Building a Unified Platform for Biological Discovery with CRISPR

Can CRISPR help build a unified platform for biological discovery?

Dr Kaivalya Shevade from the Laboratory for Genomics Research (UCSF) is developing new CRISPR-based screening methods to map gene networks, understand drug resistance, and track neuronal activity in disease. 

The research team’s innovations, including the CAT-ATAC assay and the Plexus machine learning model, show how combining biology with computation could accelerate the search for new treatments.

Read the original research: doi.org/10.1101/2025.02.11.637716

Transcript

Hello and welcome to Research Pod! Thank you for listening and joining us today. 

In this episode, we look at the work of Dr. Kaivalya Shevade, researcher at the Laboratory for Genomics Research, University of California San Francisco – in particular, his pioneering work using cellular models and CRISPR screens to reveal genetic mechanisms driving diseases. 

The Laboratory for Genomics Research, brings together experts from academia and industry to pioneer technologies and further our understanding of genetics. At the intersection of biology, innovative technology, and drug discovery, they collaborate to find new drug targets and produce treatments for a range of genetic diseases. 

The complete sequencing of the human genome achieved in the last few years revealed over 3 billion bases of DNA. Scientists are now able to interrogate this vast genome to identify genetic difference between health and disease and further our knowledge of the mechanisms driving these. 

Crucial to this endeavour was the discovery of the gene editing tool, CRISPR. These “genetic scissors” programmed to target specific locations in the genome can permanently remove a gene or temporarily turn it on or off. Such manipulations help scientists to understand the role different genes play in disease. 

An expert in this field is Dr. Kaivalya Shevade who specialises in developing stem cell derived models and using CRISPR screens to determine which genes are responsible for eliciting cellular functions. Central to this process are automative and computational abilities needed to detect and interpret the generated data. These scientists therefore integrate biology with machine learning, pushing boundaries to develop scalable technologies capable of harnessing a wealth of genetic information. 

Shevade’smost recent work detailed in two carefully reviewed research papers aims to build CRISPR screening platforms with arrayed imaging and pooled multiomic readouts to systematically identify the role of different genes in a single cell.

To understand how this research was done, let’s first hear in a bit more detail how these techniques work.

Naturally occurring in bacteria for inherent protection against foreign viruses, CRISPR-Cas9 is now engineered for use in human genomics studies. Directed to where it is needed by guide RNA, the enzyme protein, Cas9 acts as scissors to cut a specific DNA sequence and in doing so alters gene function. This alteration is known as a perturbation. Such perturbations can be performed in large-scale experiments known as CRISPR screens and enable scientists to deduce which genes are responsible for certain cellular functions or phenotypes by systematically switching them off. This approach helps scientists “find a needle in a haystack” narrowing down a large number of possible genes involved in a cellular response to a short hit list of candidate genes.

Pivotal to both the ability of CRISPR to perform its function and for gene regulation is DNA accessibility. This refers to how accessible DNA sequences are for gene regulatory factors to physically bind to it. So, in addition to determining the role of different genes, scientists want to know how genetic or chemical alterations affect DNA accessibility to better understand gene regulatory relationships. 

To achieve this, sequencing assays are used in a process called phenotypic profiling to identify phenotypic changes caused by either genetic or drug modifications. Simultaneous use of such phenotypic assays amplifies the type and amount of information generated enabling thorough exploration of the effects of perturbations using CRISPR. In addition to DNA accessibility, RNA readouts known as the transcriptome can be profiled and in doing so provide major insight into gene expression and function. 

With extensive experience in phenotypic profiling and running CRISPR screens, Dr. Kaivalya Shevade and team are moving this field forward by developing a novel technique named CAT-ATAC. This method compatible with commercially available genomic assays can simultaneously provide a multidimensional wealth of data including transcriptome and mapping DNA accessibility as well as guide RNA profiles to evaluate the CRISPR perturbation status. The researchers hope this “3 in 1” readout will majorly advance the discovery of gene-regulatory networkstransforming regulatory network inference from correlational to causal, enabling researchers to directly test how specific genes or regulatory elements influence chromatin accessibility and gene expression in individual cells. Such networks involve numerous molecular regulators including DNA, RNA and proteins that work in an intricate system to regulate gene expression. Full elucidation of these networks will be key to understanding many complex biological processes. 

In a first of its kind study, the team used their novel CAT-ATAC technique to explore the gene regulatory networks underpinning drug resistance to treatments-specifically that of Dasatinib, a drug used to treat chronic myeloid leukemia. 

This work is critical to understand the cause of Dasatinib treatment resistance so that non-drug-resistant treatments for chronic myeloid leukemia can be developed. 

The team concentrated efforts on overcoming a current technical challenge during phenotypic profiling. 

Key to the success of this profiling is the ability to simultaneously generate and interpret genetic readouts. An initial step in interpreting gene perturbation readouts from CRISPR screens requires guide RNA to be captured and correctly assigned to the corresponding cellular phenotypic readout. But when readouts occur simultaneously, complex protocols and steps are needed to correctly capture guide RNA and interpret the data. However, the researchers’ novel technique, CAT-ATAC removes the need for such complexity by improving guide capture efficiency up to a level of 77% in both cancer and pluripotent cells. 

Application of this technique has unveiled gene regulatory networks involved in Dasatinib resistance and identified two genes associated with the resistance. 

“These types of multiparametric single-cell readouts will be transformative to shape our understanding of fundamental biological processes, cell fate decisions and disease mechanisms by harnessing the heterogeneity inherent to complex cell populations instead of reducing datasets to bulk analysis,” remarked the researchers in their manuscript.

Another part of the team’s research journey done in collaboration with Dr. Parker Grosjean from Dr. Martin Kampmann’s lab at UCSF is their work on neuron activity dynamics in neurodegenerative disease specifically frontotemporal dementia. This time the team developed a novel machine learning model and applied it to high throughput phenotypic profiling to progress this field from manual engineering processes to automative. Further integration of the model with CRISPR screening enabled identification of genetic modifiers of deviant neuronal activity. 

Neuronal activity is dynamic meaning electrical and chemical signals transmitted between neurons are constantly changing. Measuring calcium levels in neurons can be used as a proxy for neuronal electrical activity. These dynamic processes are challenging to capture and interpret and up until now limited our knowledge of neurological disorders such as neurodegeneration and epilepsy. 

To overcome this, the team needed to design a semi-automated multidimensional system capable of capturing dynamic neuronal activity. But how did they do this?

Using stem cells, they developed a neuronal model that underwent high-content microscopy to measure the dynamic calcium levels in neurons and then introduced CRISPR based perturbations to identify genes that cause phenotypic changes in cells pertaining to their dynamic activity. Crucial to this multidimensional system was the addition of the team’s novel machine learning model, Plexus. Plexus is a self-supervised learning model that trains itself from the biological data it’s analysing. This novel model can specifically capture network activity dynamics outperforming other self-supervised learning approaches. Able to identify almost double the number of phenotypic changes resulting from CRISPR perturbations compared to other approaches, the researchers believe Plexus could be a game changer in the discovery of modifiers of neuronal activity in other neurological diseases. 

In terms of pharmaceutics, the team say; “…Combining genetic and small molecule perturbations with Plexus phenotyping could also provide insight into the mechanism of action of neuroactive compounds – a challenging task in drug discovery efforts.”

The scientists suggest in their paper that their new method could also be applied to other biomedical research fields. 

“While we focused on neuronal activity as a phenotype, this method could be used for any dynamic biological process, where biological context influences single-cell phenotypes,” they wrote. 

The researchers will continue their work in advancing and optimising experimental platforms and integrating these arrayed imaging platforms with transcriptomic and DNA accessibility data from CAT-ATAC to train new machine learning models to gain insights into complex diseases on a scale never seen before. The goal being to translate biomedical research into treatments to help patients. 

That’s all for this episode – thanks for listening and stay subscribed to Research Pod for more of the latest science. 

See you again soon.