“PERspectives” can have a wide range of implementations, some of which are examined and proposed in this section. From medicine, to the environment and even energy consumption, it can be the foundation of groundbreaking advances in each of the aforementioned fields and even more. As a universal tool used by future iGEM teams and generally researchers, PERspectives can provide enhanced genetic stability, reliability and long-term functionality in a variety of projects. Our system utilizes a cell consortia format in order to create a novel biosensor that allows multiple inputs recognition and activation according to their significance. The work compartmentalization to multiple bacterial populations reduces the metabolic burden that each cell would carry. Additionally, by using the quorum sensing mechanism for communication among bacterial populations, misinterpretation of the output resulting from cellular noise is reduced. The reduction of biological noise is the major advantage of “PERspectives”, in combination with its various potential implementations. Our system can be versatile to different problems, since it can be used for different genes, as presented in Figure 1. Some of the applications that will be presented in the section below can be directly plugged into our system and some of them require further editing as described in each case.
Our project’s goal is to create a universal tool that will be able to solve real-time multifactorial problems. Our bacterial perceptron will have special value in cases where the problem is too hard to reach, taking for example the human body.
What about a living therapeutic that could detect the problem in real-time and decide if a medicine is needed or not?
Non-pathogenic bacteria have been proposed and even tested in clinical trials as a potential drug delivery system. The iGEM Competition has even dedicated a project track named “Living Therapeutics”, including projects aiming to utilise a living organism as a drug delivery system. Living therapeutics are without a doubt a popular research field and the engineered genetic circuits are constantly evolving.
Nowadays, the field of precision medicine and targeted drug delivery is flourishing, promising more effective therapies and fewer side effects for patients.
How could our project ideation and design be incorporated to take this field to the next level?
As far as cancer is concerned, it is proven that tumors are not homogenous. This diversity often makes one-drug therapeutic schedules ineffective. Every tumor type includes various subtypes and even the environment inside a single tumor shows variations.
One important classification of tumors refers to the activity of the immune system inside them. Roughly explained, tumors that host an activated immune system and inflammation are characterized as immune “hot” tumors, whereas tumors with suppressed immune cells that are not inflamed are characterized as immune “cold” tumors. The immune status of a specific tumor plays a vital role regarding its response to immunotherapy, and more specifically to Immune Checkpoint Inhibitors (ICIs). These are monoclonal antibodies that bind and block immune checkpoints, like PD-L1 and CTLA-4, which are naturally expressed by cells to hinder immune systems’ overactivity. In some cancer types, these inhibitors are overexpressed and thus cancer cells escape immune surveillance. Immune checkpoint inhibitors block immune checkpoint’s activity and permit immune cells to attack the malignant cells. However, tumors are often immune excluded or exhausted, meaning that they do not host inflammatory immune cell phenotypes. Even though these tumors might express immune checkpoint inhibitors, they do not respond sufficiently to ICIs. One promising strategy is to fire up cold tumor microenvironments by combining these immunotherapeutic drugs with proinflammatory agents, that will promote infiltration of the tumor by immune cells and then, their antitumoral activity will be enhanced by the action of ICIs [12,13].
Tumour-invading bacteria are pathogenically incompetent bacterial strains, which are able to reside and multiply in the necrotic tumour environment and have been described as a potential drug delivery system to tumours [14,15]. They could potentially take the mentioned “cocktail” strategies [16] to the next level. Probiotics have been engineered to express immune checkpoint inhibiting nanobodies and release them inside tumors after synchronized lysis when they have reached a certain density, which is sensed via quorum sensing signals. In the same research, bacteria that express a proinflammatory agent were intratumorally injected together with the blockade nanobodies producing bacteria to enhance their antitumor action, similarly with the drug cocktail strategies but promising less side effects since the drugs are released inside the tumor and not systemically delivered[17]. A perceptron-inspired approach could revolutionize this strategy. The suitable therapeutic genes will be engineered into receiver cells for both hot and cold tumors. Sender cells will be engineered to sense signs of active inflammation inside the tumor and respond by producing a quorum sensing molecule. Receiver cells according to the local summation of the lactone will switch their response between the production of a proinflammatory agent or an immune blockade inhibiting nanobody. Such a system, rather complex, might provide an immunotherapeutic strategy able to adapt to each tumor's immune status and even to the immune status of each tumor’s site (necrotic immune inactive center versus a more immune active periphery). Analyses of the different metabolism between macrophage phenotypes ( inflammatory, classically activated M1 phenotype and immunosuppressive, tumor promoting alternatively activated M2 phenotype[18,19]), very important players in the tumor microenvironment, and mathematical modeling of tumor-immune cells interactions and of the different intratumoral macrophage subtypes regarding the production of small molecules already exist[20,21]. Bacteria capable of recognising small products of inflammatory cells, such as nitric oxide [22] and even the proteins Interferon-γ and Tumor Necrosis Factor-a [23] have been engineered and the collection of new bacterial sensing modules grows.
All the above are paving the way towards datasets of biochemical analyses and markers that could be used to simulate and develop a probiotic perceptron that would be able to target solid tumors, classify them and produce the more suitable therapeutic substance.
Beyond this example, such neural-like inspired biological systems would revolutionize therapies on the grounds that many diseases are multifactorial and many times difficult to be distinguished from one another. Incorporating weighted processing of inputs would lead to efficient theranostic systems promising more effective therapies and less side effects.
The human body is cohabited by microorganisms, this symbiosis is frequently referred to as the microbiome. The most studied microbiome is the one of the human gut, known as the gut microbiome[1]. Up to 1000 different bacterial species are found in the human gut. Each one of these species is essential for the function of the human gut, as it contributes to processes such as food digestion and immunity. In case that the balance between the different species gets distracted (called dysbiosis), multiple diseases could emerge. Gut disbalances are linked to diseases such as inflammatory bowel diseases (IBD), irritable bowel syndrome (IBS), diabetes, obesity, cancer, and cardiovascular and central nervous system disorders. [2]. According to the World Health Organization (WHO), probiotics are “live strains of strictly selected microorganisms which, when administered in adequate amounts, confer a health benefit on the host” [3,4]. In order to help with the treatment of gut disbalances we plan to design a bacterial perceptron that will have miRNAs or gDNAs detected in the human gut as inputs, the levels of which have been linked to gut disbalances in individuals with the disease. Regarding the gDNAs,
Identifying the cause in acute infections constitutes a common medical problem. The symptoms' similarity between viral and bacterial infections makes it difficult for health specialists to identify the optimal treatment in urgent situations. This situation often leads to the overuse of antibiotics [24] and contributes to antibiotic resistance. Research has shown that discriminating between viral and bacterial infections requires simultaneous testing for multiple factors, some of which belong to the individual's immune response and not to the pathogen itself [25]. What if we had a bacterial artificial intelligence classification system or a cell-free system following a similar principle that would simultaneously test for three different factors and classifiy for the best treatment for the patient? Our project “Perspectives” seems like a perfect candidate for this task because it can evaluate different inputs, assigning each of them the appropriate significance depending on if it recognizes the bacterial or the viral infections [26]. Three molecules that could be recognized according to literature are: NF-related apoptosis-inducing ligand (TRAIL) [27] which is upregulated in viral infection patients, IP-10 which is higher in viral patients compared to bacterial patients, and CRP (C-Reactive Protein) which is mainly encountered in bacterial infections but has been correlated to some viral strains' infection as well. In this way, our circuit could be used as a diagnostic tool, taking into account that huge research has been done on how human samples affect bacterial biosensors [28].
Poor control of type one diabetes may lead to diabetic ketoacidosis, which could be lethal. Diabetic ketoacidosis is characterized by pH lower than 7.35, high blood sugar levels and low insulin levels. Following the lead of David Ausländer [29] who designed a pH-biosensor using microencapsulated HEK293 cell line batches, a biological perceptron embedded in a living therapeutic could incorporate multiple input signals, like the aforementioned pH level, glucose level and insulin level to classify in real time the patient's status and produce a suitable amount of insulin.
As it’s known, developing new, efficient, and safe drugs is extremely time-consuming, expensive, and can take even 12 years on average. [30] The first stage of that discovery begins in the laboratory. According to the FDA, thousands of compounds may be potential candidates for development as a medical treatment at this stage. After early testing, however, only a small number of compounds seems promising and calls for further study [31]. Also, many diseases have a very complicated phenotype, often related to a complicated genotype. Therefore, the existence of a biosensing artificial neural network system could be extremely beneficial to the Research and Development (RnD) phase. [32] In the last few years, a lot of research has been done to reveal the connection between cancer and miRNA profiles[33,34]. According to them, many miRNAs are associated with cancer types, but some miRNAs have a stronger correlation to a specific type of cancer. Also, the ANNs appear as a promising tool for pharmaceutical scientists [35]. What about a biological system that could classify the new tested molecules, according to the provoked change in the phenotype of cell lines, between the possible new drugs and the molecules that are not effective or safe?
Our system’s design could play a leading role in the RnD domain in this framework. Let’s take as an example, colorectal cancer. Firstly, we
could gather a significant amount of patient samples as well as a significant amount of cancer-free individual samples and analyze them in
quantity and quality for miRNAs. After that, with the use of an Artificial Neural Network, we can select the most important miRNA molecules
and calculate the weights of all of them[36]. Then, we could translate those weights into RBS strengths by using the same weight promoters.
By using that knowledge, we could design a novel diagnostic tool for colorectal cancer. The most fascinating part though could be if we also
collect a significant number of samples of patients during the course of therapy and analyze them also with the usage of an Artificial Neural
Network [37].
After that, the potential candidate molecules will be added to colorectal cancer cell lines. By giving to our system the lysate of these cells or a portion of their medium we could have a classification between promising molecules and molecules with no efficiency or safety. In the framework of the above experiments, the scientist can use either mammalian cells [38] or bacterial cells that detect miRNAs. Toehold switches could be engineered to bind to input miRNA and respond in a weighted manner by an output production. The same principle of work could be used also for the discovery of new antidepressants [39], new anticancer drugs for Hodgkin lymphoma [40]and new anticancer drugs for breast cancer [41,42].
According to CDC: “Pharmacogenomics is an important example of the field of precision medicine, which aims to tailor medical treatment to each person or to a group of people” [43]. By pharmacogenomics, scientists reduce the likelihood of adverse drug reactions and optimise therapeutic efficacy [44]. As an example, in cancer research, several in vitro investigations describe altered patterns of miRNA expression in cells resistant to anticancer drugs. For instance, in a study by Berkers et al., they analysed miRNA expression in clear cell renal carcinoma tissue samples from patients who received sunitinib. The results showed a significant decrease in the expression of miR-141 in patients with progressive disease (that indicates maybe a high weight of that miRNA for a perceptron algorithm). Also, a subset of other miRNAs, differentially regulated between the patients who responded to sunitinib treatment versus the non-responders, included miR-520g, miR-155, miR-526b (all upregulated) and miR-144 and miR-376b (both downregulated) (that indicates maybe a medium or low weight of those miRNAs for a perceptron algorithm).
Some of the mechanisms that have been proposed for that resistance are presented below. MicroRNAs contribute to cancer drug resistance via :
In conclusion, by analysing samples of patients with miRNAs connected with the aforementioned resistance pathways for specific drugs and using a system similar to “PERspectives” we could predict if a patient will be cured by a specific medication or not[45]. In the framework of the above experiments, the scientist can use either mammalian cells [46] or bacteria cells that detect miRNAs. In either case, toehold switches appear a very promising structure for the sensing part of miRNAs. In this way, scientists could save a lot of people or improve their quality of life.
Inspired by the project S-POP [47], we propose a future implementation in order to make the future goals of S-POP become possible, by implementing our bacterial perceptron to detect water Persistent Organic Pollutants (POPs). The senders (
Another implementation of our project could be the evaluation of soil health. Soil health is very important in means of plant health and crop yield, but also affects human and animal health [50]. There are different kinds of indicators to assess soil health: Chemical Indicators, Physical Indicators and Biological Indicators. Our system could be redesigned to detect such indicators: Nitrogen found in the soil via protein degradation is an indicator of its availability for plant assimilation, there are bacterial nitrogen transporters that could be used to set Nitrogen as an input to our system [51],[52],[53]. The pH [54] of the soil is also very important in determining soil health. There are already several pH responsive promoters that could be incorporated in our system to evaluate the pH. [55]. Last but not the least, salinity is important for some environments and could be detected by salt-activated promoters engineered in our system [56]. Our system could be adjusted for the weights of each input to respond in several soil compositions. This will be a huge help for people that work in the agricultural sector to optimize their crop production.
“PERspectives” comes as an alternative, environmentally friendly method for implementing machine learning algorithms. As it is well-known, s
tate-of-the art computational systems require a large amount of data in order to be modelled and fine-tuned. Currently, the data are stored in
data centres. The operation of such centers is estimated to handle enormous amounts of data every day, as in the case of IBM’S The Weather
Company [57], which processes about 400 Terabytes of data per day in order to predict the weather in advance around the globe. Such data
mining, storage and processing flow corresponds to energy consumption of around 80MegaWatts-more than enough to power 80.000 US Households.
Another example of effective but energy-costly use of machine learning and-in particular-neural networks comes in the training process of
Natural Language Processing Software [58], where the usage of big data and specialized hardware leads to a significant increase in Carbon
Footprint Emissions. As an example, training the big “Transformer”[59] of Google-a novel neural network architecture for machine translation
tasks-required a total of 100.000 steps or 3.5 days and 8 NVIDIA P100 GPUs, leading to an estimate of 626.000 lbs of CO2 emission. As it can
be shown in Figure 1, such an emission total is 5 times more than the average emissions of a conventional car!
Concluding, there is a vast field for implementation for our project. In our opinion the first steps that should be taken in order for
some of these applications to come true are:
1. Introducing more different inputs and more layers to the system.
2. Introducing other intermediate molecules to reduce cross-talk in case of living therapeutics.
3. Introducing the system to yeast cells using the pheromone system of communication. [63]
4. Introducing the system to human cell lines using the tryptophan synthase for communication. [63]
5. Incorporating the kill switch and the negative weights as mentioned in the safety section of the implementations.