SPRAWL peripheral and central score workflow.

a) RNAs are ranked from closest to furthest from the cell-boundary to calculate the median peripheral rank of the gene of interest. For the central metric, distances from the cell centroid are used for ranking instead. b) Under the null hypothesis of each rank being equally likely, the probability mass function of the median is exactly calculable. c) The intuitive SPRAWL score per gene per cell, X, will be near +1 for highly-peripheral patterns, near 0 for randomly-peripheral patterns, and near -1 for anti-peripheral patterns. d) Peripheral significance of a gene within a cell-type is estimated from per cell SPRAWL scores using the Lyapunov Central Limit Theorem (CLT).

SPRAWL punctate and radial scores workflow.

a) The SPRAWL punctate metric relies on b) permutation testing to create a score c) that represents whether RNA molecules from the gene of interest are closer together than expected by chance. The radial metric is identically calculated, except using average angle instead of distance. The significance of gene-celltype punctate patterns is calculated using the Lyapunov CLT as in the peripheral metric. b) Depictions and interpretation of the SPRAWL punctate metric.

SPRAWL gene-celltype scores are highly correlated between biological replicates.

a) BICCN MERFISH, Vizgen Brainmap, and Vizgen Liver biological replicates (rows top to bottom) have Pearson correlation coefficients (blue) larger than 0.47 for SPRAWL peripheral, radial, punctate, and central metrics (columns left to right). Randomly permuting gene labels in these datasets eliminates underlying spatial patterning and yields insignificant Pearson correlation coefficients (orange) between biological replicates. Dotted lines indicate zero-valued SPRAWL gene-celltype scores. b) In the MOp BICCN dataset 87% of gene/cell-type pairs have positive punctate RNA patterning (blue), compared to 50% in the gene-label permuted data (orange). Similarly extreme trends of 95% and 52% are observed for the radial metric. Cldn5 RNA is consistently highly punctate and radial in all cell-types that express it, depicted by purple x-axis ticks.

SPRAWL spatial scores and 3’ UTR length are significantly correlated for a subset of genes.

a) Workflow to calculate median 3’ UTR length and spatial score per gene/cell-type. b) Slc32a1 median centrality, c) Cxcl14 radial, and d) Nxph1 punctate SPRAWL scores from the BICCN MERFISH dataset correlate significantly with 3’ UTR length determined from 10X scRNAseq data by ReadZS. The left-column boxplots show individual SPRAWL cell scores as overlaid dots. The cell-types are sorted by increasing median score marked in red. The two cell-types with the highest and lowest median SPRAWL scores are plotted individually while the remaining cell-types are collapsed into the “Other” category. Gene/cell examples are shown to the left the boxplots for each extreme cell-type group. The density plots in the middle column show estimated 3’ UTR lengths for each read mapping within the annotated 3’ UTR, stratified by cell-type. Lengths were approximated as the distance between the annotated start of the 3’ UTR and the median read-mapping position. Each density plot is normalized by cell-type to show relative shifts in 3’ UTR length with median lengths depicted with red lines. The scatterplots show the significant correlations between median SPRAWL score and median 3’ UTR length. The two cell-types with the highest, and the two with the lowest SPRAWL median scores are highlighted.

Timp3 alternative peripheral localization across MOp cell-types is statistically correlated with ReadZs differences in 3’ UTR length.

a) ReadZs detects two major alternative 3’ UTRs in mouse Timp3 from 10X scRNAseq which correspond to miR-181c-5p and miR-221-3p binding sites. Reads from L6 CT cells predominantly map to a novel upstream shortened 3’ UTR while endothelial cells primarily express the longer annotated 3’ UTR. The UCSC genome browser placental animal sequence conservation shows highly conserved regions in blue. Fisher’s exact test was highly significant between the two peaks denoted by the dotted lines between the two cell-types. b) Timp3 mean periphery score is significantly correlated with Timp3 median ReadZs score across MOp cell-types with Pearson correlation coefficient of -0.91 and p << 0.05. c) Fraction of Timp3 RNA full-length 3’ UTR reads, gray box and d) barplots, decreases during human lung tissue culture.

Shorter Timp3 3’ UTRs become relatively more abundant in pericyte cell culture while Timp3 protein production remains stable.

a) Experimental setup for human pericyte cell culture with reverse-transcriptase quantitative PCR (RT-qPCR) and extracellular Timp3 protein ELISA readouts at four time points. b) Timp3 protein secretion per cell per hour does not significantly change throughout culture time, even though the total protein measured by BCA does change. c) qPCR experiment design with proximal and distal qPCR primers to distinguish long and short 3’ UTR isoforms. The proximal qPCR primer can detect both long and short isoforms while the distal primer can only amplify the long 3’ UTR. d) The ratio of distal to proximal primer template abundances significantly decreases throughout culture time, implying increased usage of the short Timp3 3’ UTR compared to the long isoform. e) Timp3 3’ UTR abundance, normalized by 18s housekeeper abundance, fluctuates from halving to doubling between culture timepoints for both distal and proximal primers.

Counts of unique, significant, and opposite-effect genes in each experiment/metric combination. Genes are defined as significant if they are observed to be significant in at least cell-type in any replicate. Opposite-effect genes are those observed to have at least one significantly positive SPRAWL gene/cell-type score, and one significantly negative SPRAWL gene/cell-type score.

SPRAWL metrics have high specificity and lack bias. a) SPRAWL scores for permuted null datasets, reds, have expected mean values of zero regardless of either the number of cells per cell-type or the gene abundance. The permuted datasets have expectedly lower variance for higher cells per cell-type and gene abundance. The real data, blue, shows expected means near 0 for the central and peripheral metrics, but higher scores for the punctate and radial metrics. b) Under null simulations, red lines, all gene/cell-type pairs are deemed insignificant at an alpha level of 0.05 (vertical dashed line) for the four metrics. In the real data, blue lines, more gene/cell-type pairs are significant, after Benjamini-Hochberg correction, with higher cell-type and RNA abundance. c) The fraction of significant gene/cell-type pairs in the BICCN samples are consistent across abundance levels measured as gene/cell-type median spot counts. d) Peripheral and central scores are strongly anti-correlated for gene/cell-type scores while the radial and punctate scores are positively correlated. e) To test whether peripheral localization patterns were driven artifactually by incorrect cell boundary calling, the cell boundary locations were computationally shrunk by a factor of 0.8 in the x and y direction, discarding spots that fell outside the new boundaries. In both the BICCN MOp and Vizgen Brainmap datasets, a Pearson correlation coefficient of greater than 0.85 was observed between the shrunk and original median gene/cell-type periphery scores. f) SPRAWL scores are not conflated with cell size g) Similar fractions of gene/cell-types are significant between the different datasets and metrics.

Vizgen Liver Showcase scores are highly correlated between replicates. a) The Vizgen Liver showcase dataset provides spatial information for 2 mouse livers with 2 slices each. Cell annotation data was not provided in the Vizgen Liver Showcase, instead clusters produced from off-the-shelf Leiden clustering (python scanpy package) were used as pseudo cell-types. All four datasets were combined without reference to biological or technical replicate by first normalizing the read counts per cell, identifying highly variable genes, reducing to the first 10 principle components (b) and then computing a neighbor graph with n = 40 which resulted in 100 clusters which had similar number of cells from each animal c). As well as having high Pearson correlation coefficient between mice (d), the technical replicates were highly correlated within both Liver 1 (e) and Liver 2 (f).

ReadZS detects Tabula Sapiens Lung differential 3’ UTR length Timp3 and decrease in Timp3 expression throughout culture a) ReadZS detects statistically significant 3’ UTR length differences in the human Timp3 3’ UTR across endothelial cell-types from the Tabula Sapiens consortium datasets. The eCDF plot below the gene annotation shows slight variation in read buildup over all cell-types below which individually show read-density. HuR binding sites from PAR-CLIP are shown above the Timp3 gene structure diagram. The last track shows high vertebrate sequence conservation throughout the UTR. b) Normalized expression of Timp3 against Actin and Gapdh show decreasing expression of Timp3 throughout increasing culture direction in all tissue compartments.

Computationally predicted miRNA binding sites in the 3’ UTRs of Slc32a1, Cxcl4, and Nxph1 and additional 3’ UTR correlated genes a) Subset of computationally predicted 3’ UTRs from the miRWalk database for Asic4, Slc32a1, and Nr2f2 indicate a potential mechanism of regulation for 3’ UTRs of different lengths. b) Three genes Ubash3b, Igfbp4, and Wipf3 show significant negative correlation between various SPRAWL metrics and estimated 3’ UTR length.

SPRAWL scores do not correlate with presence of signal recognition peptide, but do correlate with nuclear enrichment: a) Genes encoding signal recognition peptides do not have significantly differential SPRAWL scores while b) genes such as Wipf3 and Slc30a3 have significantly lower peripheral scores in cell-types with higher nuclear expression. Satb2 shows the opposite unexpected correlation.

qPCR primer efficiencies for Timp3 3’ UTR were estimated by using 2-fold cDNA dilutions of the same 6-hour timepoint sample. Of the two proximal and two distal primer pairs, only proximal primer 1 had low efficiency at 81.9%. The remaining three primers showed nearly perfect 100% efficiency, where a cDNA dilution of 2X resulted in a critical threshold value increase of 1. Dots indicate CT readings of technical replicates done in triplicate with shaded regions between them. Dashed lines indicate 100% efficiency curves.