Figure 5. Mechanisms contributing to low noise and high information content of homeobox TFs.

(A) Example expression patterns of a LIM class homeobox TF (Lhx1) and a calcium binding protein (Calb2) with similar overall expression levels. Sample key as in Figure 1—figure supplement 1–3. (B) (upper) OFF state noise (defined as standard deviation (std) of samples with FPKM<1) plotted against maximum expression. (lower) HUGO gene groups enriched in the region indicated by red dashed box in the upper panel (see Figure 5—figure supplement 1 for PANTHER and Gene Ontology enrichments). (C) Average (replicate n = 2) ATAC-seq profiles for the genes shown in A. Some peaks are truncated. Expression levels are plotted at right (gray bars). (D) Length-normalized ATAC profile for genes with high ($>$ 0.3, blue dashed box in B, n = 853) and low ($<$ 0.2, red dashed box in B, n = 1643) OFF state expression noise. (E) Each circle represents the orthogonality of expression patterns calculated using HUGO gene groups. Orthogonality is a measure of the degree of non-redundancy in a set of expression patterns. Since the dispersion of orthogonality depends on family size, results are compared to orthogonality calculated from randomly sampled groups of genes (green solid lines: mean and Std. Dev.; green dashed lines: 99% confidence interval). Families, Z-scores, family size: 1. GPCR: 17.1, n = 277; 2. Homeoboxes: 16.6, n = 148; 3. Ion channels: 10.7, n = 275; 4. C2 domain containing: 7.8, n = 159; 5. Zinc fingers: 6.9, n = 1002; 6. Immunoglobulin superfamily domain containing: 6.7, n = 292; 7. PDZ domain containing: 6.3, n = 144; 8. Fibronectin type III domain containing: 5.9, n = 143; 9. Endogenous ligands: 5.1, n = 165; 10. Basic helix-loop-helix proteins: 4.9, n = 77.

Figure 5—figure supplement 1. Properties of Low OFF noise genes.

PANTHER (A) and Gene Ontology (GOM: Gene Ontology Molecular functions category) (B) enrichments for low OFF noise genes defined by red dashed region in Figure 5B. (C) Histogram of max expression for Low OFF noise genes and high OFF noise genes (genes in blue dashed region in Figure 5B). Low OFF genes have slightly higher max expression values than high OFF genes, p=0.002, Students’ t-test. Red and blue vertical lines indicate mean values (5.31 and 5.27, respectively). (D) Histogram of gene length for Low/high OFF genes. Low OFF noise genes are significantly shorter than high OFF noise genes, p=0 (below machine precision), Student’s t-test. Red/blue vertical lines indicate mean values (3.47 and 4.24 respectively). (E) Orthogonality, calculated as in Figure 5E, but using the PANTHER gene families.

Figure 5—figure supplement 2. Homeobox TFs form a combinatorial code.

(A) Heatmap showing expression patterns of 8 homeobox TFs that distinguished 99% of pairs. A minimal gene set algorithm (see Materials and methods) was used to select these TFs. Each GACP expressed an average of 4.1 ± 1.3 (Std. Dev.) of these TFs. (B) Combined DM (differentiation matrix, see Figures 3 and 4C) constructed by allowing GACP pairs to be distinguished on the basis of expression of any of 8 homeobox TFs in the minimal set (left) or by any homeobox TFs (right). White indicates distinguishable pairs and black indicates indistinguishable pairs. (C) Heatmap showing expression patterns of minimal gene sets for GPCR capable of distinguishing 99% of pairs.

Figure 5—figure supplement 3. OFF noise in single-cell datasets.

(Left column) OFF noise calculated as in Figure 5B from the standard deviation of cluster averages, plotted against the maximum expression. Red dots are homeobox transcription factors, black dots are all other genes. (Middle, Right columns). HUGO gene groups and PANTHER protein families over-represented in the dashed red boxes in the OFF noise plots. Datasets are from (Zeisel et al., 2015; Tasic et al., 2016; Zeisel et al., 2018; Saunders et al., 2018; Tasic et al., 2018).

Figure 5—figure supplement 4. OFF noise and gene length in Homeobox subfamilies.

(Left) Scatter plot of mean gene length and mean FCR for homeobox subfamilies. Subfamilies are as defined according to HUGO gene groups. (Right) Scatter plot of mean gene length and mean OFF noise for homeobox subfamilies.