Cell Systems
Editorial Genotype-Phenotype Mapping Meets Single Cell Biology As we look forward to 2017, Cell Systems expects that advances in technology will redefine an age-old problem: the mapping of phenotype onto genotype. Several landmark papers published in recent weeks describe experimental methods to address this problem—Perturb-seq (Dixit et al., Cell 167, 1853–1866; Adamson et al., Cell 167, 1867–1882), CRISP-seq (Jaitin et al., Cell 167, 1883–1896), and CREATE (Garst et al., Nat. Biotechnol. 35, 48–55). They illustrate the power of linking CRISPR-Cas9 guide RNAs to distinct molecular sequence identifiers incorporated into the resulting edited cells. Barcoding strategies have been used before, but the key advance of Perturb-seq and CRISP-seq involves performing tens to hundreds of thousands of single-cell RNA sequencing experiments on the edited cells. This provides a uniquely rich dataset of transcriptome-scale phenotypic measurements associated with designer genetic changes—a genotype-phenotype association writ large at single-cell resolution. Collectively, these papers provide concrete examples of broad trends discussed in this issue’s Commentary on the state of the field of systems genetics (pp. 7–15). For example, Aime´e Dudley and Joe Nadeau describe a grand challenge of ‘‘transitioning from description to experimentation.’’ We see this as an evolution from high-throughput measurement alone to high-throughput perturbation coupled to high-throughput measurement—exactly the capability demonstrated by Perturb-seq, CRISP-seq, CREATE, and other recent studies. Jef Boeke, Lars Steinmetz, Allan Jones, Marcus Noyes, and Sibylle Vonesch comment that such experimental approaches should allow researchers to move beyond catalogs of genetic variation to mechanistic studies of gene function. An example of such a study is highlighted in this month’s Cell Systems Call (pp. 3–6, ‘‘Genome-scale screening for functional lncRNAs’’). We believe that many signs point toward single cells as an emerging unit of experimentation. Several years ago during early days of single-cell analyses, it was reasonable to ask why single cells mattered, especially when bulk sequencing measurements yielded ‘‘superior’’ data in terms of coverage and noise. One may have assumed that only single-cell behaviors should be studied at the single-cell level. Perturbseq, CRISP-seq, and CREATE reveal that this is not so. A previously unsung attribute of single-cell-based technology is simply that it supports a level of parallelization, and therefore scale, that is more difficult to achieve when working with larger samples. On the basis of these advances in massively parallel experimentation for functional analysis in single cells, we predict several trends in 2017. Enter Cell Biology Perturb-seq, CRISP-seq, and CREATE can be viewed as a marriage of DNA synthesis capacity and genome-scale readout of molecular phenotypes using DNA sequencing. How-
ever, as our experts point out in the Commentary (pp. 7–15), phenotyping of cell biological features through automated, multiparametric imaging allows rich phenotypes above the molecular level to be measured. This is the domain of the cell biologist. Combining systems genetics techniques with quantitative cell biology will help create a genotype-phenotype map across the dimensions of space and time. Single-Cell Manipulation As single cells become a fundamental unit of experimentation, the development of new approaches to manipulate single cells will be a priority. For example, in Perturb-seq, single cells were processed via a commercial droplet-based platform, and in CRISP-seq, single cells were sorted into 384well plates using flow cytometry. These recently developed workflows allow single-cell RNA-seq to be performed at scale. Future advances in microscale instrumentation and parallelized cell imaging would support cell biological measurements, such as analysis of cell shape (Sero and Bakal, pp. 84–96). In addition, molecular strategies to perturb single cells in new ways (see ‘‘Stealthy control of proteins and cellular networks in live cells’’ and ‘‘Thinking fast and slow: synthetic posttranslational circuits for dynamical control of cell fate decisions’’ in the Cell Systems Call, pp. 3–6, for example) would also allow new questions to be asked. Rethinking Missing Heritability A defining feature of systems genetics is the notion that a better understanding of networks and higher-order features is needed to understand genotype-phenotype relationships (see the Commentary, pp. 7–15). Could the cell biological features identified by the marriage of cell biology, genomics, and synthetic biology represent manifestations of these network properties? We expect that they will. Reformulation of Computational Challenges The authors of Perturb-seq and CRISP-seq highlight analysis of their genotype-phenotype datasets as an open area of future research. Notably, both used various dimensionality reduction and linear modeling techniques to interpret their data. It is tempting to speculate that learning higher-order, non-linear features using deep neural networks or other methods may be helpful. In addition, it may be important to combine multiple machine learning techniques, a key feature of recent high-profile successes in artificial intelligence such as AlphaGo (Silver et al., Nature 529, 484– 489) and differential neural computing (Graves et al., Nature 538, 471–476). Despite the scale enabled by Perturb-seq, CRISPR-seq, and CREATE, these methods can still only sample a limited number of combinations of perturbations, so we see a need for computational approaches that address sampling issues head-on and rethink how to exploit structure in the problem and data.
Cell Systems 4, January 25, 2017 ª 2017 Published by Elsevier Inc. 1
Cell Systems
Editorial Altogether, recently published work suggests that cell biology will converge with genomics, DNA synthesis, computer science, and microscale instrumentation in 2017 and provide pillars for understanding how the genome is manifest within organismal biology.
2 Cell Systems 4, January 25, 2017
H. Craig Mak Editor, Cell Systems
Quincey Justman Scientific Editor, Cell Systems http://dx.doi.org/10.1016/j.cels.2017.01.008