three Da-Gal4 Hb TaDa experiments; the Hb ChIP-seq ratio 658 file was used as the signal file (see under peak calling for details). Figure 3G peak files for Hb, 659 Bcd and Ftz were downloaded from BDTNP and were lifted-over from dm3->dm6 using 660 CrossMap; the Hb TaDa signal was converted to bigwig using 'bedGraphToBigWig' from kentUtils 661 (https://github.com/ENCODE-DCC/kentUtils). Figure 5B peak file was downloaded from BDTNP and 662 was lifted-over from dm2->dm6 using CrossMap; the Da-Gal4 CaTaDa signal was converted to 663 bigwig using 'bedGraphToBigWig' from kentUtils. Figure