Cell specific eQTL analysis without sorting cells

Cell specific eQTL analysis without sorting cells

Harm-Jan Westra, Danny Arends, Tõnu Esko, Marjolein J. Peters, Claudia Schurmann, Katharina Schramm, Johannes Kettunen, Hanieh Yaghootkar, Benjamin Fairfax, Anand Kumar Andiappan, Yang Li, Jingyuan Fu, Juha Karjalainen, Mathieu Platteel, Marijn Visschedijk, Rinse Weersma, Silva Kasela, Lili Milani, Liina Tserel, Pärt Peterson, Eva Reinmaa, Albert Hofman, André G. Uitterlinden, Fernando Rivadeneira, Georg Homuth, Astrid Petersmann, Roberto Lorbeer, Holger Prokisch, Thomas Meitinger, Christian Herder, Michael Roden, Harald Grallert, Samuli Ripatti, Markus Perola, Adrew R. Wood, David Melzer, Luigi Ferrucci, Andrew B. Singleton, Dena G. Hernandez, Julian C. Knight, Rossella Melchiotti, Bernett Lee, Michael Poidinger, Francesca Zolezzi, Anis Larbi, De Yun Wang, Leonard H. van den Berg, Jan H. Veldink, Olaf Rotzschke, Seiko Makino, Timouthy Frayling, Veikko Salomaa, Konstantin Strauch, Uwe Völker, Joyce B.J. van Meurs, Andres Metspalu, Cisca Wijmenga, Ritsert C. Jansen, Lude Franke

Expression quantitative trait locus (eQTL) mapping on tissue, organ or whole organism data can detect associations that are generic across cell types. We describe a new method to focus upon specific cell types without first needing to sort cells. We applied the method to whole blood data from 5,683 samples and demonstrate that SNPs associated with Crohn’s disease preferentially affect gene expression within neutrophils.


2 thoughts on “Cell specific eQTL analysis without sorting cells

  1. This is a very nice paper. Thanks for posting the pre-print!

    I have a few questions/comments:
    1. On the classifier. Did you attempt any other approaches? Could you provide more details on how the cell type specific probes were selected? Were they simply the 58 most correlated probes? Were they selected in any way to maximize independent signals? Did you consider the NNLS approach used by Battle et al.?
    2. Limiting search to previously identified, additive cis-eQTLs. As the authors note, limiting the search for interaction-eQTLs (as opposed to searching the full set of cis-SNPs) almost certainly downwardly biases the fraction of interaction-eQTLs identified, because they previously selected for SNPs that explain a large proportion of the variance in an additive model. Of course, this effect will be even more dramatic for the eQTLs with larger effects in the minority cell population. Could the authors perform the analysis on the full set of SNPs on at least a subset of the data?
    3. Characteristics of interaction-eQTLs. The analysis of GWAS overlap is nice, but it would be great if the authors pushed this a bit further. Are these genes enriched for genes that are differentially expressed between cell types? Are the distributions of SNP to TSS distances different between the different classes? Are there cell type specific regulatory element (e.g. DHS?) datasets that can be used to interpret these findings?
    4. Cell type gene expression signature. Are there genetic variants that are associated with the proportion of neutrophils? When you include the cell type term in the model, do you increase the number of identified additive cis-eQTLs? (I.e., does heterogeneity add noise to the standard analysis?) Relative to uncorrected expression data? Relative to covariate-naïve latent variable removal?
    5. Examples. It would be nice to see a few plotted examples of particular gene-SNP combinations that have significant interaction eQTLs. In particular, it would be nice to see the data from one or both of the datasets where the cell type counts were directly quantified.

  2. Pingback: Most viewed on Haldane’s Sieve: February 2014 | Haldane's Sieve

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s