Fig 1.
Pulse labeling experiment types to measure degradation rates.
The conventional approach as in [18] utilizes biochemical separation, which does not preserve the fraction ratio (labeled vs. unlabeled) in the read counts. Alternative novel approaches (e.g. [12]) induce reverse transcription signature events (nucleotide conversions, typically T-to-C). Individual reads can be classified by the presence or absence of this characteristic nucleotide conversions. In an ideal case, the fraction ratio is well reflected by the read counts, however in practice a relatively low 4sU incorporation rate of 1:40 has to be taken into account ([12], [9]).
Fig 2.
The key characteristics of metabolic RNA labeling experiments.
A: The diagonal term of the Fisher information matrix (FIM) , as a function of the ratio of labeling time t to the characteristic time of degradation τ = 1/δ for the case of SLAMseq experiment. Read counts follow the Poisson distribution, the expression level is μ = 1 and the degradation rate is δ = 1. B: 95% confidence interval (CI) relative width of the degradation rates for different sets of time points included in the simulation of the SLAMseq experiment. We simulated counts for a range of rates δ and assumed for simplicity that normalization factors are perfectly known but not the rates and expression levels. Smoothed data from 10 simulation runs is shown. C: Relative standard deviation (
) of the MLE for δ as a function of measurement time at different values of the overdispersion parameter k. With increasing overdispersion, the profile of the dependency flattens. However, near the optimal time point, variance of the estimation is more sensitive to time of labeling, which complicates the optimal design choice for different δ ranges. Expression level is fixed to μ = 100 reads in this example, the degradation rate is assumed to be δ = 1. The FIM
is calculated for n = 1. D: Relative standard deviation (
) for a model with overdispersion (k = 100, solid line) or with no overdispersion (k → ∞, dashed line). The degradation rate is δ = 1, the labeling time is t = 1. The FIM
is calculated for n = 1.
Fig 3.
Estimates for pulse-chase SLAM-seq data [12].
Degradation rates and 95% confidence intervals are shown for different chase time points. For short chase times, the majority of genes have poorly identified degradation rates (see subsets [0, 0.5], [0, 3], [0, 6] hr). On the other hand, longer chase times do not allow to precisely estimate rates for unstable genes ([0, 12] hr).
Fig 4.
Application to the SLAMseq experiment.
A: Diagonal term of the FIM as a function of chase time. Similar to Fig 2, we normalize it as
, so it corresponds to the lower boundary of the relative variance
. Using time points with low
values results in higher variance of
. In this example, as values of
and
, we use medians of their estimations from the model fitted to the full set of points. B: Relative width of 95% confidence intervals (CI) for the rate estimations
. We use the genes with
located between 40%-60% percentiles (i.e. near the median). Genes, which have ratio close to the optimum t/τ ≈ 2.9 (subfigure (A)), have smaller relative CI for
.
Fig 5.
Purification of labeled and unlabeled RNA fractions.
MCF-7 cells were pulse labeled with 4sU for up to eight hr as indicated. Total RNA was spiked with in vitro transcribed 4sU-labeled FLuc and unlabeled RLuc, biotinylated with MTSEA-biotin and subjected to streptavidin purification. (n = 3). A: Dot blot-based detection of biotinylation with streptavidin-HRP in input and flow through of streptavidin purification. B: The amount of RNA enriched by the streptavidin purification was determined by absorption measurement. C: In vitro transcribed spike in RNAs 4sU-labeled FLuc and unlabeled RLuc in the flow through and biotin-enriched fraction were measured by RT-qPCR analysis and normalized to a standard curve given in S3 Fig.
Fig 6.
Application to experimental data from the MCF-7 pulse labeling time course experiment.
A: We plot the diagonal term of the FIM computated at estimated parameter values and multiplied by ,
, to illustrate contributions from labeled and unlabeled fractions to estimations of degradation rates for different experimental points (MCF-7 experiment, 2, 4, and 8 hr) and fractions (labeled and unlabeled). The black lines are the limiting values for the
according to Eq 20. B: The modified FIM term
is computed for a range of labeling times for one of the fastest (at the 0.1% quantile) and one of the slowest (at the 99,9% quantile) genes (δfast = 0.79hr−1, δslow = 0.019hr−1). The normalization coefficient for the labeled and unlabeled fractions is adjusted in such a way that their sequencing depth (total mean read count) at time t equals the sequencing depth of the total sample.