Goal: Eliminate Competition between Food & Clean Air
We need food and air to live. Photosynthesis provides both.
Farms provide food; forests provide clean air. But, forests and farms are in accelerating competition for land - together they make up a 57% (and growing) share of Earth's total land area. The rest is desert, tundras and glaciers - not friendly places.
On balance, forests have been losing out to more farms. These land use changes account for 1/3 of human emissions. Despite the move from forests to farms, we're still on pace to underserve demand in rice & wheat by 30% within 30 years.
Photosynthesis may be life's engine, but that engine only operates at 0.1-2% efficiency. All life on earth - animals, humans, plants, microbes etc - is hindered as a result. To make food abundant and control CO2 levels, 10+% efficiencies are required. That is our goal.
Table of Contents
We begin with a history of food & CO2 on Earth in four charts. Food is created by taking CO2, keeping the C and releasing the O2. So, naturally, their stories are deeply entwined.
We then outline how we plan to achieve 10% efficiencies by co-evolving photosynthetic genetics and growing environment.
Over the last 1000 years, the average person has become 40x more wealthy. Despite a 100x increase in the number of people, the price of wheat has fallen by 5x. However, the price has stopped dropping over the last twenty years.
Food production is number of acres times production per acre.
We might think to double our allocation of land used for agriculture. We can't.
That leaves increasing production per acre. While real yields have continued a slow 0.3% increase year over year. However, max potential yields per acre have stagnated since 1982. Since we can't add land, we must increase output per acre.
The problem is not input energy. Even a single photon has enormous energy relative to biological requirements. A photon of red light - a low energy wavelength - carries 3 ATPs worth of energy.
1000x more solar energy hits a bacterial cell than the bacterial cell requires to live.
It's unclear what the optimal level of CO2 on Earth is.
In the last two hundred years, global co2 has grown at an all-time fast rate due to human emissions. The projections are not bright.
Agriculture emerged ~10,000 years ago - we believe an increase of global CO2 from 230ppm to 300ppm was required to make farming feasible.
So, optimal CO2 levels are not 0 ppm.
Earth CO2 levels have varied enormously. Improvements in photosynthesis can dramatically change CO2 levels and CO2 levels are a principal evolutionary driver for photosynthetic organisms.
Around 400 million years ago, CO2 levels plummetted from 150,000 ppm to 10,000 ppm very quickly. While no causal link has been demonstrated, this reduction coincides with the evolution of carbon dioxide storage in photosynthetic bacteria.
100 million years ago, CO2 levels started to fall significantly. As a result, a new, more efficient photosynthesis (C4) pathway evolved to handle low CO2.
Many plants are not optimized for these higher CO2 levels reducing their capacity to remove CO2 levels and grow quickly.
Fixing Photosynthesis' Leaky Funnel
Why bother with environmental control and especially environmental optimization when the goal is accelerating photosynthesis genomically?
Environments and genomes are inextricably linked. Environments are a primary selection mechanism for genomes and can change the genome itself. Genomes are tailored to environments and small changes in photosynthetic genomics change the environment.
Determining optimal environmental conditions for a given genotype leads to better understanding of the phenotype. For instance, a 59bp change in a cyanobacterial strain led to a 3x increase in photosynthetic efficiency only under 1000mmol/m^2/s of light. By probing light levels, the improvement was traced back to the electron transport chain.
The physiology of a plant is changing every second. It's optimal light cycle is 10 microseconds on / 10 miliseconds off. Instead, we learn a neural net that inputs info about the plant and determine the appropriate environment.
So, genomes and environments should be co-optimized. In alternating cadence, the genome is optimized to suit the environment and the environment is optimized to suit the genome.
Photosynthesis as currently constructed is very wasteful. It operates at 0.1-2% efficiency.
Parts of photosynthesis are remarkably efficient. The solar cell has 83% efficiency in converting light energy into electrical energy. Unfortunately, converting electrical energy into carbohydrate energy is very inefficient.
Photosynthesis can be decomposed into 5 major subparts. Capturing photons, converting photons into electrons, transporting electrons, converting electrons into energy, and creating sugars.
Photosynthesis takes place within cells on plants' leaves.
Throughout the day, photons hit the plants leaves.
Within these halls of antennae, incident photons generate electrons via the photoelectric effect.
Generated electrons are then immediately drawn into a biological wire (literally proteins with attached conductive metals).
These electrons provide the energy required to create life's intermediate energy forms (ATP & NADPH).
CO2 is breathed in and harvested energy is used to chain long carbon chains of sugars for use by the plant.
The overall process has 0.5-2% efficiency. Some transitions are remarkably efficient (83% quantum yield of photons to electrons); while others (Rubisco turnover rate) are not. In any chain process, addressing the current limiting factor yields a new one. In our repairman's tour of photosynthesis, we'll detail potential fixes; refactor for each transition as well as consider possibilities of cutting entire steps.
Below is a "sales" funnel of energy. At each step, some amount of energy is lost. Overall, we start with 1000 Joules and end up with 13 Joules!
The current approach to improving plants requires too much watching grass grow. Phenotyping the results of a genomic change requires waiting until the plant is fully grown and most measurements are currently done by hand and written down on paper.
We are building a pipeline from FASTA files of genomes to deployment in our customer's facilities. Every step is automated to create a high throughput of candidates. Every step is expressed in software so that machine learning algorithms can filter candidates for the next change.
Real-world data and experiments is required. However, the more build-test-iterate cycles that can be executed in tissue culture (plant cells) vs full-grown plants the faster improvements will be made.
There are three main stages. Plant genomes, plant cells and full-grown plants. Plant genomes are added to plant cells which are then grown into full plants via tissue culture.
Entirely in software, hundreds of thousands of candidate genetic edits are generated. Machine learning algorithms then accept (green) or reject these candidates. Accepted genes are then expressed in cells (green arrow).
The build-test-iterate pipeline has historically been futile in plants since the cycle requires 3+ months to wait for the plant to grow. Instead, we incorporate detailed photosynthetic efficiency measurements at the tissue culture stage.
Convolutional neural nets then evaluate the photosynthetic efficiency of every cell in every plant in order to accept or reject candidates for deployment in full-scale production.
Rejects again provide the data for training models at the tissue culture and gene candidate stage so that hopefully more can be evaluated and rejected on the day timeframes of tissue culture vs months timeframes of full plants.
Overall, at every stage candidates are evaluated by machine learning models and either progressed (green) or rejected (red). Every reject yields data (blue) with which to train the prior step. By automating, the flywheel moves more quickly; by training accurate models, the flywheel becomes more efficient.