Replication dynamics of recombination-dependent replication forks

Replication forks restarted by homologous recombination are error prone and replicate both strands semi-conservatively using Pol δ. Here, we use polymerase usage sequencing to visualize in vivo replication dynamics of HR-restarted forks at an S. pombe replication barrier, RTS1, and model replication by Monte Carlo simulation. We show that HR-restarted forks synthesise both strands with Pol δ for up to 30 kb without maturing to a δ/ε configuration and that Pol α is not used significantly on either strand, suggesting the lagging strand template remains as a gap that is filled in by Pol δ later. We further demonstrate that HR-restarted forks progress uninterrupted through a fork barrier that arrests canonical forks. Finally, by manipulating lagging strand resection during HR-restart by deleting pku70, we show that the leading strand initiates replication at the same position, signifying the stability of the 3′ single strand in the context of increased resection.


Supplementary Figure legends
Supplementary Figure 1 Polymerase usage following HR-restart at 4 different loci. Pu-seq traces at three loci on ChrI and one locus on ChrII where the RTS1 barrier has been integrated close to a strong origin in an early replicating region of the chromosome that boarders a late replicating region. RTS1 barrier activity on (rtf1 + ). The usage of Pol d (blue) and Pol e (red) are shown on the Watson and Crick strands. Note the switch from Pol e to Pol d at the RTS1 site that is indicative of a change in polymerase usage from Pol e to Pol d on the leading strand when RTS1 barrier activity is on and replication is restarted by HR.
Supplementary Figure 2 Calculating the delay to replication at rRFB barrier. Pu-seq traces of a locus on ChrI with and without the integration of 10xTer2-Ter3 sequence (rRFB barrier). a. Top Panel: Fork direction calculated from the Pu-seq traces (see 17 ). When the rRFB barrier construct is integrated, there is a decrease in the proportion of left-right moving forks downstream of the barrier. This is caused by a delay to left-right moving forks, and a concomitant increase in right-left moving forks. Bottom panel. Polymerase delta usage averaged across both strands is shown, demonstrating that (unlike for RTS1) forks only pause and are not restarted by HR. b. Selected outputs (4 -8 mins, 1 minute intervals) of the Monte Carlo model for the indicated delays to the left-right replication forks at rRFB are compared to the experimental data. C. Error between the experimental data and the model output for a delay between 0 and 8 minutes (15 second intervals) measured by the Euclidean distance between the two signals; orange curve is a smoothed representation of the errors using a Savitzky-Golay filter.
Supplementary Figure 3 Calculating the delay at RTS1 barrier. Using the replication model, different delay times of replication at the RTS1 barrier before HR-dependent restart (between 0 and 30 minutes, 1 minute steps) were fitted and compared to the experimental data. The best fit is an 11 minute delay at the barrier. a. Selected outputs from the model for usage of Pol d on Watson strand are compared to the experimental data. Note: both when replicated by left-right HR restarted forks, or when replicated by right-left converging canonical forks, the Watson strand is replicated by Pol d. b. Selected outputs from the model for usage of Pol d on Crick strand are compared to the experimental data. Note: A decrease in Pol d usage on the Crick strand reflects replication by Pol e from right-left converging forks. Thus, because an increased delay results in more replication by converging canonical forks, this is reflected by a decrease in Pol d usage. c. Error between the acquired data and the model output measured by the Euclidean distance between the two signals; orange curve is a smoothed representation of the errors using a Savitzky-Golay filter.
Supplementary Figure 4 Comparative polymerase usage following HR-restart between rnh201-d and rnh201-RED. Pu-seq traces of the RTS1-rRFB locus on ChrII. Top two traces: RTS1 barrier activity off (rts1-d). Bottom two traces: RTS1 barrier activity on (rtf1 + ). The usage of Pol d (blue), Pol e (red), and Pol a (green) are shown on the Watson and Crick strands in rnh201-d and rnh201-RED.
Supplementary Figure 5 Efficiency of the RTS1 barrier depends on the expression level of rtf1 + . a. The endogenous rtf1 + promoter results in 70% barrier efficiency. Left: Selected outputs of the model for different efficiencies of delay are compared to the experimental data. Right: Error between the experimental data and the model output (5% steps) measured by the Euclidean distance between the two signals. b. adh1-rtf1 constitutive overexpression results in 90% barrier efficiency. Left: Selected outputs of the model for different efficiencies of delay compared to the experimental data. Right: Error between the experimental data and the model output (5% steps) measured by the Euclidean distance between the two signals.
Supplementary Figure 6 Delay at second RTS1 barrier after HR restart at the first RTS1. Using the replication model, different delays for replication of HR-restarted forks at the second RTS1 barrier (between 0 and 30 minutes, 1 minute steps) were fitted to the experimental data. a. Selected outputs from the model for usage of Pol d on the Watson strand compared to the experimental data. Note: both when replicated by left-right HR restarted forks, or when replicated by right-left converging canonical forks, the Watson strand is replicated by Pol d. b. Selected outputs from the model for usage of Pol d on Crick strand compared to the experimental data. Note: A decrease in Pol d usage on the Crick strand reflects replication by Pol e from right-left converging forks. Thus, because an increased delay results in more replication by converging canonical forks, this is reflected by a decrease in Pol d usage. c. Error between the experimental data and the model output measured by the Euclidean distance between the two signals; orange curve is a smoothed representation of the errors using a Savitzky-Golay filter.

Supplementary Figure 7
Construction and testing of rnh201-RED. In S. cerevisiae it has been shown that the synthetic growth defect between the combined rnh201 and rnh1 deletions with deletion of TOP1 is alleviated by restoration of the poly-ribonuclease activity of Rnh201 (i.e. rnh201-RED). a. Alignment of S. cerevisiae and S. pombe Rnh201 with highlighted amino acid changes. b. Relative DNA fragmentation at incorporated rNTPs after alkaline treatment (for details see the methods section). c. Spot test: top1-d rnh1-d rnh201-d grow slower than the wild type control (rnh201 + ) and top1-d rnh1-d rnh201-RED.

Supplementary Figure 8
Delay at RTS1 before HR-restart in absence of Pku70. Using the replication model different delay times of replication at the RTS1 barrier before HR-dependent restart for the pku70-d background were fitted (30 minutes, 1 minute intervals) and compared to the experimental data. The best fit is a 14 minute delay. a. Selected outputs from the model for usage of Pol d on Watson strand compared to the experimental data. Error between the experimental data and the model output measured by the Euclidean distance between the two signals; orange curve is a smoothed representation of the errors using a Savitzky-Golay filter. Left panel: pku70-d rnh201-d. Right panel: pku70-d rnh201-RED.
Supplementary Figure 9 Position of replication restart in the absence of Pku70. Using the replication model, different start points of HR-dependent replication restart were fitted with 100 base steps downstream or upstream and compared to the experimental data. a. Selected outputs from the model for usage of Pol d on Watson strand compared to the experimental data. Note that a 100 bp change in the position of restart shifts the transition of Pol d usage. b. Selected outputs from the model for usage of Pol d on Crick strand compared to the experimental data. c. Error between the experimental data and the model output measured by the Euclidean distance between the two signals; orange curve is a smoothed representation of the errors using a Savitzky-Golay filter. For this plot the primary sequence data was reanalysed (Bowtie2) with a bin size of 100 and the bin sizes in the model adjusted accordingly. The experimental data becomes more noisy as bin size deceases, and thus the experimental data is smoothed with a sliding window of 5.