1 2 aim
Treating Duchenne Muscular Dystrophy: The Promise of Stem Cells, Artificial Intelligence, and Multi-Omics

#apaperaday: Treating Duchenne Muscular Dystrophy: The Promise of Stem Cells, Artificial Intelligence, and Multi-Omics

In today’s #apaperaday, Prof. Aartsma-Rus reads and comments on the paper titled: Treating Duchenne Muscular Dystrophy: The Promise of Stem Cells, Artificial Intelligence, and Multi-Omics.

Today’s pick is from Frontiers in Cardiovascular Medicine by Vera et al on the promise of stem cells, artificial intelligence and multi-omics for Duchenne therapy development. 3 interesting topics but the review is high level and not very critical. Doi: 10.3389/fcvm.2022.851491

Duchenne is caused by lack of dystrophin, which results in many pathological processes in skeletal muscle and heart, which jointly lead to reduced regeneration and replacement of muscle by fibrosis. Markers of pathology can be seen in blood.

Authors outline that two strategies are being pursued: restoring dystrophin and reducing pathology/improving muscle quality. Authors stress the second one is important as well, since only ‘fixing dystrophin’ will not counteract accumulated muscle damage. Fully agree!

Authors also outline that the field has learned a lot from studying model systems. Agree again. They first focus on the use of human stem cells to further study the disease. Animal models are not optimal, as the most used mdx mouse model does not fully recapitulate Duchenne.

They give the example that onset of cardiomyopathy in mdx mice is late, while most patients die from cardiac failure. I think this is a bit of a shortsighted statement.

  1. Mdx mice do have heart problems as well and can die from stress since the heart cannot cope.
  2. Most patients die due to heart failure because multidisciplinary care (fortunately) avoids patients dying from other causes, such as respiratory failure, pneumonia etc

Having said that, mdx mouse is not a perfect model, but no model systems are perfect.

Authors offer human induced pluripotent stem cells (IPSCs) as a better alternative. IPSCs are cells that have the potential to become all cell types (pluripotent) – however you have to ‘tease’ them into become a specific cell type (e.g. heart of skeletal muscle).

You can make IPSCs from a patient’s cells (fibroblasts from a skin biopsy are common, but you can also make them from cells in urine!). You can also use IPSCs from a healthy individual and generate a mutation with genome editing (CRISPR/Cas9) – the latter allows better comparison.

These IPSC models differentiated into the heart of skeletal cells can then be used to study pathology (how do they differ from cells making dystrophin) and to discover treatments (do the deficits improve with treatment?). e.g. Duchenne IPSC heart cells improve with beta-blockers.

What the authors do not indicate (and this is the lack of criticism I mentioned in tweet 1) is that these cell models are not models for postnatal heart or skeletal muscle but for fetal heart and skeletal muscle. So that is a limitation.

Secondly, you have only a single cell type and are unable to look at the complete picture, which also involves inflammatory cells, fibroblasts etc. I do not say IPSCs are not useful and cannot provide us with information – however, authors should have outlined limitations.

Often it will still be needed to also test in animal models as there you do see the complete picture in postnatal animals. Next authors go into the potential of artificial intelligence and machine learning. With all the models and all the data accumulating this can help.

One example this can be exploited is by comparing drug profiles with pathology profiles and ‘match’ them. Companies doing this are now established and authors show an example for Duchenne. It contains many known treatments/approaches (good sign), but also some that did not work.

Again authors are not critical and do not go into requirements that are needed for this type of work: it relies on high quality data! There is a lot of data available for Duchenne models, but the quality sadly varies. So one has to check carefully before using a dataset.

This is the famous ‘crap in crap out’ mantra. When using good quality data artificial intelligence can suggest candidate drugs for drug repurposing for Duchenne. However, also here there are things to be aware of: e.g. is the dose needed feasible in Duchenne patients?

Has the drug been tested in children? Are there off target effects that might not be a problem in the disease for which the drug is approved, but which could be a problem in Duchenne (due to the pathology or use of e.g. steroids). Finally, do they add anything on top of steroids?

The last aspect of the review is multi omics. Omics is the approach of generating a lot of data in an untargeted fashion (so without targeting a specific protein or transcript or gene). However, the data generated varies for proteins, transcripts, metabolites etc.

This makes it difficult to integrate them and compare them, and connect them, while this would be very useful. Authors outline this is challenging and it is, but they do not offer suggestions to solve this, e.g. through statistical modeling.

They also do not mention that the multi omics data should also include functional data in order to correlate and predict. Shoutout to Pietro Spitali and Mirko Signorelli for generating models for this, like this example.

Authors end by stating that a multidisciplinary approach is needed for the multi omics and artificial intelligence. I agree, and George Paliouras and our group are trying to get funding for this type of interdisciplinary work.

All in all the review is nice because it highlights some important topics, but the authors only scratch the surface and mainly highlight opportunities and when they do mention challenges they do not suggest solutions or inroads towards this. So a bit ambiguous.