I am the proud owner of MJ and Ella. MJ is quite skilled at swiping food off the counter, whether it’s freshly-baked cake, my Chipotle bowl, or a juicy hamburger. Ella is a professional shoe thief, and I’ve spent too much time searching for them. I named Ella after Legionella pneumophila (legionELLA), a microbe that I worked on for many days (and nights) during my undergrad studies

Kim and I had the pleasure of speaking with Dr. Vanessa Hale at The Ohio State University (OSU), who researches the canine urine microbiome, termed the "urobiome." The Hale Lab approaches the urobiome with a OneHealth perspective, where they recognize the interconnectedness of humans, animals, and environmental health. 

For example, the dog is an excellent model to study bladder cancer. Did you know that the presentation of bladder cancer in humans and dogs is very similar? Bladder cancer is a disease linked to environmental exposures, and pets interact in the same environment that humans do, whether inside or outside the home. Dogs are exposed to many of the same exposures that we are, including, but not limited to, smoking, pollution, pesticides and drinking water quality. Given their naturally shorter lifespans, relative to humans, dogs provide a unique opportunity to study disease trajectories that would take decades to observe in humans. 

Furthermore, dogs and their human companions share microbes, with some human-originating strains appearing in their dogs’ ‘biome.’ That’s right! The next time your dog gives you a surprise kiss on the face or gets all up into your personal space, remember all those microbes swapping back and forth in your microbiomes. This shared microbial landscape highlights how our pets are truly a part of our hearts and families, not just our homes. 

Dr. Hale’s innovative research focuses on the microbes within the bladder, and how these microbes might mediate cancer development through their metabolic activity. The bladder was once thought to be sterile, however, advancements in NGS and improved sample collection revealed a healthy and stable urobiome. 

We are still in the early stages of elucidating the role of the urobiome in health and disease, but with advancements in protocols and sequencing, and willing pet-parents, the potential for scientific advancements in the urobiome space is enormous. Even though urobiome research is still new, its future is exciting. 

Dr. Hale is particularly excited about exploring the gut-bladder axis, and asks “can oral administration of microbe-based treatments influence the urobiome and therefore improve bladder cancer outcomes?” She is also excited about the potential of urobiome transplants (like fecal microbiota transplants, but for the bladder)--but emphasized the critical need for more research to better understand which strains are safe and effective. 

One important takeaway that Dr. Hale wishes all pet parents knew was to be cautious about the overwhelming amount of information and packaging promoting pet probiotics. Although the marketing for pet probiotics is massive, more scientific validation is needed to know the effectiveness of these pet probiotics. She recommended seeking evidence-backed solutions and a trained vet professional on whether or not your pet requires these pet probiotics. 

Working with and being a pet-parent to dogs has its unique rewards and challenges. Dr. Hale shared that her favorite part of her work is the deep connection between dogs and their owners. She loves hearing updates from pet parents on how their dog is doing post clinical trials. On the other hand, we know the trouble that our furry friends can get into. 

I can’t count the number of times that MJ has been caught red-handed, stealing cake off the kitchen counter or my other dog, Ella, has been stealing my shoes to start her own collection on my bed. In one of Dr. Hale’s pet-parent stories, she shared that her dog, Kota, ate a corn cob out of the composting bin, which got stuck in her digestive tract. Dr. Hale immediately took her dog to the Ohio State Veterinary Medical Center where Kota had surgery to have the corn cob removed. 

Thanks to Dr. Hale’s quick actions and the veterinary team at the Ohio State University, Kota is healthy and happy. Her dog’s legacy will live on in the Ohio State Vet School Curriculum, where her dog's abdominal radiograph presents a unique picture of “Corn Cob Gut Obstruction”. Look closely where the yellow circle is to see the corn cob. Note that Kota was adopted as an adult dog with an unknown history. The small bright white circles in this radiograph are bird shot. We didn’t know until we saw this radiograph that she’d previously been shot and she has no signs of being affected by the bird shot otherwise. Rest assured, she is a very happy, healthy, loved dog!

This National Dog Day, let’s celebrate not just our pets, but also the incredible scientific advances that they are contributing to. Happy Dog Day to all! 

A special thank you to Dr. Vanessa Hale for giving us her time and her stories. You can connect with her here on LinkedIn or via email (hale.502@osu.edu). If you’re interested in learning more about her work, you can view her ongoing bladder cancer trial here, as well as her work with Dr. Jenessa Winston on two dog owners who love microbes! 

Curious about how pet-parents are instructed to collect urine samples from their pets? Watch this video (skip to minute 3:20) and see how Dr. Vanessa Hale followed her dog, Kota, around to collect a urine sample. Ah, the things we do for our pets (and science).

Machine-guided design of cell-type-targeting cis-regulatory elements

Gosai SJ, Castro RI, Fuentes N, et al.  |  Nature 634: 1211–1220
doi: 10.1038/s41586-024-08070-z

Researchers from Yale University, the Broad Institute of MIT and Harvard, and The Jackson Laboratory developed a method to effectively control gene expression by using AI-designed DNA switches, allowing them to "flip genes on and off" in specific cell types within an organism, potentially leading to new avenues for targeted gene therapies and medical research.

The technology utilizes an AI platform called "Computational Optimization of DNA Activity (CODA)" capable of designing synthetic DNA sequences that act as highly specific gene switches. These AI-designed switches can be used to activate genes in only certain targeted cell types. This could revolutionize gene therapy by allowing for more precise delivery of treatments, potentially targeting diseases like metabolic disorders without causing unwanted side effects in healthy cells. 

Longitudinal multi-omics analysis of host microbiome architecture and immune responses during short-term spaceflight

Tierney BT, Kim J, Overbey EG, et al.  |  Nat Microbiol 9:  1661–1675
doi: 10.1038/s41564-024-01635-8

As the number of commercial space missions increase and space tourism grows, more people with diverse medical histories and microbiomes will be sent into space. It is critical to gain an understanding of the complex interactions between the space environment and the composition of the hosts' microbiota in order to be able to avoid putting lives at risk.

NASA is employing metagenomics to identify and quantify microbes on spacecraft surfaces and in cleanrooms. The goal is to assess potential risks associated with microbes that can survive in space, such as those that are resistant to radiation, desiccation, and ionization.

In this article, results are presented from the analysis of swabs collected from ten different areas on the body to study three types of microbial ecosystems (oral, nasal, skin) at eight different time points. Astronauts were found to experience changes in their microbiota and immune response during spaceflight. For example, the amount of skin bacteria associated with skin rashes or hypersensitivity increased. The oral microbiome displayed mission-dependent changes in metatranscriptomic expression of bacteria associated with tooth decay and biofilm formation. The number of gut bacteria associated with intestinal inflammation increased, while the number of bacteria with anti-inflammatory properties decreased.

These changes can be transient or more long-lasting and vary by astronaut, study, and body location. Astronaut microbiota, however, typically returns to its natural baseline when they return to Earth.

Additional experiments and missions are needed to further test the microbiome-derived theory of spaceflight-associated immune changes.

Brain aging patterns in a large and diverse cohort of 49,482 individuals

Yang Z, Wen J, Erus G, et al.  |  Nat Med 30: 3015–3026
doi: 10.1038/s41591-024-03144-x

It has long been known that aging causes changes in brain anatomy, some of which can be detected using magnetic resonance imaging (MRI). However, the patterns of systematic brain changes are invisible to the human eye and remain largely unexplored.

Studies using data from a restricted number of people have shown that machine learning methods can help to solve this problem by extracting subtle fingerprints of aging from MRI data. In this paper, the deep-learning algorithm, Surreal-GAN -- previously developed by the first author of the article -- was trained on MRI scans of the brains of 1,150 healthy people aged 20 to 49 and 8,992 older people, many of whom showed reduced cognitive abilities. The trained algorithm was subsequently applied to nearly 50,000 people participating in various studies of aging and neurological health.

This analysis identified five discrete patterns of brain atrophy, various combinations of which were associated with different types of age-related brain degeneration. For example, dementia and its precursor, mild cognitive impairment, were associated with three of the five patterns. Other patterns were associated with conditions such as Parkinson's disease and Alzheimer's disease, and one combination of three patterns was highly predictive of mortality.

The data analysis also showed that the identified models could potentially be used to estimate the likelihood of greater brain degeneration in the future. Five distinct patterns of brain atrophy associated with aging and neurodegenerative diseases were also linked to factors such as smoking and alcohol consumption, genetic factors, and blood markers that influence health and disease risk.

This new approach can significantly advance our understanding of the subtle mechanisms of age-related changes in brain anatomy and advance the possibility of their targeted control.

Nucleotide Transformer: building and evaluating robust foundation models for human genomics

Dalla-Torre H, Gonzalez L, Mendoza-Revilla J, et al.  |   Nat Methods 22: 287–297
doi: 10.1038/s41592-024-02523-z

The authors of the article developed the "Nucleotide Transformer",  a series of the generalized AI models for genomics applications. Such models are pre-trained on DNA sequences using self-supervised learning, which then extracts context-sensitive representations of nucleotide sequences suitable for accurate predictions of molecular phenotypes from genotypes from publicly available genomic data.

The training was performed on 128 GPUs across 16 compute nodes. This process lasted 28 days, during which models with 2.5 billion parameters were generated. Following this, a robust benchmarking and statistical evaluation pipeline was created to evaluate the analysis design used for various tasks related to the identification of chromatin features, DNA regulatory elements, and splice sites in the human genome.

Benchmarking demonstrated Nucleotide Transformer models to perform better than self-supervised approaches such as Hyena-DNA and DNABERT-2. During the pre-training process, Nucleotide Transformer models acquired biological knowledge applying, for example, to regulatory elements and differentiation between coding and non-coding regions.

Results confirm the promise of the developed approach, despite a number of limitations related to the limited sizes of nucleotide sequences that they could process, how effectively they capture long-range interactions, and the lack of cell type-specific information.

It should be noted that Nucleotide Transformer models are currently used in agricultural applications for annotating genomes at single-nucleotide resolution and for solving specific problems on proteins from their corresponding coding sequences. The paper also provides other examples of the acquisition of knowledge about genomic elements by Nucleotide Transformer models.

Training and application of fundamental models in genomics provides a widely applicable approach for accurate prediction of molecular phenotype from DNA sequence.

Personalized pangenome references

Sirén J, Eskandar P, Ungaro MT, et al.  |  Nat Methods 21: 2017–2023
doi: 10.1038/s41592-024-02407-2

A pangenome is the complete set of common and unique genomes present in a given species. It combines the genetic information of all sampled genomes, resulting in a large and diverse range of genetic material. The pangenome concept--first proposed in 2005 for bacterial studies--is applicable to many other species, including humans.

Pangenomes reduce reference bias by representing genetic diversity better than a single reference sequence. However, when comparing a new sample to a pangenome, misleading variants may arise, causing, for example, false read mappings. Such irrelevant variants are usually rare and were previously addressed by filtering them out. The same time, this approach did not remove all irrelevant variants and lost some relevant ones.

The authors developed a k-mer-based method for sampling local haplotypes similar to the sequenced genome. Using a subgraph based on sampled haplotypes as a reference for read mapping yields more accurate results than the full graph or a frequency-filtered graph containing only common variants.

The approach is implemented in the vg toolkit (https://github.com/vgteam/vg) for the Giraffe short read aligner and its accuracy was assessed while compared with modern methods using human pangenome graphs from the Human Pangenome Reference Consortium.

The authors showed that the proposed approach reduces small variant genotyping errors by four times compared to a set of genome analysis tools, making genotyping of structural variants of short reads competitive with long read variant detection methods. This represents an improvement over previous filter-based methods, but there are some limitations to overcome.