Network Neuroscience for the Clinic: OHBM2022 Keynote Interview with Dr. Juan (Helen) Zhou
PhD Candidate at the Neural Systems and Behaviour Lab, Monash University, Australia
Dr. Juan (Helen) Zhou is an Associate Professor and Principal Investigator of the Multimodal Neuroimaging in Neuropsychiatric Disorders Laboratory in the Centre for Sleep and Cognition, Yong Loo Lin School of Medicine, National University of Singapore (NUS). She also holds a joint appointment with the Department of Electrical and Computer Engineering, NUS, and she currently serves as the Deputy Director for the Centre for Translational Magnetic Resonance Research at Yong Loo Lin School of Medicine. She recently finished her term as Council Secretary and a member of the Program Committee of the Organization for Human Brain Mapping.
Across these roles, Dr Zhou’s research focuses on the network-based vulnerability hypothesis in disease. Specifically, her lab studies the neural bases of human cognitive functions and the associated vulnerability patterns in ageing and neuropsychiatric disorders using multimodal neuroimaging methods, psychophysical techniques, and machine learning approaches.
Dr. Zhou presented a keynote address at OHBM 2022 in Glasgow—read on to learn about her research, career path, and hopes for the future of neuroimaging!
Alexander Holmes (AH): Welcome Dr. Zhou, thank you so much for joining us here—it is an honour to have you with us. Can you first tell us about your pathway into science and how you got to where you are now?
Helen Zhou (HZ): Ah, do you want the short answer or the long answer? When I was doing my undergraduate studies at the School of Computer Science and Engineering in Singapore, I was a part of this accelerated Masters program. During our final year, we needed to do some research, which was where I became interested in algorithms, neural networks, and image processing. When I tried these machine learning projects, it was my first hands-on experience using these algorithms to solve real problems. So, there were many ups and downs (Laughs).
I then realised that besides bioinformatics, my supervisor also researched brain imaging using MRI. At the time, I had several PhD offers in fairly different fields, but I decided to read further into brain MRI and found a long article introducing MRI, brain anatomy, and function. I still clearly remember the last sentence which said, ‘There are still many unknowns in this field’. This was good to know, as that means there's a lot of room for improvement and new research, right?
I chose to do my PhD with Professor Jagath Rajapakse and developed some novel computational methods to investigate variability of human brain anatomy and structure using MRI. During my PhD, one of the research assistants in charge of a collaborative project with clinicians from the National Neuroscience Institute left suddenly, and there was no one to actually perform the fMRI experiments. I agreed to help, as at the time—around 2004—fMRI was less commonly available. It was a good experience: this became the first brain imaging project related to Parkinson's disease in Singapore, so it was a very rare opportunity. Interacting with neurologists and radiologists in this project helped expand my horizons. After completing my PhD and working with many different algorithms, I kept thinking about how analytical approaches within MRI can improve our understanding of the human brain and brain disorders.
In 2008, I was very fortunate to have the opportunity to work as a postdoc studying memory and ageing with Professor William Seeley at UCSF. It was a really good environment for me to grow, with so many neurologists, researchers, scientists, nurses, PhDs, and post-docs interacting with each other. After that, I moved to New York, to do some neurodevelopmental research using diffusion MRI; in both roles I was immediately fascinated by the network-based neurodegeneration hypothesis and the need to use MRI techniques to study this.
Since returning to Singapore, I’ve set up my own lab in multimodal neuroimaging in neuropsychiatric disorders, within a neuroscience and behaviour disorder program in the Duke NUS Medical School at the National University of Singapore. We’re part of the Centre for Cognitive Neuroscience led by Professor Michael Chee, and we operate at the interface between computational approaches and their applications in behavioural disorders. I feel like I'm really at the right place. Within that supportive environment my lab has grown, gotten more funding, and we now train a lot of excellent PhD students and postdocs; so I'm very grateful for the journey.
AH: That’s a very interesting journey! It really highlights how you can have a lot of diverse fields coming together in what we study.
HZ: I also forgot to mention, but three years ago—right before COVID—my lab and the whole Centre moved to the main campus of the University to join the largest medical school in Singapore, the Yong Loo Lin School of Medicine. There we set up two centres: the Centre for Sleep and Cognition and the Centre for Translational Magnetic Resonance Research. So as you said, it's really a multidisciplinary research environment.
AH: You briefly touched upon it, but a lot of your work so far has focused on understanding ageing and neurodegeneration. What aspects of these areas interest you the most?
HZ: Well, no one is getting any younger! Previously, at UCSF, I scanned patients with frontotemporal dementia and other dementia subtypes, and realised these patients need a lot of support. After so many years studying ageing and ageing-related disorders, it's very clear that we cannot just disassociate the two. They go hand in hand.
Our current longitudinal ageing study is our signature program in the school, investigating early detection, early intervention, and the preclinical disease stage in dementia. Based on traditional evaluations, many pre-clinical cases are currently considered healthy, so we have to instead consider how different risk factors, different pathology, or different genetic backgrounds influence their current status and future progression. Ideally, years before they have any symptoms, you could obtain a comprehensive baseline and a quantifiable brain measure that reliably estimates the likelihood of developing these disorders in the future.
AH: One of your most impactful papers investigated network connectivity in Alzheimer’s disease and in frontotemporal dementia. What aspects of these neurodegenerative disorders best lend themselves to being explored through the lens of connectivity at a network-wide scale?
HZ: These two disorders are really very, very interesting. With Alzheimer’s disease, patients obviously lose their memory and visuospatial function, but they also become more sensitive to emotional stimuli. In contrast, frontotemporal dementia patients don't experience much memory loss, but they experience socioemotional deficits, such as loss of empathy and disinhibition. The two disorders have opposing symptom deficit profiles. This is a perfect test case for a network-based neurodegeneration hypothesis. If we can see the contrasting patterns in the brain supporting these opposing symptoms, then that supports the network-based dysfunction model. And that’s what we found.
This has now been replicated in many other studies, including in MCI with and without behavioural problems. We strongly believe that this reciprocal network system, like the salience network and the Default Mode network, are anti-correlated and segregated. If one goes up, then the other goes down, or vice versa. Extending this to some of our recent work, we tested the importance of the salience network in the context of ageing and psychosis, where we found that this function was also really important. We keep seeing some kind of segregation between the Default Mode network and task positive networks including the salience and executive control network.
AH: Many of your recent papers have been investigating these brain-behaviour relationships and trying to predict disease progression. Can you tell me more about that?
HZ: In one study, we looked at healthily ageing people at risk for dementia, and we found that they lost this segregation between the networks—and that this loss was related to the rate of decline in processing speed over time.
In another study, we're investigating different subtypes of Alzheimer’s disease with and without comorbid cerebral vascular disease. Alongside the traditional memory problems, Alzheimer’s disease is also characterised by executive function or processing speed deficits, but with comorbid cerebral vascular disease (CVD), we see an additive effect on cognitive decline. The most interesting part, however, is that we also find differing network phenotypes between patients with and without CVD. This is further supported by a recent longitudinal study we have done in collaboration with Korean colleagues revealing how amyloid and CVD divergently influence brain networks in the prodrome stage. This is quite exciting work, and it demonstrates that across the Alzheimer’s disease spectrum, brain networks will behave very differently depending on the degree of pathology.
We’ve also seen that patients with Alzheimer’s disease who have behavioural problems—not as severe as in frontotemporal dementia, but behavioural problems nonetheless—tend to have a faster disease progression and divergent neuropathology. Because there is great inter-subject variability, you have to acknowledge both the condition and the behaviour. Having the same disease doesn't guarantee that you will have the same pattern of network breakdown; it depends on what symptoms you have.
One last example from my lab, which was also one of the oral presentations at OHBM this year, used graph convolutional networks to encode the network information and nodal measures from fMRI, and then predict future levels of regional atrophy over time. That information can then predict someone’s future condition or symptom changes and we can identify types of activity in specific networks that are more predictive of disease progression.
AH: As multimodal research is also a key part of your lab’s work, how would you say the growing number of larger, open-access datasets with multiple modalities has shaped this work?
HZ: That's a very good question. I think we should really thank the Alzheimer Disease Neuroimaging Initiative (ADNI) and many other initiatives. When I was at New York University, I was under the supervision of Professor Michael Milham, who led the field in open science and data sharing, having shared the Nathan Klein Institute Rockland Sample dataset and organised the ADHD 200 dataset around that time. Now, there are many sites sharing their data, like the UK Biobank dataset which also includes multiple modalities, for example.
Data sharing has really helped us as a field, especially when using computational approaches, machine learning, and other related techniques. Whenever we think about any interesting questions, I will immediately tell my team, ‘Can you please look at the public datasets too?’ Because replication is important, right? We have our local datasets and we can have exciting findings, but I always ask my lab members to try to find a public dataset and then replicate their findings.
AH: What would you say are the most exciting things that your lab is working on right now?
HZ: First, I'm part of this five-year research program, which aims to recruit 1200 participants at risk for vascular cognitive dementia to complete a two-year clinical trial in Singapore. We are going to evaluate the efficacy of this multi-domain lifestyle intervention at reducing vascular impairment. The interesting thing is that we’re investigating these multimodal biomarkers including, brain imaging, retinal imaging, blood biomarkers, and genetics, which we’re integrating together through machine learning to identify who can benefit from intervention and why.
In parallel, we are also researching large scale, ageing cohorts in collaboration with many other PIs. We plan to recruit over 2000 elderly individuals and conduct a wide variety of multi-organ evaluations, including brain imaging. This is a very promising cohort, as they’re all healthy, so we want to see how various factors influence their physical and mental health. We aim to specifically detect early changes in the critical stage in Alzheimer’s, but will also longitudinally track the progression of different measures over time. Eventually, we want to integrate these different sources of information to make it more reliable and easier for patients too.
AH: Very interesting! It seems like these areas will be very fruitful and productive. Given your wide range of work, what do you consider to be your greatest scientific achievement?
HZ: I think there's no singular greatest achievement, rather a combination of work that comes together to tell a story. But if I have to pick from when I was a junior researcher, I would say the 2010 Brain paper and the 2012 Neuron paper. They come together to demonstrate both the differential phenotype of dementia subtypes as well as the possibility of using the functional connectome to predict progression.
Since setting up my own lab in Singapore, I would also say that two big contributions we have made include the 2016 Neuroimage paper, which demonstrates how functional segregation changes over time with ageing and its relationship with declining processing speed; alongside another paper published in 2017 in Brain, which was the first time demonstrating that Alzheimer’s Disease activity had very differential network phenotypes. My team is still following up on those lines of research, and we hope to have more exciting findings soon.
AH: Looking ahead then, what do you believe will be the major future directions and applications of neuroimaging and brain connectivity?
HZ: We need to continue investigating the relationship between brain and behaviour in health—or, in certain disorders, between the brain and symptoms. Rather than simply predicting behaviour, we need to actually understand the underlying relationship. In one recent study published in eLife in 2022, we used fluorodeoxyglucose (FDG)-positron emission tomography (PET), which provides very rich information about the brains’ pathology and metabolism. Integrating MRI with PET— using deep learning, for example—can tell us more about these mechanisms.
We also need to consider problems within current fMRI sequences This has led to the development of multi-echo fMRI and new efforts to improve the test-retest reliability, signal to noise ratio, and reproducibility of fMRI research. But these new techniques need new analytical approaches: methods to capture the subtle changes in our higher order networks at a layer-specific level. I believe the biological and the computational techniques should inform each other, and we should integrate these tools in a more meaningful manner.
My team is also actively working on intervention strategies. The connectome has proven quite useful in terms of TMS studies treating depression. We are trying to develop imaging markers which can help us resolve the inter-subject variability in these disorders and optimise intervention strategies. For example, we can use imaging markers to identify high-risk participants and intervene early when they are most responsive to treatment; i.e., when the most severe or irreversible changes haven’t taken place yet.
Finally, I think there will be more interest towards integrating multiple modalities, like we mentioned before. Nowadays, digital phenotyping is quite popular, but we need to consider how we extract these measures from additional phenotypes and how we integrate them with other brain measures to make meaningful conclusions..
AH: Finally, how would you summarise your keynote lecture at OHBM 2022 for our readers?
HZ: Okay, this is going to be a short answer. In the talk, I focus on differential brain network phenotyping in neurological disorders, particularly neurodegenerative and cerebral vascular disorders. I talk about progressions from the preclinical, prodromal stage to the dementia stage and share some new findings from our group, so please check out the recording on OnDemand.
AH: I definitely look forward to re-watching that talk. Thank you once again for taking the time to talk with me today.
HZ: Thank you very much for having me.