rnabioco/aplysia repository overview

⁠RNA-Encoded Molecular Memory (REMM)

NSF Ideas Lab Collaborative Grant: RNA-Encoded Molecular Memory PIs: Aaron Johnson (CUAMC), David Glanzman (UCLA), Katie Berry (Mt. Holyoke), Saad Bahmla (GA Tech).

The PIs hypothesize that the expression or post-transcriptional state of one or more non-coding RNAs (ncRNAs) is selectively altered by sensitization training and that this molecular change mediates persistent sensitization memory in the model animal Aplysia californica. To test this hypothesis, the PIs will perform differential RNA-seq on purified RNA from trained and untrained animals and identify species of ncRNA with changes that correlate significantly with LTM. Initial experiments will determine whether the expression levels of specific ncRNAs change as a consequence of sensitization and identify candidate noncoding sequence that may be poised to change structure to mediate expression change. Based on these results, additional aspects of ncRNA structure, modification state or protein interactions will be probed to explore whether sensitization training induces persistent changes in these ncRNA properties that could mediate memory encoding in Aplysia. This work would set the stage for a future project in which Candidate ncRNAs could then be screened for mnemonic potency by disrupting them with antisense oligonucleotides (ASOs) and assessing the effects on neuronal excitability and synaptic connectivity of Aplysia sensory and motor neurons, as well as the effect of the ASOs on LTM in intact animals.

Done to Date: Collaborators at UCLA have recently done a series of experiments and performed RNA-Seq with their sea snail model (isolated neural ganglia cultures) to identify differential expression after serotonin treatment. They have sequenced (Illumina) and aligned (HISAT2) 18 samples that encompass three biological replicates of a +/- serotonin treatment of the dissected and cultured central nervous system, with three ganglia populations separated after serotonin for RNA-Seq prep. Two different animal sources were used, one that is bred for research (Miami “mari-cultured”) and one wild-caught (Cal). Average ~80% alignment with 20-40M total reads per sample.

RBI Fellows Request for Collaboration: Our collaborative group had recently discussed the Hani Goodarzi preprint on human riboswitches (attached) and realized we may be able to use the approach they developed to identify candidate 3’UTRs and noncoding RNAs that change abundance upon serotonin and have thermodynamic properties of a riboswitch. The scope of the work I’m hoping to get RBI Informatics Fellow help with initially would mainly be calling differentially-expressed genes and plugging the noncoding parts of those transcripts (5’ and 3’ UTRs and fully noncoding transcripts) into the Goodarzi SwitchFinder. There are decent annotations for transcripts and a solid genome for Aplysia, though my collaborators use two different animal populations for their experiments, including wild-caught animals. The biggest challenge may be calling UTRs that extend beyond the transcript annotation, but hopefully the data themselves will be able to guide any extensions needed. While this will cast a fairly wide net, using the known important genes that are affected upon the serotonin treatment, we aim to identify specific candidates to follow up with, including possibly some targeted structure analysis, which would be for further work.

⁠Project data description

• data: Raw (FASTQ) and processed (BAMs, salmon outputs) data files. See readme in data.

• dbases: Various resources (FASTA files, star indexes, etc) used for analyis

• docker: A docker container containing the necessary software for the pipeline was built. This container does not have the R and switchfinder software, only the pipeline dependencies. The docker image is hosted here (https://hub.docker.com/repository/docker/rnabioco/aplysia/general⁠) and can be run on bodhi as a singularity container.

• docs: various notes and ppts.

• hub: a UCSC trackhub with various data files for visualization. Currently hosted at http://amc-sandbox.ucdenver.edu/User33/remm/hub/⁠

• pipeline: A basic RNA-seq pipeline that runs STAR, featurecoutns, and salmon. The pipeline is written in snakemake and was run on bodhi.

• results: Individual analyses conducted for the project. See the readme for more details about the relevant directories.

• shared-results: results communicated to Aaron or collaborators.

⁠Methods

RNA-seq libraries were prepared from Aplysia californica ervous system ganglia. Paired-end libraries were sequenced on an Illumina XXX. Illumina Tru-seq sequencing adapters were removed using cutadapt (v4.5). Reads were aligned to the Aplysia californica genome (GCF_000002075.1) using STAR (2.7.11a). Genomic alignments were used only for visualization. Transcript abundance for was estimated used salmon (v1.10.2) either using the Aplysia californica transcriptome (GCF_000002075.1_AplCal3.0.ncbiRefSeq.gtf), or a custom transcriptome supplemented with non-coding transcripts, and additional transcripts from the Orvis et al paper.

Gene-level differential expression was conducted using tximport to summarize transcript counts to gene-level, followed by differential expression testing using DESeq2. For the custom transcriptome custom pre-filtering of the gene counts was performed to remove genes poorly represented in the dataset. Genes with a high level of counts from few or signle sampels were excluded, by removing genes with a maximum count accounting for greater than or equal to a proportion of 0.4 of the total counts across all samples were excluded.

In R code: rmax <- rowMax(norm_counts) / rowSums(norm_counts) genes_to_keep <- which(rmax < 0.4))

This filter was added after observing spurious differential expression due to high read counts in few samples, potentially being due to inter-sample genetic variation leading to biased transcript quantification.

To construct the custom transcriptome, transcripts from the Orvis et al study (https://sra-download.ncbi.nlm.nih.gov/traces/wgs04/wgs_aux/GJ/YY/GJYY01/GJYY01.1.fsa_nt.gz⁠), and non-coding transcripts supplied from the Abrams lab (transrate.pasa.contigs.filtered.short.fasta and transrate.trinity-dn.contigs.filtered.short.fasta) were aligned to the AplCal3.0 genome. A gtf file was then made from the BAM, which was compared against the NCBI gtf using gtfcompare (v0.12.9). Transcripts with class codes i (intronic) and u (no match to reference) from gffcompare were appended to the NCBI/Broad annotation. Transcripts with the same splice sites as the reference were added only if the novel transcripts increased UTR lengths. Classes j, k, n, m, and y were added as new transcripts for the matching reference gene. Transcripts from other classes were ignored. The output of this filtering was the "aplCal_with_nc_orvis_et_al.fa" file.

Riboswitches were predicted on non-coding RNA sequences (ncRNAs, 5'UTR and 3'UTR) using the switchfinder software from https://github.com/goodarzilab/SwitchFinder⁠ . Shuffled sequences were included as a control which were generated using dinucleotide shuffling of the sequences, as implemented by scripts in the meme software suite. Known riboswitch sequences, taken from the SwitchFinder repositiory (https://raw.githubusercontent.com/goodarzilab/SwitchFinder/main/example_data/seed_riboswitches.fa⁠), were also included.