Tech Note: Simplified protocol for SMARTer Pico kit

Sample input, not fragmentation time, determines library similarity

In order to determine the effect of the fragmentation time and/or sample input on the data, the libraries from Dataset 1 were clustered. The dendrogram is generated from the normalized gene counts through edgeR. Euclidean distances between log2 normalized CPM values are then calculated and clustered. The clustering result shown in Figure 3 indicates that libraries from samples with 10 ng input are more similar to each other than to those with 1 ng. Furthermore, the fragmentation time of 2 or 4 mins has little impact on the library data when 10 ng input is used, as those samples fall within the same clades. Lower input leads to lower reproducibility, regardless of the fragmentation time. 

We assessed the percentages of reads aligned to the reference genome, as well as the percentage of reads assigned to features in the genome (Figure 4). We found no large differences among the datasets suggesting that regardless of the input and fragmentation time, the proportion of useful reads remains the same. 

Library insert size is affected by fragmentation time

To test if longer fragmentation times led to shorter library inserts, we estimated the insert size distribution based on the data from the inner distance module within the MultiQC report. The distance reported is the mRNA length between two paired reads. To this value, the length of the paired reads was added to get the insert size. Figure 5 shows the frequency of the insert size for the libraries from Dataset 1. The samples subjected to longer fragmentation times (red and green) have shorter inserts, on average. However, this is smaller than the difference in average size of the library obtained from the capillary electrophoresis (see Table 1); the difference there was as large as 500 bp compared to 80 bp in the sequencing data. This could be explained by the difference in clustering efficiency between short and long fragments, leading to an underrepresentation of the very long fragments in the sequencing data. Overall, the data shows that the fragmentation time does reduce the size of the library inserts.

Library complexity depends on sample input and is not affected by fragmentation time  

Library complexity is a measure of the number of unique molecules in the library. For any application, the goal is to make the libraries as complex as possible. This parameter is helpful in determining where sensitivity loss can occur (detection of low expressed genes) or where sequencing errors are created (duplicates). It is therefore important to determine if the fragmentation time affects library complexity and if so, to what degree. 

Library complexity was estimated using the preseq module from the MultiQC report. Figure 6 shows the rarefaction curves for Dataset 1 (blue, red and green), Dataset 2, the user sample (orange), and Dataset 3, from Takara (grey). One can see that the library complexity is more dependent on the sample input, and not so much on the fragmentation time.

rRNA depletion efficiency is independent of sample input and fragmentation time 

One of the key features of the SMARTer Pico kit is the depletion of rRNA in mammalian samples. The removal of the rRNA is dependent on probes that target these sequences and are then enzymatically degraded (Figure 1). According to the manufacturer, excessive fragmentation can lead to an inefficient depletion process, but we did not expect a difference in the depletion efficiency when fragmentation was skipped on good quality samples from Dataset 1 or Dataset 2. 

The depletion efficiency was estimated from the percentage of reads overlapping the genomic region spanned by the rRNA feature. The higher the percentage of reads mapped to rRNA, the lower the depletion efficiency. 

Figure 7 shows that neither the fragmentation time nor the sample input affect the depletion efficiency. On the other hand, sample quality seems to be crucial since the libraries from Dataset 3, generated from FFPE samples, show lower depletion efficiency than the high-quality samples from Dataset 1 or  Dataset 2.

Mitochondrial rRNA depletion efficiency is affected neither by sample input nor fragmentation time 

Another feature of the SMARTer Pico kit is the depletion of Human Mitochondrial rRNA (Mt rRNA). The removal of the Mt rRNA is also dependent on probes that target these sequences and are then enzymatically degraded (Figure 1). Similar to the analysis above, a higher percentage of reads mapping to Mt rRNA indicates lower depletion efficiency. Figure 8 shows that the depletion of the Mt rRNA is sample dependent (tissue or cell type) and it is neither clearly affected by the fragmentation time nor the sample input.

Gene body coverage is biased towards the 5´ end when fragmentation is not performed 

Gene Body Coverage calculates read coverage over gene bodies. This is used to check if read coverage is uniform and if there is any 5′ or 3′ bias. Figure 9 shows that samples from all datasets have very similar gene body coverage, except for the high-quality samples that were not subjected to fragmentation (blue). These samples show a bias to the 5´ end of the gene body but with a similar coverage on the rest of the gene body when compared to the rest of the samples. Other inputs or fragmentation times do not change the gene body coverage.

Conclusion

In this tech note, we aimed to determine the effects of different fragmentation times and input amounts on the data produced by the SMARTer Pico prep. Although we lack statistical power, this approach can guide us to the following conclusions: The shorter than recommended fragmentation times neither affect the depletion efficiency (Figures 7 and 8) nor the library complexity (Figure 6). Skipping the fragmentation resulted in longer libraries when examined in the capillary electrophoresis (Figure 2) but this is compensated for during sequencing, as long fragments are less likely to form clusters during sequencing (Figure 5). Skipping the fragmentation also affected the gene body coverage, where a bias to the 5´ end was observed (Figure 9) but this compromised neither the data quality (Figure 4), the complexity (Figure 6) nor the reproducibility (Figure 3). Additionally, using 16 PCR cycles seems to have little effect on the library complexity (Figure 6). Overall, we can see that sample input is the key to library complexity and reproducibility, while fragmentation time has less impact on data.