There’s more to come in this space, but I am thrilled to see this work finally published. Huge thanks to first authors Kaden Southard and Rico Ardy, and co-authors Anran Tang, Deirdre O'Sullivan, Eli Metzner, and Karthik Guruvayurappan.

We posit that this state antagonism can potentially be exploited therapeutically to ablate the disease-associated inflammatory state. More broadly, as perturbation atlases grow there may be an opportunity to map a regulatory graph of states defined by antagonistic interactions.

Inflammatory fibroblasts secrete collagen. This enabled a striking experiment visualizing “state antagonism.” The pro-inflammatory TF EGR3 increases collagen expression, while the pro-universal TF KLF4 decreases it. When both are activated together, the effects cancel out.

Takehome 4: There is a regulatory logic underlying transcriptional states. We noticed that TFs that promoted the universal fibroblast state often appeared to be repressors of the inflammatory state. What then happens if we try to drive cells into both states at the same time?

These comparisons let us identify TFs driving four fibroblast states described in the literature: universal, inflammatory, myofibroblast, and antigen presentation. Our in vitro signatures flag these subpopulations across four independent fibroblast atlases in a new main figure.

During revision, several new fibroblast cell atlases were released, allowing us to compare our perturbation results to fibroblasts observed in diverse tissues and disease contexts.

Takehome 3: Cell atlases make single-cell functional genomics data interpretable. Our goal in doing all these perturbations was to see if we could push primary fibroblasts into the transcriptional states they exhibit in vivo, some of which are linked to function or disease.

A second challenge we encountered was surprisingly strong off-target effects from dCas9 binding at 3-5 nt sequences matching the protospacer seed region, which we could observe directly via CUT&RUN. This is a critical consideration for studies of enhancers and high MOI designs.

I do think there is an excellent opportunity here for a community or consortium effort to scale this approach genome-wide and systematically define active guides. These inactive reagents are a pernicious and widespread challenge in functional genomics.

We overcame this issue through the elegant strategy of applying brute force. Since Perturb-seq can now be scaled to measure >10,000 perturbations, we simply tested six guides each for 1,836 human TFs. We report active guides for 1319 of them.

Takehome 2: CRISPRa has some practical challenges. Here I must include what may be the most popular part of the preprint, now known as Extended Data Fig. 1C. It shows the high failure rate of CRISPRa guides targeting a model target locus, CD45.

A second advantage, which partly motivated this work, is that CRISPRa makes it easy to multiplex perturbations by expressing multiple guides simultaneously. This is important for TFs as combinatorial regulation is a major area of interest for us and others.

Takehome 1: CRISPRa is a great tool for perturbing transcription factors (TFs). We observe that it usually leads to physiological expression levels, which is a key advantage over ORF overexpression for some use cases.

A tweetorial describing our preprint is linked at thenormanlab.com, so I’ll focus here on key improvements (as the paper got quite a lot better in review) and some lingering thoughts.

rdcu.be/ezw15