Single-cell transcriptomics of dynamic cell behaviors

Despite advances in spatial transcriptomics, the molecular profiling of dynamic behaviors of cells in their native environment remains a major challenge. We present a method, termed behavioral transcriptomics, that allows us to couple physiological behaviors of single cells in an intact tissue to deep molecular profiling of individual cells. This method enabled us to establish a novel molecular signature for a striking migratory cellular behavior following tissue injury.

3c). 143 We wondered whether the identified phenotypes may be a common feature of injury-144 induced epithelial regeneration. We examined published data of an independent injury 145 model (Borthwick et al., 2001) and analyzed the prevalence of these signatures during 146 repair after polidocanol injury. As predicted, the M basal cell signature is strongly 147 enriched 24 hours post injury (hpi), declines at 48 hpi and 72 hpi, and returns to 148 baseline at 1 week after injury (all p < 10 -16 , Mann-Whitney U test, Figure 4a). Similarly, 149 the NM signature is decreased at 24 hpi when cell migration is presumed to be active, 150 increases at 48 hpi and 72 hpi when cell migration is presumed to be diminished, and 151 returns to baseline at 1-week post-injury when regeneration is complete (Figure 4a). The rapid progress in spatially resolved transcriptomics is enabling the discovery and 169 characterization of transcriptionally heterogenous cells in diverse tissue contexts (Lee et 170 al., 2021;Ståhl et al., 2016). However, these methods do not capture the dynamics of 171 cell behaviors that often define the unique biological processes that occur in the tissues.

172
To address this gap, we developed an approach to examine the association of 173 molecular and behavioral phenotypes of single cells in their native tissues. We first 174 established a respiratory organ explant culture that maintains tissue dynamics for an 175 extended length of time, and subsequently combined this platform with live imaging in order to observe distinct lung cellular behaviors at a broad time scale, spanning cell migration, cell division, and ciliary beating.
To link cell behavior to molecular analysis, we used photoconversion to mark cells that 179 display distinct cell behaviors for subsequent single-cell genomics analysis. We found 180 that a subpopulation of basal stem cells migrates within the lung during early 181 regeneration. We used recently developed single-cell RNA-sequence approaches to 182 establish molecular signatures for moving and nonmoving basal cells. Furthermore, we 183 found these distinct cell signatures across independent lung regeneration models, 184 suggesting that M and NM cell behaviors are likely not only conserved cellular features 185 of early epithelial regeneration, and but also that live imaging-guided single-cell profiling 186 approach can discover general principles of tissue biology. 192 193 mT-mG (stock no. 007676), nT-nG (stock no. 023035), CAGs-LSL-rtTA3 (stock no.

347
We restricted the expression matrix to the subsets of variable genes and high-quality 348 cells noted above, and values were log2-transformed, and then centered and scaled 349 before input to PCA, which was implemented using the R function 'prcomp' from the  To identify cell types within the data, unsupervised hierarchical clustering was used 360 using the 'Ward.D2' metric in the 'hclust' R package. Pearson's correlation was used as 361 a distance metric. This produced 3 clusters, 2 were clearly identifiable as Basal and 362 Club cells, based on disjoint expression of known markers Krt5 and Scgb1a1, 363 respectively, while the third was distinguished by much lower technical quality (an 364 average of 2373 genes detected per cell compared to 5193 for the Basal and 5480 for 365 the club clusters respectively, p=0.0004, Mann-Whitney U-test). These low-quality cells 366 were not used for DE testing.

368
To identify the signature of moving vs non-moving basal cells (Figure 3c) we ran 369 differential expression tests between cells in the Basal cluster between the two 370 conditions (moving and non-moving), and selected genes that were differentially 371 expressed (FDR<0.05). Differential expression tests were carried using a two part 372 'hurdle' model to control for both technical quality and mouse-to-mouse variation. This 373 was implemented using the R package MAST (Finak et al., 2015), and P values for 374 differential expression were computed using the likelihood-ratio test. Multiple hypothesis 375 testing correction was performed by controlling the false discovery rate using the R 376 function 'p.adjust'.

Re-analysis of polidocanol injury dataset 379 380
Previously published single-cell RNA sequencing data from mouse trachea injured   from a total of 5 mice at 4 different time points (mouse origin is color-coded). A two-way ANOVA was run to examine the effect of time post SO2 injury and different mice on the mean speed determined by PIV. There were 22 ROIs analyzed from 5 mice over 4 time-points. There was a significant interaction between time and the mean speed, F(2.219,42.91)=16.12, p <0.0001, but no significant difference between mouse and mean speed, F(4,17) =2.193, p=0.113. A Tukey post-hoc test revealed significant pairwise differences between 26 and 50 hr, 26 and 62 hr, 38 and 50 hr, as well as 38 and 62 hr. ** p<0.01. (D) Frequency distribution of injury-induced cell movements measured at 26-and 38-hours after injury identifies "mover" and "non-mover" regions. analysis for a rapid "mover" and a slow "non-mover" region. (D) Cell movement vectors 5 computed from PIV analysis were fit to a von Mises distribution to compute the circular variance.