Comprehensive catalog of dendritically localized …

Comprehensive catalog of dendritically localized mRNA isoforms from sub4 cellular sequencing of single mouse neurons

“Comprehensive catalog of dendritically localized mRNA isoforms from sub-cellular sequencing of single mouse neurons” Sarah A. Middleton, James Eberwine and Junhyong Kim

https://doi.org/10.1101/278648

I selected this article for review for the following reasons:

  • In neurons, RNA localization is a fundamental way to control protein expression. To my knowledge this is the first attempt to map dendritically localized RNAs at single cell resolution
  • The findings reported are very intriguing, in particular the potential role of SINE elements in localization of RNA to dendrites.

Both reviewers were enthusiastic about the application of a state-of-art technology such as single-cell RNA seq to answer a question that has not yet found a consistent answer: which RNAs are transported to the dendrites in neurons. Concerns were raised regarding methodological issues that may confound the interpretation of the results, for example technical challenges with extractions of RNA from dendrite versus some and issues with data normalization and quality. Moreover, the fact that these studies were conducted at a specific developmental stage in culture may limit how generalizable the conclusions are in vivo. Nonetheless the authors present a valuable resource that can guide future studies.

I want to thank the authors for sharing their work on bioRxiv before it has undergone peer review. I also want to take an opportunity to thank the reviewers who have donated their time to evaluate the novelty, strengths, and limitations of this work. One reviewer chose to remain anonymous, and one chose to be named. Both reviewers were faculty. The two reviews are available in full below.


Reviewer 1 (Davide Risso)

Middleton et al. present a comprehensive catalog of dendritically localized mRNAs by performing single-cell RNA-seq of 16 culturally derived hippocampal neurons. For each neuron, the authors are able to extract RNA from both the soma and the dendrites, resulting in a paired design in which for each neuron they have access to both RNA measurements.

The paper is an exciting application of state-of-the-art technologies that allow the measurement of RNA at a sub-cellular resolution. The resulting set of dendritically localized genes and isoforms are bioinformatically analyzed and some promising patterns emerge, including the potential role of SINE retrotransposons in dendritic localization.

The authors extract three lists of genes from the RNA-seq data: what they call “deDend” (derived from a differential expression analysis), “consDend” (expressed in most dendrites) and “isoDend” (derived from a differential 3’ UTR analysis). It is unclear how these three sets relate to each other from a biological perspective, but the authors state that the three lists are largely non-overlapping.

Lastly, the authors combine their list of dendritic RNAs with a meta-analysis list derived from six publications that use microarray or RNA-seq to profile dendritic transcriptomes. They found little overlap at the gene level between the seven studies, but recurrent themes emerge, such as ribosomal proteins.

Overall, I enjoyed the paper, but there are a few points that I would like the authors to address that may lead to an improved analysis.

  1. While it is obvious how to interpret the results from the differential expression (DE) and differential isoform analysis, I’m having troubles understanding the value of the set of genes expressed in 90% of the soma. Given the technical challenges of single-cell RNA (including the presence of drop-outs), it looks dangerous to me to focus on a set of genes by just looking at their presence/absence pattern. What if there are additional genes that are expressed in dendrites and are lost because of drop-outs or transcriptional bursting? Conversely, what if there are different sub-population of neurons that express different RNAs in the dendrites? In addition, RNA-seq is only able to provide relative expression measures as GC-content, transcript length and other gene-specific features can potentially bias the list of expressed genes that the authors work with. That is why using the soma as a comparison is a good idea, and why I trust the DE results better – which incidentally is something that can be done precisely because the authors measured RNA in both soma and dendrite, which I think is a very good aspect of the dataset.
  2. Data normalization is a very important issue in the analysis of single-cell RNA-seq. Especially so when one is comparing two groups that do not express the same number of genes, as is the case here: in fact soma express twice as many genes as dendrites. How will that affect the DESeq2 analysis, given that the DESeq2 normalization method assumes that the majority of the genes is not DE? The authors included a set of ERCC spike-ins in the protocol, but they do not use them in the analysis: although the behavior of ERCC spike-ins for RNA-seq is suboptimal, the authors could at least use them for quality control, to insure that normalizing the data does not create obvious problems.
  3. More generally, exploratory data analysis to show that the quality of the data is good is missing and the authors should consider adding it since this is a relatively new protocol which might lead to noisy data. For instance, would an unsupervised clustering of the data lead to dendrites vs. soma or does any other aspect of the data (cultures, days, etc.) explain the data better? If so, the authors might consider adding those as covariates in the DE analysis.
  4. The identification of highly variable genes seems problematic: the authors state that the high variability genes had lower median expression than the low variability genes, but that is exactly what one expects from the Negative Binomial distribution, for which it has been shown many times that low count genes tend to show more (technical) variability than high count genes. Many method exists for the accurate identification of highly variable genes in single-cell RNA-seq, some of which use the ERCC spike-ins as anchor points (see for instance [1] and [2]). Furthermore, in the isoform analysis the authors state that “61.1% of the isoDend genes had a more variable DF in the soma than in the dendrites”, but it is unclear how this variance was computed.
References:
  1. Lun ATL, McCarthy DJ and Marioni JC. A step-by-step workflow for low-level analysis of single-cell RNA-seq data with Bioconductor [version 2; referees: 3 approved, 2 approved with reservations]. F1000Research 2016, 5:2122 (doi: 10.12688/f1000research.9501.2)
  2. Buettner F, Natarajan KN, Casale FP, et al.: Computational analysis of cell-to-cell heterogeneity in single-cell RNA-sequencing data reveals hidden subpopulations of cells. Nat Biotechnol. 2015; 33(2): 155–160

Reviewer 2 (anonymous)

Previous studies have definitively shown that some RNAs can be transported and localized in dendrites of neurons. However, recent studies have found little consensus on which RNAs and how many are localized to dendrites. In this paper, the authors set out to characterize the full repertoire of dendritically localized RNAs and whether there were any clear targeting motifs. The authors used single cell deep RNA sequencing from cultured hippocampal neurons and ran multiple types of analyses to determine the composition and variability of RNAs in this data set. Interestingly, one of the novel enriched sequences in dendritically localized RNAs are retrotransposon SINES.

While this paper provides further characterization of putative dendritically localized RNAs, it still suffers from similar drawbacks of previous studies in the purity of the dissociation between soma and dendrites. There are also significant limitations in taking a single cell approach of cultured cells at one particular age during development, which limits the conclusions. However, the paper does provide a rich data set for follow up studies that could validate or corroborate specific targeting elements and enriched RNAs.

It is disconcerting for the field that the author’s meta-analysis of 6 different studies showed that not one RNA was identified in all 6 studies to be dendritically localized.

Major concerns

  1. As with most of these types of studies, it’s hard to control for a clean dendrite vs. soma dissociation. The fact that many housekeeping genes were found in both compartments suggests that this preparation is potentially not well dissociated nor is there protein expression data (e.g. MAP2 levels) to validate the dissociation.
  2. Only 16 neurons were used for deep sequencing. Obviously, the cost of sequencing limits the numbers of cells that can be characterized, but there is no information on how the neurons were picked for inclusion. What markers were used to distinguish excitatory vs. inhibitory cells for example? CA1 vs. DG neurons? The 16 neurons may not be at all homogenous. Primary cultured neurons are notoriously variable across different cultures. It is not clear how many different cultures (i.e. dissections) were used to provide the 16 neurons. The authors should determine the variability across cultures vs. within a culture.
  3. The authors based their dendritically localized RNAs solely on having an arbitrary higher enrichment than found in the soma. As the authors point out, this would miss potentially relevant RNAs that are highly expressed in the soma and then are transported out into the dendrites. Thus, the temporal dynamics of transport are missed with this paradigm but it is not clear how this is solved by identifying the most consistently localized RNAs across neurons, given point 2 above that would add considerable variability to the data set. Thus, a lot of false positives may result from this analysis.

Minor concerns

  1. It’s not clear that the role of dendritically localized RNAs is solely for synaptic potentiation nor is it generally accepted that synaptic potentiation requires dendritically localized RNAs.
  2. The supplemental tables and analyses were not included in the preprint PDF or linked online.

Lucia Peixoto is an Assistant Professor at the College of Medicine and a member of the Sleep and Performance Research center at WSU Spokane. She is working to understand the genetic links to autism and related symptoms, such as learning disabilities and disturbed sleep.