Questions Regarding sRNA Clusters and Differential Expression

[JavierRamos01]

Hello Dr Axtell

I have become familiar with using Shorstack for identifying and quantifying sRNA, but I do have some questions I need to clarify.

1. I am confused about what a cluster is trying to represent. For miRNA in the results.gff output says miRNA_hairpin, while the counts and results.txt show it as cluster. For miRNAs is the cluster represents the precursor hairpin, and for proper terminology, for any publications, should we use the term precursor miRNA or mature miRNA?

2. Another thing I am not sure about is the clusters for siRNAs. How does Shorstack define where a cluster starts and ends? I know that siRNAs are around 21-24 nt in length, and in plants, a majority of siRNAs will be 24 nt long, which is shown by the dicercall. However, I wanted to know why the cluster is so large at around 400+ nt when siRNAs are much smaller. Is it because multiple siRNAs can derive from the same cluster, and if so, is that what the dicer min and dicer max columns represent, and how does it differ from miRNA clusters?

3. Lastly, for the differential expression of miRNAs and siRNAs using Deseq2, would it be better to use all of the raw counts in a replicant, both miRNAs, siRNAs, and unknown sRNA locus at the same time, or do differential expression of miRNAs and siRNAs separately and disregard the unknown sRNA locus? On a final note, when we are doing differential expression of sRNAs, are we doing it on a cluster or the mature/ majorRNA sequence that is shown in the results.txt?

[MikeAxtell]

Thanks for the questions

1. All small RNA loci that I know of (MIRNA loci, plant RDDM loci, plant secondary siRNA loci, animal piRNA loci, and non-AGO small RNAs like breakdown products from rRNAs) will present as "clusters" in sRNA-seq data. Multiple different sRNA sequences will align in close genomic proximity. For instance, MIRNA loci make mature microRNAs, but also microRNA stars, and usually lots of minor length and position variants. One core idea of ShortStack is the treat the cluster as the unit of expression, rather than individual small RNAs. The other core idea of ShortStack is that only by looking at cluster properties can one confidently distinguish the different types of loci. ShortStacks "clusters" are NOT necessarily corresponding to the exact start and stop of precursor RNAs. The term precursor RNA has a specific meaning (a MIRNA hairpin precursor after Drosha cleavage in animals) .. that is NOT what ShortStack reports. For MIRNA loci ShortStack instead tries to guess the entire hairpin plus some extra flanking sequence.
2. If you look at plant sRNA-seq data aligned to a genome on a genome browser you will see that 24nt / RDDM type of loci almost always appear in local clusters. It's not just one siRNA being made; a bunch are made in close proximity. For this class it's though that Pol IV fires at multiple closely linked positions as a general rule. ShortStack just uses some basic data smoothing to define cluster start and stops. The aggresiveness of the smooting can be controlled with the --pad option. The --dicermin and --dicermax options are global settings that are applied to every cluster that is found. >=80% of aligned reads must be in the size window set by these two options, or else the locus is flagged as type "N" or "other", reflecting that it is likely not a Dicer / Ago related small RNA cluster. These are usually bits of degraded, abundant RNAs.
3. For DE we use the raw counts from every cluster as input to standard programs like DESeq2 or edgeR. Similarly to regular mRNA-seq DE, we wouldn't want to exclude certain classes of loci because that may skew the analysis. Once you perform DE you find go back and categorize the results (for instance, to find the microRNAs or near microRNAs).

[JavierRamos01]

Thank you for answering my questions, Dr Axtell.

I just have one more thing I am unsure about. When performing differential expression of the siRNA clusters, is the whole cluster containing multiple siRNAs produced by RNA Pol IV? While the differential expression of the miRNA clusters would account for the miRNA hairpin and the extra flanking sequences?

Currently, I am a master's student working with plant sRNAs; many of these concepts were new to me before starting, and I just wanted some further clarification.

I much appreciate your insights.