Seeing the light: automatically identifying key anatomical changes in light sheet microscopy images of plant roots
Lead Research Organisation:
University of Nottingham
Department Name: Sch of Biosciences
Abstract
To meet an increasing demand on food production, there is more need now than ever before to understand and improve the efficiency of crop growth and yield. Here, we consider the anatomical changes of roots under low phosphate levels that give rise to important architectural changes in the root system. The study of these anatomical changes in features such as cell divisions over time is only possible to due to recent technological advances in imaging and expertise in developing analysis software. For the first time, we aim to find the origins of these anatomical changes.
Phosphorus is one of the key macronutrients (alongside nitrogen and potassium) required for healthy plant growth, and is a widely used constituent of fertilizer used in commercial crop production. Phosphorus is a limited strategic resource: it is derived from a finite natural supply. Understanding phosphorus use and its effects on growth in plants is therefore of key importance to Global Food Security. Phosphorus mobility in the soil is limited by slow diffusion, and so areas of low phosphate concentration are created around root system. Much work has been carried out examining the overall architecture of root systems under differing conditions; however, much less is understood about the anatomical changes, and how they develop. There are only a few instances where root anatomical and architectural traits have been combined in a systematic way to select for plants with enhanced nutrient acquisition, but when this has been done the improvements have been staggering . For example, one study in Mozambique selected common bean varieties with shallow root angle (to enhance topsoil exploration) and enhanced root hair growth (to increase contact with the soil), this led to an almost 300% increase in plant biomass on low phosphorus soils, twice that expected from the additive benefits of these traits in isolation. This shows a great importance in understanding anatomical traits and understanding how they respond to low nutrient environments.
There are several reasons for our lack of knowledge in adaptive anatomical responses in roots. First, the equipment has not been available to image the plants growing over the time periods these anatomical changes take place. Second, the ability to alter the growth conditions (such as nutrient levels) around the root whilst being imaged has not been possible. Third, discovering subtle anatomical changes in the huge datasets that would be generated is a very challenging manual task. Our work will allow us to overcome these challenges, allowing study of anatomical changes in low phosphorus media dynamically.
In this proposal, we use a cutting-edge microscope, a light sheet fluoresence microscope (or LSFM) to image cellular anatomy as plants grow. A key challenge with LSFM time series data is the huge volume of data produced. This requires new analysis methods to permit us to gain biological insight. One such problem is identifying formative divisions that give rise to anatomical patterns in 3D datasets of roots resolved over time. This proposal seeks to automatically identify particular anatomical changes in big datasets. With LSFM it is possible to acquire gigabytes of data per minute. Time course experiments can easily generate tens of gigabytes of data. This turns visualisation and analysis into a bottleneck. We propose a solution. We will use machine learning approaches to allow new software to identify regions of interest within these datasets. We will build visualisation tools which will use the results of these approaches to allow biologists to navigate the data in meaningful ways, rather than blindly moving through the whole dataset. We will use these tools to investigate how root anatomy is altered in plants grown in low nutrient environments.
Phosphorus is one of the key macronutrients (alongside nitrogen and potassium) required for healthy plant growth, and is a widely used constituent of fertilizer used in commercial crop production. Phosphorus is a limited strategic resource: it is derived from a finite natural supply. Understanding phosphorus use and its effects on growth in plants is therefore of key importance to Global Food Security. Phosphorus mobility in the soil is limited by slow diffusion, and so areas of low phosphate concentration are created around root system. Much work has been carried out examining the overall architecture of root systems under differing conditions; however, much less is understood about the anatomical changes, and how they develop. There are only a few instances where root anatomical and architectural traits have been combined in a systematic way to select for plants with enhanced nutrient acquisition, but when this has been done the improvements have been staggering . For example, one study in Mozambique selected common bean varieties with shallow root angle (to enhance topsoil exploration) and enhanced root hair growth (to increase contact with the soil), this led to an almost 300% increase in plant biomass on low phosphorus soils, twice that expected from the additive benefits of these traits in isolation. This shows a great importance in understanding anatomical traits and understanding how they respond to low nutrient environments.
There are several reasons for our lack of knowledge in adaptive anatomical responses in roots. First, the equipment has not been available to image the plants growing over the time periods these anatomical changes take place. Second, the ability to alter the growth conditions (such as nutrient levels) around the root whilst being imaged has not been possible. Third, discovering subtle anatomical changes in the huge datasets that would be generated is a very challenging manual task. Our work will allow us to overcome these challenges, allowing study of anatomical changes in low phosphorus media dynamically.
In this proposal, we use a cutting-edge microscope, a light sheet fluoresence microscope (or LSFM) to image cellular anatomy as plants grow. A key challenge with LSFM time series data is the huge volume of data produced. This requires new analysis methods to permit us to gain biological insight. One such problem is identifying formative divisions that give rise to anatomical patterns in 3D datasets of roots resolved over time. This proposal seeks to automatically identify particular anatomical changes in big datasets. With LSFM it is possible to acquire gigabytes of data per minute. Time course experiments can easily generate tens of gigabytes of data. This turns visualisation and analysis into a bottleneck. We propose a solution. We will use machine learning approaches to allow new software to identify regions of interest within these datasets. We will build visualisation tools which will use the results of these approaches to allow biologists to navigate the data in meaningful ways, rather than blindly moving through the whole dataset. We will use these tools to investigate how root anatomy is altered in plants grown in low nutrient environments.
Technical Summary
This proposal seeks to identify the particular anatomical changes that occur in low phosphate conditions, using light sheet microscopy (LSFM) and novel image analysis approaches.
Preliminary results show that root anatomy is significantly altered under low phosphorus conditions. In our static images of 14 day old roots we can observe additional endodermis, cortex and epidermal cells. As a result of the increase in the number of cortical cells many more epidermal cells than usual overlay clefts between cortex cells. It is well established that epidermal cells overlaying the clefts between cortical cells go on to form root hair bearing cells, and consistent with this we see an increase in the number of hair-bearing versus non-hair bearing cells. In bean, it has been well documented that increased root hair density results in improved growth under low phosphorus. Therefore these anatomical changes are likely to cause important environmental adaptations.
Whilst we can generate a hypothesis about what benefit increasing the number of cortical cells may bring, we also see other anatomical changes (such as those relating to the organization of the meristem) that we currently don't understand. Our novel software-based analysis of these datasets will allow us to pinpoint precise changes, such as an increase in number or size of specific cell types, as well as organization of the root structure, by primarily identifying cell divisions in 3D and over time, as well as other secondary features such as the quiescent centre, and cell type. This approach has several significant advantages from other studies into plant stress responses as through automation we can consider divisions occurring throughout the root tip and over long periods of time as opposed to focusing on a few cells. This will give us a holistic view of how the root responds to environmental changes. To do this, we need to develop novel machine learning approaches to analyse the large LSFM datasets.
Preliminary results show that root anatomy is significantly altered under low phosphorus conditions. In our static images of 14 day old roots we can observe additional endodermis, cortex and epidermal cells. As a result of the increase in the number of cortical cells many more epidermal cells than usual overlay clefts between cortex cells. It is well established that epidermal cells overlaying the clefts between cortical cells go on to form root hair bearing cells, and consistent with this we see an increase in the number of hair-bearing versus non-hair bearing cells. In bean, it has been well documented that increased root hair density results in improved growth under low phosphorus. Therefore these anatomical changes are likely to cause important environmental adaptations.
Whilst we can generate a hypothesis about what benefit increasing the number of cortical cells may bring, we also see other anatomical changes (such as those relating to the organization of the meristem) that we currently don't understand. Our novel software-based analysis of these datasets will allow us to pinpoint precise changes, such as an increase in number or size of specific cell types, as well as organization of the root structure, by primarily identifying cell divisions in 3D and over time, as well as other secondary features such as the quiescent centre, and cell type. This approach has several significant advantages from other studies into plant stress responses as through automation we can consider divisions occurring throughout the root tip and over long periods of time as opposed to focusing on a few cells. This will give us a holistic view of how the root responds to environmental changes. To do this, we need to develop novel machine learning approaches to analyse the large LSFM datasets.
Planned Impact
The project aims to deliver three main outputs:
1) A novel software approach to locating and labelling features of interest - eg. cell divisions - in large, 3D timeseries datasets.
2) An answer to the biological question, where do anatomical changes originate in response to phosphorus depletion in Arabidopsis?
3) Light sheet protocols for imaging live Arabidopsis over long periods of time
A further output, not on the critical path, is the release of the holder design. It may be possible to commercialise some aspects of this, and we will investigate opportunities as they develop.
Who will benefit from these outputs, and how?
Plant biologists (Academic/Industry):
The new software tool (3) will directly impact lab biologists working with similar light sheet datasets. Previous software tool releases to the biological community has seen a large worldwide uptake (e.g. over 4000 download for a previous tool, RootTrace), including evidenced use by industry.
It is anticipated that other researchers will use the software with similar data, but to answer different biological questions, without the software requiring major coding amendments.
Plant biologists around the world, will have access to the knowledge gained in answering the question in deliverable (2) above, and protocols in (3).
The tool will support the future concept of a high throughput phenotyping platform involving the light sheet microscope. Potential future enhancement of the microscopy system with robotics will demand high throughput analysis. This could be used to uncover new adaptive responses that are beneficial under other environmental stresses. Therefore, our software of use to industrial sectors developing such enhancements, as well as forming the basis of future grant applications to national and European funding agencies.
Computer scientists and Image analysts (Academic/Industry):
Importantly computer scientists will have access to the source code behind the tools, allowing them to alter their function or extend their use to new datasets.
We will foster new collaboration between computer scientists working in this field by inviting them to the proposed workshop on "Developing Software for the Biological Sciences".
The two PDRAs on this project will gain interdisciplinary skills. This will leave them well placed for employment in UK systems biology groups, for example, both within academia and the commercial sector.
Industrial input (industry):
As we are already working with Leica to develop the environmental imaging components and profusion system, we are well placed to discuss commercialising aspects of the software. Whilst the main code will be open source and available to all for free, opportunities such as software training sessions, or the development of specific plugins may be appropriate to charge for.
Crop breeders (Industry):
Our preliminary results on the effects of phosphate deficiency in cortical cells, suggest the adaptive response in increased cortical cell number is specific to the Brassicaceae family. This includes important crop plants grown in the UK, such as rapeseed, cabbage, broccoli and turnip. In the long term, this work could help farmers to uncover new adaptive responses that allow growth of plants in soils with lower phosphate. These responses could be targeted by breeders to select for crops that require less intensive use of fertilizers in the UK and abroad. In the short term, we use the University's strong connections with commercial breeders to share knowledge.
Consumers (Public):
Ultimately, advances such as this will help increase crop production efficiency and hence may have an impact on food supply and prices in the long term future.
General Public:
We plan to publicise the machine learning aspects of this project. Machine learning has seen increased publicity recently, and we can capitalise on that interest to demonstrate the advantages of such approaches to improve food security
1) A novel software approach to locating and labelling features of interest - eg. cell divisions - in large, 3D timeseries datasets.
2) An answer to the biological question, where do anatomical changes originate in response to phosphorus depletion in Arabidopsis?
3) Light sheet protocols for imaging live Arabidopsis over long periods of time
A further output, not on the critical path, is the release of the holder design. It may be possible to commercialise some aspects of this, and we will investigate opportunities as they develop.
Who will benefit from these outputs, and how?
Plant biologists (Academic/Industry):
The new software tool (3) will directly impact lab biologists working with similar light sheet datasets. Previous software tool releases to the biological community has seen a large worldwide uptake (e.g. over 4000 download for a previous tool, RootTrace), including evidenced use by industry.
It is anticipated that other researchers will use the software with similar data, but to answer different biological questions, without the software requiring major coding amendments.
Plant biologists around the world, will have access to the knowledge gained in answering the question in deliverable (2) above, and protocols in (3).
The tool will support the future concept of a high throughput phenotyping platform involving the light sheet microscope. Potential future enhancement of the microscopy system with robotics will demand high throughput analysis. This could be used to uncover new adaptive responses that are beneficial under other environmental stresses. Therefore, our software of use to industrial sectors developing such enhancements, as well as forming the basis of future grant applications to national and European funding agencies.
Computer scientists and Image analysts (Academic/Industry):
Importantly computer scientists will have access to the source code behind the tools, allowing them to alter their function or extend their use to new datasets.
We will foster new collaboration between computer scientists working in this field by inviting them to the proposed workshop on "Developing Software for the Biological Sciences".
The two PDRAs on this project will gain interdisciplinary skills. This will leave them well placed for employment in UK systems biology groups, for example, both within academia and the commercial sector.
Industrial input (industry):
As we are already working with Leica to develop the environmental imaging components and profusion system, we are well placed to discuss commercialising aspects of the software. Whilst the main code will be open source and available to all for free, opportunities such as software training sessions, or the development of specific plugins may be appropriate to charge for.
Crop breeders (Industry):
Our preliminary results on the effects of phosphate deficiency in cortical cells, suggest the adaptive response in increased cortical cell number is specific to the Brassicaceae family. This includes important crop plants grown in the UK, such as rapeseed, cabbage, broccoli and turnip. In the long term, this work could help farmers to uncover new adaptive responses that allow growth of plants in soils with lower phosphate. These responses could be targeted by breeders to select for crops that require less intensive use of fertilizers in the UK and abroad. In the short term, we use the University's strong connections with commercial breeders to share knowledge.
Consumers (Public):
Ultimately, advances such as this will help increase crop production efficiency and hence may have an impact on food supply and prices in the long term future.
General Public:
We plan to publicise the machine learning aspects of this project. Machine learning has seen increased publicity recently, and we can capitalise on that interest to demonstrate the advantages of such approaches to improve food security
Publications

Janes G
(2018)
Cellular Patterning of Arabidopsis Roots Under Low Phosphate Conditions.
in Frontiers in plant science

Khan FA
(2020)
Volumetric Segmentation of Cell Cycle Markers in Confocal Images Using Machine Learning and Deep Learning.
in Frontiers in plant science

Orosa-Puente B
(2018)
Root branching toward water involves posttranslational modification of transcription factor ARF7
in Science

Ovecka M
(2018)
Multiscale imaging of plant development by light-sheet fluorescence microscopy.
in Nature plants


Pound MP
(2018)
Erratum to: Deep machine learning provides state-of-the-art performance in image-based plant phenotyping.
in GigaScience

Pound MP
(2017)
Deep machine learning provides state-of-the-art performance in image-based plant phenotyping.
in GigaScience
Description | We have analysed 3D geometry of Arabidopsis roots grown on low phosphate. We see a proliferation of cortex cells and no change in epidermal cell number [1]. This leads to modulation of root hair patterning. Note currently this is manual analysis using 3D light sheet data; automated cell scale analysis is still in development. However, protocols for the growth and imaging have been developed as part of this award. We have also developed novel software annotation tools for 3D/4D data, to enable annotation of events in 3D light sheet or confocal data, to train AI systems. This presents data in an easy-to-annotate orthogonal format, allowing spherical-like objects (e.g. plant cell nuclei) to be easily annotated in 3D or 4D datasets. This has been used as part of a paper which explored machine versus deep learning analysis of 3D datasets [2]. We have assembled 3D and 4D datasets of lateral root development on the light sheet microscope, and have sections of the data annotated to provide input to deep networks. We have built a deep network approach to detect and label cell cycle markers, comparing manual annotation with annotation drawn automatically from biological markers; this is currently in a paper draft. We have acquired funding ("Data CAMPP - innovative training in data capture, analysis and management for plant phenotyping", 2021) from UKRI to train bioscientists in data science and image analysis techniques, with a focus on plant phenotyping. This award was one of several which built the foundation for that project. [1] Janes G, von Wangenheim D, Cowling S, Kerr I, Band L, French AP, Bishopp A. (2018). Cellular Patterning of Roots Under Low Phosphate Conditions.. Frontiers in plant science, pp. 735 [2] https://doi.org/10.3389/fpls.2020.01275 |
Exploitation Route | Demonstrates practical protocols for imaging Arabidopsis plants under varying environmental conditions on the LSFM. This will be used to underpin timeseries datasets in the future. Our 3D/4D nuclei marker annotation tool will likely be of use in other labs in academia and industry. We have received a lot of interest in the deep learning (AI) development we are carrying out in general, leading to further deep learning projects both for microscopy images and plant image analysis in general. We have collected and annotated datasets of plant growth on the LSFM, which may also be of use in the machine learning domain as training data for new techniques. |
Sectors | Agriculture Food and Drink Digital/Communication/Information Technologies (including Software) Other |
Description | As mentioned in Further Funding, this work is partially responsible for us securing significant further funding from industry in the domain of using AI for plant image analysis. |
First Year Of Impact | 2016 |
Sector | Agriculture, Food and Drink |
Impact Types | Economic |
Description | Data Intensive Bioscience - Expert Working Group |
Geographic Reach | National |
Policy Influence Type | Membership of a guideline committee |
Description | EMBL-EBI BioImaging Ecosystem group |
Geographic Reach | Europe |
Policy Influence Type | Membership of a guideline committee |
Description | Membership of the Joint BBSRC-MRC Supercomputing task force |
Geographic Reach | National |
Policy Influence Type | Participation in a guidance/advisory committee |
Description | Part of UKRI AI Review - AI for Science & Research Workstream |
Geographic Reach | National |
Policy Influence Type | Contribution to a national consultation/review |
URL | https://www.ukri.org/news/ukri-artificial-intelligence-regional-workshops/ |
Description | Data CAMPP (Innovative Training in Data Capture, Analysis and Management for Plant Phenotyping) |
Amount | £808,900 (GBP) |
Funding ID | MR/V038850/1 |
Organisation | United Kingdom Research and Innovation |
Sector | Public |
Country | United Kingdom |
Start | 04/2021 |
End | 04/2023 |
Description | Deep learning for Image Analysis |
Amount | £350,000 (GBP) |
Organisation | Syngenta International AG |
Sector | Private |
Country | Switzerland |
Start | 09/2018 |
End | 10/2021 |
Title | New destination plasmids for observation of multiple fluorescent markers in roots |
Description | We have created a set of multiple markers that highlight specific tissues in plants and couple GFP with a red plasma membrane marker. This is built around the greengate cloning system, and involved the creation of bespoke destination vectors as well as individual entry clones. |
Type Of Material | Biological samples |
Year Produced | 2018 |
Provided To Others? | No |
Impact | This is fantastic for creating live movies of gene expression via light sheet microscopy. We aim to submit a publication with this information within the next 6 months. The plasmids were created as part of BB/L023555/1 but the imaging and data acquisition is part of this project. |
Title | Confocal cell cycle marker dataset adapted for machine learning |
Description | Confocal cell cycle marker dataset adapted for machine learning. With annotated ground truth. Provided in formats to support machine learning. |
Type Of Material | Database/Collection of data |
Year Produced | 2020 |
Provided To Others? | Yes |
Impact | Supported the paper https://doi.org/10.3389/fpls.2020.01275 |
URL | https://gitlab.com/faraz.khan1/volumetric-segmentation-of-confocal-images |
Description | Working with Syngenta |
Organisation | Syngenta International AG |
Department | Syngenta Ltd (Bracknell) |
Country | United Kingdom |
Sector | Private |
PI Contribution | As part of work presented from this and other plant-science related deep learning projects in the lab, we now have a close working relationship with Syngenta, and indeed now have a running industrially funded project with them (as reported in the Further Funding section). We have provided presentations at Syngenta, and attended scientific meetings hosted by Syngenta to discuss our machine and deep learning work with them and other partners. We have co-run image analysis training schools with representatives from Syngenta. We have an active, industrially-funded project with Syngenta where we are developing deep learning tools and apps for a variety of plant science applications. |
Collaborator Contribution | Providing plant science challenges to apply our deep learning solutions to. Access to data. Access to expertise. |
Impact | Syngenta-funded research grant (please see Further funding section). This is multidisciplinary between plant science and computer science. |
Start Year | 2016 |
Title | "Orthogonal Pixels" annotation tool |
Description | A plugin for ImageJ/FIJI which allows easier annotation of confocal/lightsheet datasets by providing orthogonal views of a 3D/4D dataset, allowing the marking of spherical-like objects e.g. nuclei. |
Type Of Technology | Webtool/Application |
Year Produced | 2020 |
Open Source License? | Yes |
Impact | Can be used to annotate data to support machine learning, e.g. it was used in https://www.frontiersin.org/articles/10.3389/fpls.2020.01275/full |
URL | https://www.frontiersin.org/articles/10.3389/fpls.2020.01275/full |
Description | Deep learning of plants |
Form Of Engagement Activity | A talk or presentation |
Part Of Official Scheme? | No |
Geographic Reach | Regional |
Primary Audience | Undergraduate students |
Results and Impact | Invited talk at the Aberystwyth University, Monday, 19th March 2018 Abstract: Deep learning seems to be succeeding in bringing AI-based methods into the limelight, particularly for challenging, image-based datasets. In this talk, I will discuss our use of deep learning at Nottingham to phenotype various image features in plant-based data - from microscope images, to crops in the field (nearly). Deep learning will be introduced, from an image analysis perspective, and our experience of using it in a variety of plant-related domains will be discussed, including some challenges we have faced in the process. We have found it to work extremely well where traditional image analysis approaches would struggle, but this comes at a cost, and with some drawbacks which I will highlight. I presented in particular the light sheet data sets in the context of challenging data, and presented some ideas about how to avoid pitfalls of deep learning. |
Year(s) Of Engagement Activity | 2018 |
Description | Deep learning: taking a deeper look at biological images |
Form Of Engagement Activity | A talk or presentation |
Part Of Official Scheme? | No |
Geographic Reach | National |
Primary Audience | Postgraduate students |
Results and Impact | Invited talk at the Plymouth Imaging Network at the University of Plymouth. I overviewed the pros and cons of machine learning when working with biological images, and discussed approaches we are taking to analyzing events in the light sheet data |
Year(s) Of Engagement Activity | 2018 |
URL | https://twitter.com/plymimagingnet?lang=en |
Description | Engagement with the plant phenotyping community via the PhenomUK network |
Form Of Engagement Activity | Participation in an activity, workshop or similar |
Part Of Official Scheme? | No |
Geographic Reach | National |
Primary Audience | Other audiences |
Results and Impact | Networking and knowledge sharing between a diverse audience of researchers and industry working in the area of plant and crop phenotyping in the UK. |
Year(s) Of Engagement Activity | 2019 |
URL | https://www.phenomuk.net/ |
Description | Invited talk: INDEPTH meeting Cost Action meeting, Prague, Czech Republic |
Form Of Engagement Activity | A talk or presentation |
Part Of Official Scheme? | No |
Geographic Reach | International |
Primary Audience | Postgraduate students |
Results and Impact | Invited talk: INDEPTH meeting Cost Action meeting, Prague, Czech Republic. Presentation of the deep learning approaches used in this project, and the work so far on light sheet imaging. Networking with other groups discussing potential of deep learning on their projects. |
Year(s) Of Engagement Activity | 2019 |
URL | https://www.brookes.ac.uk/indepth/news/prague-indepth-abstract-book/ |
Description | Plant Organ Growth Symposium 2019, Bordeaux, France. April 2019. Poster presentation: Optimising light sheet microscopy for automatic annotation of anatomical changes in growing Arabidopsis roots |
Form Of Engagement Activity | Participation in an activity, workshop or similar |
Part Of Official Scheme? | No |
Geographic Reach | International |
Primary Audience | Other audiences |
Results and Impact | Poster presentation at a subject-specific international symposium: "The aim of this meeting is to bring together top experts in the field of organ growth regulation in order to share recent advances and expertise on this exciting topic. Our objective is to discuss state of the art knowledge on growth regulation at multiple spatial scales." |
Year(s) Of Engagement Activity | 2019 |
URL | https://symposium.inrae.fr/pogs2019/PROGRAM |
Description | Talk at BBSRC DTP training school: BBSRC Biomolecular Skills School 2019 |
Form Of Engagement Activity | A talk or presentation |
Part Of Official Scheme? | No |
Geographic Reach | Regional |
Primary Audience | Postgraduate students |
Results and Impact | Gave presentation at the BBSRC Biomolecular Skills School 2019, Nottingham, 8/1/2019. Introduced image analysis for the biosciences, and discussed the machine deep learning techniques which are the basis for this project, and also discussed the light sheet microscope and light sheet data properties. Audience was about 50 postgraduate DTP students. |
Year(s) Of Engagement Activity | 2019 |
Description | Techniques workshop: Confocal and light sheet microscopy, for PhD students |
Form Of Engagement Activity | Participation in an activity, workshop or similar |
Part Of Official Scheme? | No |
Geographic Reach | Local |
Primary Audience | Postgraduate students |
Results and Impact | This was a techniques workshop hosted by Tony Bishopp (Co-I) at Nottingham. It explored confocal and light sheet microscopy, covering the theory, the equipment we have and some examples of use in research. |
Year(s) Of Engagement Activity | 2019 |