Trajectory analysis using monocle 3

To reconstruct lymph gland hematopoiesis in Drosophila, hematopoietic cells in the blood lineage were collected after filtering the PSC, DV, RG, and neuron subclusters (n = 19,143; Supplementary Table ⁴). We then followed the Monocle 3³⁷ analysis pipeline described in the website documentation (https://cole-trapnell-lab.github.io/monocle3/) using custom parameters predetermined from repetitive analyses. We normalized the dataset by log-transformation and size factor correction, following three covariates, sequencing library, UMI count, and mitochondrial gene contents (the proportion of mitochondrial genes in transcriptome), and scaled using the preprocess_cds() function with 75 PCs. UMAP dimension reduction (reduce_dimension) was performed with custom parameters umap.min_dist=0.4 and max_components=3, and clustering resolution (cluster_cells) was set to 0.001 which assigned all cells into a single partition. After graph learning was performed (learn_graph), the cells were ordered using order_cells() to set a node embedded in the PH1 subcluster as a start point. All the trajectory graphs were visualized using the plot_cells() function with or without a trajectory graph.

Monocle 3 offers several approaches for differential expression analyses using regression or graph-autocorrelation. In this study, we identified co-regulated genes along the pseudotime by graph-autocorrelation and modularized them. To detect co-regulated genes, the graph-autocorrelation function graph_test() was specified with a “principal_graph” parameter and significant genes were selected (q value < 0.05). Modularization was performed using find_gene_modules() with default parameters but only passing a list of resolution values from 10⁻⁶ to 0.1 for automatic parameter selection. A total of 51 gene modules were detected from the normal lymph gland trajectory, however, three modules were excluded in that they were unable to be characterized by the enrichment analysis with biological process gene ontology terms or KEGG pathways using g:Profiler (https://biit.cs.ut.ee/gprofiler/gost).

The trajectory analysis of infested lymph gland (Fig. 7) followed a similar pipeline as that of normal lymph gland with slightly different parameters. The dataset was normalized and corrected for covariates, sequencing library, UMI count, and mitochondrial gene contents in the same manner, though using 50 PCs. We then used 0.5 for the minimum distance in UMAP and 0.005 for the clustering resolution. Gene modules were also explored using a complete dataset and the result generally agreed with the previous gene modules using normal lymph glands. So, we focused on the LM trajectory in the analysis. To collect LM trajectory we cropped cells in 2-dimensional UMAP graph (1,058 cells satisfying UMAP 3 > 8 * UMAP 1 + 16). One GST-rich cell was discarded in the analysis. Modularization of co-regulated genes was performed as previously described using these cells.