Briefing Document: Pan-Antiviral Activity of the Hsf1-sHsp Cascade in Mosquito Cells

Source: Qu, J., Schinkel, M., Chiggiato, L., Machado, S. R., Overheul, G. J., Miesen, P., & van Rij, R. P. (2025). The Hsf1-sHsp cascade has pan-antiviral activity in mosquito cells. Nature Communications Biology, 8(1), 74. https://doi.org/10.1038/s42003-024-07435-4

Date: Received - 05 July 2023 | Accepted - 20 December 2024 | Published - 25 January 2025

1. Executive Summary:

This research paper investigates the antiviral immune responses in Aedes aegypti mosquitoes, a primary vector for several arboviruses. Through multi-omics data integration, the study identifies a novel early-responsive antiviral cascade involving heat shock factor 1 (Hsf1) and a cluster of eight small heat shock protein (sHsp) genes. This "Hsf1-sHsp cascade" demonstrates pan-antiviral activity against multiple arboviruses, including chikungunya, Sindbis, dengue, and the insect-specific Agua Salud alphavirus in Ae. aegypti cells, and also against chikungunya and O’nyong-nyong viruses in other mosquito species (Aedes albopictus and Anopheles gambiae cells). The findings suggest that Hsf1 could be a promising target for developing novel intervention strategies to limit arbovirus transmission by mosquitoes.

2. Main Themes and Important Ideas:

• Identification of a Novel Antiviral Cascade: The study reveals a previously uncharacterized early immune response in Ae. aegypti cells triggered by chikungunya virus (CHIKV) infection. This response involves the transcription factor Hsf1 activating the expression of eight sHsp genes located within a single topologically associated domain (TAD) in the mosquito genome.
• "By integrating multi-omics data, we find that heat shock factor 1 (Hsf1) regulates eight small heat shock protein (sHsp) genes within one topologically associated domain in the genome of the Aedes aegypti mosquito."
• Hsf1 Drives Early sHsp Expression: RNA-seq and chromatin profiling data indicate that Hsf1 binds to the promoter regions of these eight sHsp genes, leading to their rapid upregulation upon CHIKV infection. RNAi-mediated silencing of Hsf1 significantly diminished the induction of all eight sHsps, confirming its regulatory role.
• "We found the DNA binding motif of the transcription factor heat shock factor 1 (Hsf1) to be highly enriched at promoter regions of these eight sHsp genes..."
• "Hence, the Hsf1-sHsp cascade functions as an early transcriptional response to CHIKV infection in Aag2 cells."
• The Hsf1-sHsp Cascade Exhibits Anti-CHIKV Activity: Silencing either Hsf1 or all eight sHsp genes using RNAi resulted in elevated CHIKV RNA levels in Ae. aegypti cells, indicating that this cascade plays a role in suppressing CHIKV replication.
• "To explore the antiviral potential of the identified Hsf1-sHsp axis, we silenced Hsf1 using RNAi and observed elevated CHIKV RNA levels at 8 hpi..."
• "...we therefore used two sets of dsRNA targeting the conserved Hsp20 domain to knock down all eight sHsp genes and observed significantly increased CHIKV RNA copies at 8 hpi..."
• Modulation of Hsf1 Activity Impacts CHIKV Replication: Using a pharmacological inhibitor (KRIBB11) of Hsf1 led to increased CHIKV replication, while an activator (hsfa1) resulted in a significant reduction in viral RNA levels. Interestingly, DTHIB, reported as a human Hsf1 inhibitor, acted as an activator in mosquito cells and also repressed CHIKV replication, further supporting the antiviral role of the sHsps.
• "KRIBB11 treatment significantly downregulated most sHsp genes after 12 h...led to more than 5-fold increase of viral RNA levels in Aag2 cells..."
• "Hsfa1 treatment significantly increased Hsf1 expression and strongly activated all eight sHsp genes, leading to about a 25-fold reduction in CHIKV copies at 12 hpi..."
• The Cascade Targets an Early Post-Entry Stage of Viral Replication: Time-of-addition assays with the Hsf1 activator and inhibitor suggest that the Hsf1-sHsp cascade interferes with an early stage of the CHIKV replication cycle, occurring around 4 hours post-infection, after virus entry but before significant RNA replication.
• "When the Hsf activator hsf1a was added at time points before inoculation and up to 4 hpi, CHIKV replication was strongly inhibited..."
• "This indicates that the Hsf1-sHsp cascade interacts with the early stage of CHIKV RNA replication but does not affect subsequent stages."
• Pan-Antiviral Activity Across Multiple Arboviruses and Mosquito Species: The study demonstrates that the Hsf1-sHsp cascade is not specific to CHIKV. Activation of Hsf1 with hsfa1 reduced replication of Sindbis virus (SINV), Agua Salud alphavirus (ASALV), and dengue virus (DENV) in Ae. aegypti cells. Furthermore, activating the cascade showed antiviral effects against CHIKV in Aedes albopictus cells and the RNAi-deficient C6/36 cells, and against O’nyong-nyong virus (ONNV) in Anopheles gambiae cells.
• "Consistently, KRIBB11 treatment led to increased virus replication for SINV and ASALV, whereas hsfa1 treatment significantly reduced virus replication for SINV, ASALV and DENV..."
• "The fact that the heat shock response was antiviral in both U4.4 cells and the RNAi-deficient C6/36 cells indicates that the Hsf1-sHsps cascade functions independently of RNAi."
• Hsf1 as a Potential Target for Intervention: The broad antiviral activity of the Hsf1-sHsp cascade across different arboviruses and mosquito vectors highlights Hsf1 as a promising target for developing novel strategies to block arbovirus transmission at the mosquito stage.
• "Overall, our work uncovers an antiviral cascade Hsf1-sHsp and identifies Hsf1 as a potential target for developing transmission-blocking strategies against arboviruses at the mosquito stage."
• "Considering the pan-antiviral activity of the Hsf1-sHsp cascade, our findings may open the way for the development of a broad-spectrum drug to prevent arbovirus transmission."
• Potential Mechanism of Action: The study suggests that sHsps, known for maintaining protein homeostasis by preventing protein aggregation, might exert their antiviral effects by interacting with conserved viral or host protein complexes involved in early viral replication. Proteomic approaches are proposed to identify sHsp client proteins and elucidate the underlying mechanism.
• "Our time-of-addition assays suggest that the response inhibits an early post-entry stage of the viral replication cycle."
• "It has been suggested that sHsps maintain protein homeostasis by binding proteins in non-native conformations, thereby preventing substrate aggregation."

3. Implications and Future Directions:

• The discovery of the Hsf1-sHsp cascade adds a new layer to our understanding of mosquito antiviral immunity, complementing the well-established role of RNAi and other innate immune pathways.
• Targeting Hsf1 in mosquitoes could offer a novel, broad-spectrum approach to reduce the transmission of multiple medically important arboviruses. This strategy could potentially be less susceptible to viral resistance mechanisms compared to targeting specific viral proteins.
• Further research is needed to validate the in vivo relevance of this cascade in whole mosquitoes and to understand the precise molecular mechanisms by which sHsps inhibit viral replication.
• Investigating the interaction between the Hsf1-sHsp cascade and other m...

Briefing Document: "Recombinant venom proteins in insect seminal fluid reduce female lifespan"

Source: Article "Recombinant venom proteins in insect seminal fluid reduce female lifespan" by Samuel J. Beach & Maciej Maselko, published in Nature Communications (DOI: 10.1038/s41467-024-54863-1).

Date: Current date (based on when the briefing document is created).

Prepared for: [Intended Audience - e.g., Pest Management Stakeholders, Research Community, Regulatory Bodies]

Executive Summary:

This research presents a novel genetic biocontrol technology termed the "toxic male technique" (TMT). The study demonstrates, using Drosophila melanogaster as a model, that genetically engineering males to express insecticidal venom proteins within their seminal fluid, which are then transferred to females during mating, significantly reduces the lifespan of the mated females. This intragenerational approach contrasts with existing mating-based genetic biocontrol technologies that primarily focus on reducing offspring viability or skewing sex ratios. Agent-based modeling of Aedes aegypti suggests that TMT has the potential to be significantly more effective in rapidly reducing blood-feeding rates, a key factor in disease transmission, compared to leading intergenerational biocontrol methods like female-specific Release of Insects carrying a Dominant Lethal gene (fsRIDL). TMT offers a promising avenue for rapidly controlling outbreaks of disease vectors and agricultural pests with potentially reduced reliance on chemical insecticides and lower risks of transgene persistence compared to gene drive technologies.

Main Themes and Important Ideas/Facts:

• The Need for Alternative Pest Management Tools:
• The emergence of insecticide resistance is a major challenge.
• Current genetic biocontrol technologies (GBTs) have limitations, particularly a generational lag in their impact, leaving mated females capable of causing harm (e.g., disease transmission) in the interim.
• The study highlights the significant human and economic costs associated with insect pests and the negative impacts of widespread insecticide use on non-target species and the environment.
• "The emergence of insecticide resistance has increased the need for alternative pest management tools."
• Introducing Intragenerational Genetic Biocontrol: The Toxic Male Technique (TMT):
• TMT aims to directly reduce the lifespan of mated females by expressing toxic proteins in the male reproductive tract (male accessory glands - MAG) that are transferred during mating.
• This is presented as the "first example of an intragenerational GBT."
• "Here we describe the toxic male technique (TMT), which is—to the best of our knowledge—the first example of an intragenerational GBT. By genetically engineering pest species to express toxic, low molecular weight compounds within the MAG, the survival of mated females can be reduced (Fig. 1)."
• Proof of Concept in Drosophila melanogaster:
• Seven insect-specific venom proteins were selected based on criteria including invertebrate specificity, low mammalian toxicity, and low molecular weight.
• MAG-specific expression was achieved using the antr-GAL4 driver in the UAS/GAL4 system.
• Mating with males expressing Γ-CNTX-Pn1a (from a spider) or δ-AITX-Avd2a (from a sea anemone) significantly reduced the median lifespan of wild-type females by 37-64% compared to controls.
• Increasing the ratio of TMT males to females (3:1) resulted in an even greater reduction in female lifespan (50% and 64% respectively).
• "The median lifespans of wild-type females mated to two of the TMT strains are reduced by 37–64% after initial exposure compared to control females mated to wild type males."
• Functionality and Specificity of Recombinant Venoms:
• Functionality of the venom proteins was confirmed through ubiquitous expression assays, where functional venoms resulted in non-viable offspring.
• The antr-GAL4 driver showed strong and specific expression in the male accessory glands with minimal off-target expression.
• Single-pair courtship assays indicated that TMT males were able to court females as effectively as wild-type males, suggesting no significant impact on mating behavior.
• However, competitive mating assays revealed reduced competitiveness of the transgenic males likely due to the GAL4-UAS genetic background.
• Modeling the Impact on Aedes aegypti Populations:
• Agent-based models simulating Aedes aegypti release programs compared TMT to SIT and fsRIDL.
• The model predicted that TMT could reduce blood-feeding rates by a further 40-60% during release periods compared to fsRIDL.
• TMT achieved faster population reduction (PR50 and PR95) compared to fsRIDL in the simulations.
• The median number of gonotrophic cycles (related to disease transmission potential) was significantly lower in the TMT simulations.
• "Agent-based models of Aedes aegypti predict that TMT could reduce rates of blood feeding by a further 40 – 60% during release periods compared to leading biocontrol technologies like fsRIDL."
• Impact of Mating Lethality and Male Sterility in Models:
• Simulations exploring reduced mating lethality showed that fertile TMT males required ≥80% mating lethality to outperform fsRIDL in reducing blood-feeding.
• Sterile TMT males (combined with a sterility mechanism) only required ≥40% mating lethality to outperform fsRIDL.
• This highlights the potential benefit of combining TMT with male sterilization to prevent transgene escape.
• Sensitivity Analysis of Model Parameters:
• Sensitivity analysis revealed that polyandry (multiple mating by females) and release ratio of TMT males had significant positive effects on TMT's relative performance.
• Density-dependent mortality had a negative effect, although this could be mitigated by increased release ratios.
• Higher rates of polyandry are predicted to enhance the effectiveness of TMT.
• Potential and Future Directions of TMT:
• TMT offers a rapid response to pest outbreaks, making it a potential alternative to chemical pesticides as a first-line response.
• The taxonomic selectivity of venom proteins allows for targeted pest control.
• Co-expression of multiple toxins could help prevent resistance.
• Future research needs to focus on identifying optimal venom proteins (considering molecular weight and toxicity), improving expression levels, and engineering stable, conditional expression systems for mass rearing.
• Determining the mating competitiveness of TMT males in target species under wild-type genetic backgrounds is crucial.
• Further studies are needed to understand the epidemiological effects and the impact on non-target organisms.
• "TMT is inherently self-limiting due to strong selection pressure against venom expression, which maybe advantageous from a regulatory perspective as, unlike GBT such as gene drives, there is little risk of substantial transgene introgression into non-target populations, and TMT transgenes are predicted to be rapidly lost from all populations in the absence of ongoing releases."

Quotes:

• "Here, we demonstrate intra-generational genetic biocontrol, wherein mating with engineered males reduces female lifespan."
• "The toxic male technique (TMT) involves the heterologous expres...

Briefing Document: Leveraging AHP and Transfer Learning for Infectious Disease Outbreak Prediction

Dates: Received - 10 August 2024 | Accepted - 26 November 2024 | Published - 31 December 2024
Citation: Abdallah, R., Abdelgaber, S. & Sayed, H.A. Leveraging AHP and transfer learning in machine learning for improved prediction of infectious disease outbreaks. Sci Rep 14, 32163 (2024). https://doi.org/10.1038/s41598-024-81367-1
Keywords: Infectious diseases, AHP, Transfer learning, Risk factors, Machine learning, Dengue, Chikungunya, Zika, Outbreak prediction

Executive Summary:

This study investigates the application of advanced machine learning techniques, specifically an ensemble model integrating Random Forest, XGBoost, and Gradient Boosting, enhanced by Analytic Hierarchy Process (AHP) for feature selection and transfer learning, to predict outbreaks of Dengue, Chikungunya, and Zika. Utilizing a comprehensive dataset of climate and socioeconomic factors from Colombia (2007-2017), the researchers demonstrate that their integrated approach significantly improves the accuracy and reliability of outbreak predictions. The ensemble model achieved notable success, particularly in predicting Zika outbreaks with an accuracy of 96.80% and an AUC of 0.9197. The study highlights the importance of considering a broad range of risk factors, leveraging expert knowledge through AHP, and utilizing transfer learning to overcome data scarcity and improve model generalizability across related diseases.

Main Themes and Important Ideas/Facts:

• The Critical Need for Precise Outbreak Prediction:
• Infectious diseases pose a significant threat to global health and economic stability, underscoring the crucial need for accurate outbreak predictions to enable effective mitigation strategies.
• "Infectious diseases significantly impact both public health and economic stability, underscoring the critical need for precise outbreak predictions to effictively mitigate their impact."
• Early and precise prediction allows public health authorities to deploy timely control and prevention strategies, optimizing resource allocation and minimizing the impact of outbreaks.
• "Early and precise prediction of such outbreaks is essential for public health authorities to deploy effective control and prevention strategies, thereby mitigating impacts on public health and resources."
• Limitations of Traditional Prediction Methods and the Promise of Machine Learning:
• Traditional methods face challenges due to the complex dynamics of infectious diseases and limitations in available data.
• Machine learning (ML) has emerged as a powerful tool capable of analyzing diverse datasets and identifying complex relationships influencing disease dynamics.
• "Machine learning (ML) has emerged as a powerful tool, capable of analyzing various and diverse datasets to identify complex relationships among the various factors that influence disease dynamics."
• ML algorithms excel at identifying patterns and trends that may be missed by human analysis, enabling timely prediction based on continuous monitoring of risk factors.
• The Role of Analytic Hierarchy Process (AHP) for Feature Selection:
• AHP, a multi-criteria decision-making approach, is used to systematically identify and prioritize the most influential risk factors for infectious disease outbreaks.
• "The analytic hierarchy process (AHP), a multi-criteria decision making (MCDM) approach developed by Thomas L. Saaty, organizes factors into a hierarchical structure, proving invaluable for decision-makers navigating complex scenarios."
• AHP incorporates expert domain knowledge through pairwise comparisons to assign weights to various climatic, socioeconomic, and demographic factors.
• The study identifies "barriers to health services," "dependency rate," and "no health insurance" as the highest-ranked risk factors.
• "Based on the previous rankings of risk factors shown in Fig. 2, it is concluded that the highest-ranked factors; barriers to health services, dependency rate, and lack of health insurance play a critical role in disease outbreaks."
• Consistency checks (Consistency Index and Consistency Ratio) are employed to ensure the reliability of the weight assignments.
• Addressing Data Scarcity with Transfer Learning:
• Transfer learning is utilized to leverage knowledge from one domain (Dengue outbreak data) to improve prediction in related domains (Chikungunya and Zika).
• "Transfer learning, which utilizes knowledge from one domain to solve related issues in another, addresses this by transferring models, weights, or features from one disease context to another."
• The model is pre-trained on a comprehensive Dengue dataset, which shares similarities with Chikungunya and Zika in terms of transmission vectors and influencing factors.
• Fine-tuning is then performed using the specific data for Chikungunya and Zika.
• The Effectiveness of Ensemble Machine Learning Models:
• The study employs Random Forest, XGBoost, and Gradient Boosting algorithms, along with an ensemble technique that combines their predictions.
• The ensemble model demonstrated the highest overall predictive performance.
• For Zika, the ensemble model achieved the highest accuracy (96.80%) and AUC (0.9197).
• "The result reveals that the ensemble model is particularly effective, achieving the highest accuracy rate of 96.80% and an AUC of 0.9197 for predicting Zika outbreaks."
• For Chikungunya, the ensemble model achieved a balanced performance with an accuracy of 93.31%, precision of 57%, and recall of 63%, highlighting its reliability.
• "Notably, in the context of Chikungunya, this model achieves an optimal balance between precision and recall, with an accuracy of 93.31%, a precision of 57%, and a recall of 63%, highlighting its reliability for effective outbreak prediction."
• Methodology Overview:
• The proposed model consists of six layers: data source, preprocessing, feature engineering (including AHP for selection and 75th percentile method for outbreak threshold definition), data splitting (80% train, 20% test), modeling (Random Forest, XGBoost, Gradient Boosting, Ensemble with transfer learning), and evaluation.
• A predefined search technique across major online databases was used to identify common risk factors.
• The dataset comprised climate and socioeconomic data from NASA, DANE (Colombia), and SIVIGILIA (Colombia), totaling 1716 instances with 27 features for Dengue, Chikungunya, and Zika.
• Data preprocessing involved EDA, handling missing values (e.g., mean imputation for average temperature), and data transformation (Min-Max normalization and one-hot encoding).
• Model Evaluation Metrics:
• Model performance was evaluated using accuracy, precision, recall, F1-score, and Area Under the ROC Curve (AUC).
• Confusion matrices were used to visualize the prediction performance of each model.
• Comparison with Existing Research:
• The study distinguishes itself from prior research by implementing AHP for more precise feature selection, addressing the limitations of predefined feature selection libraries.
• The integration of transfer learning to overcome data scarcity is another key contribution.
• Limitations of the Study:
• Limited labeled datasets for Chikungunya and Zika.
• Lack of comprehensive clinical data at the patient lev...

Briefing Document: Spatial and Temporal Analysis of West Nile Virus Infection in Ontario, Canada (2003-2022)

Citation: Talbot, B., Ludwig, A., O’Brien, S.F. et al. Spatial and temporal analysis of West Nile virus infection in mosquito and human populations based on arboviral detection testing data. Sci Rep 14, 31343 (2024). https://doi.org/10.1038/s41598-024-82739-3

Dates: Received - 09 July 2024 | Accepted - 09 December 2024 | Published - 28 December 2024

Prepared for: Saleh Lab

Executive Summary:

This study analyzed over two decades of data from Ontario, Canada, to understand factors influencing West Nile virus (WNV) infection in humans. Researchers integrated mosquito surveillance, human blood donation testing, and environmental and socioeconomic data. Their findings indicate that higher habitat suitability for the primary mosquito vector in southern regions and lower socioeconomic status are significant risk factors for human WNV infection. The research also identified climatic variables like temperature and precipitation as important for mosquito populations and WNV transmission. Ultimately, this spatio-temporal analysis aims to inform public health interventions to reduce the risk of mosquito-borne diseases in Canada.

Main Themes and Important Ideas/Facts:

1. Endemic Nature and Surveillance of WNV in Ontario:

• WNV was introduced to North America in 1999 and Canada in 2001, rapidly spreading and becoming endemic in much of the continent.
• Ontario has experienced an increase in WNV cases over time with interannual fluctuations and several epidemics.
• Since 2003, regular WNV detection testing has been conducted on most blood donations in Canada to prevent transmission to recipients.
• While not perfectly representative of the general population, blood donor testing data offers a valuable "representative snapshot of spatio-temporal variations in WNV infection risk in the human population."
• Traditional human case reporting has limitations, including underreporting of asymptomatic infections.

2. Key Mosquito Vectors and Their Dynamics:

• Culex pipiens/restuans and Aedes vexans are identified as important mosquito vectors for WNV in northeastern North America, acting as bridges between avian hosts and humans.
• Cx. pipiens/restuans had a higher minimum infection rate (MIR) for WNV compared to Ae. vexans. The overall MIR was higher in the latter half of the study period (2013-2022).
• The abundance and occurrence of Ae. vexans generally decreased across the study years, while these remained relatively stable for Cx. pipiens/restuans.
• Peak abundance and occurrence of both vector species, as well as WNV infection prevalence in mosquitoes and human blood donors, occurred in August and September.
• "No WNV-positive mosquito pools were observed from November to May of any year."
• Interestingly, Culex tarsalis, a WNV vector more common in western North America, was also observed in Ontario.

3. Environmental Factors Influencing WNV Risk:

• Ecological niche modeling revealed that temperature was the most important explanatory variable for the habitat suitability of both mosquito species and WNV.
• Precipitation had a strong effect on Ae. vexans habitat suitability, with average values being optimal.
• Land cover had a moderate importance, with wetlands and residential areas with low to medium green cover showing high suitability for both mosquito species and WNV. Grasslands were suitable for mosquitoes but not WNV.
• The highest projected habitat suitability for both mosquito species and WNV was concentrated in southern Ontario, particularly the corridor from Windsor to Toronto. WNV habitat suitability was more confined to major urban centers.

4. Socio-economic and Demographic Factors Associated with Human WNV Infection:

• Statistical modeling at the census subdivision level identified lower income (proportion of households earning less than $20,000) as a strong positive predictor of human WNV infection (IRR = 2.8).
• Conversely, a high proportion of residents spending 30% or more of their income on shelter showed a strong negative association with WNV infection (IRR = 0.7).
• Habitat suitability index for Cx. pipiens/restuans was also a significant positive predictor of human WNV infection (IRR = 2.0).
• "Our results show that high habitat suitability of the main WNV vector in southerly locations of this region, Cx. pipiens/restuans (IRR = 2.0), and variables related to lower income (IRR = 2.8), and shelter infrastructure spending (IRR = 0.7), were key risk factors associated with WNV infection among blood donors from 2003 to 2022 across Ontario (R2 = 0.67)."
• Population density and residential land cover did not show a significant effect on human WNV infection in the multivariable models in this study, contrasting with some previous research.

5. Individual-Level Factors:

• Analysis at the individual blood donation level found that donor age, year of detection test, and month of detection test had statistically significant positive associations with WNV infection. However, the model had poor statistical power due to the low number of positive cases.
• "Donor age, year of detection test and month of detection test had a slope coefficient between 1.002 and 1.003 (Table 4), suggesting a positive association of these variables with WNV infection."
• The cumulative WNV infection rate was slightly higher for male donors and much higher for donors older than 65.

6. Implications for Public Health Interventions:

• The study highlights modifiable risk factors, such as housing infrastructure and socio-economic status, that can be targeted for interventions.
• "For example, our study supports the need for government education campaigns and incentives facilitating renovations aimed at reducing mosquito habitat and/or exposure to mosquitoes, especially in areas with highly suitable habitat for Cx. pipiens/restuans in Southern Ontario..."
• The findings underscore the need for rigorous surveillance in southern Ontario due to the strong influence of temperature on mosquito vector habitat and subsequent human infection risk, especially in the context of climate change.
• The study provides a model for using blood donor testing data for widespread arboviral surveillance.

7. Limitations:

• Nucleic acid tests used have a potential for cross-reactivity with other viruses in the Japanese encephalitis serocomplex.
• Mosquito sampling and human donor data are highly concentrated around urban and metropolitan areas.
• Blood donor population may not perfectly represent the general population.

Quotes:

• "While the blood donor population isn’t absolutely synonymous with the general population, data from WNV detection tests on blood donations offer a representative snapshot of spatio-temporal variations in WNV infection risk in the human population."
• "Mosquito species of the genus Culex, such as Culex pipiens and Culex restu...

Briefing Document: Explainable Deep Learning for Molecular Discovery
Citation: Wong, F., Omori, S., Li, A. et al. An explainable deep learning platform for molecular discovery. Nat Protoc (2024). https://doi.org/10.1038/s41596-024-01084-x
Dates: Received - 11 March 2024 | Accepted - 26 September 2024 | Published - 09 December 2024
Source: Excerpts from "An explainable deep learning protocol for de novo molecular discovery" (s41596-024-01084-x.pdf)

Executive Summary

This briefing document summarizes a research protocol for leveraging explainable deep learning (DL) to accelerate the discovery of novel molecules with desired properties, particularly focusing on antibiotic discovery as a case study. The protocol utilizes Graph Neural Networks (GNNs) and a software package called Chemprop to build predictive models of molecular properties based on experimental data. A key innovation is the integration of Monte Carlo Tree Search (MCTS) to provide "rationales" – specific chemical substructures that explain the model's predictions. This explainability contrasts with traditional "black box" DL approaches and allows researchers to gain chemical insights, efficiently narrow chemical spaces, and prioritize compounds with promising structural features for experimental validation. The protocol encompasses data generation, model training and benchmarking, rationale analysis, and experimental validation, and is designed to be accessible even without extensive coding expertise or specialized hardware.

Main Themes and Important Ideas

1. The Power and Limitation of Deep Learning in Molecular Discovery

• DL approaches have demonstrated significant success in modeling compounds, leading to increased true discovery rates and the identification of drug candidates across various therapeutic areas.
• "DL approaches have accurately modeled compounds, with increased true discovery rates, and resulted in the discovery of drug candidates including antibiotics7,17–22, senolytics23 and anticancer and antiviral combinations24,25."
• However, a major drawback of many DL models is their "black box" nature, lacking transparency in how they arrive at predictions.
• "However, a major limitation to DL approaches is that they are typically black box in nature or unable to provide reasoning behind model predictions."

2. Explainable Deep Learning (xDL) as a Solution

• Explainable DL is an emerging field focused on providing the reasoning behind DL model predictions, contrasting with interpretable DL which aims to reveal the decision-making steps.
• "Explainable DL is an emerging field that aims to open up the black box by providing this reasoning7,34–37. Explainable DL contrasts with interpretable DL, which aims to reveal patterns of decision-making steps the models perform to arrive at their predictions38 (see Box 2 for further comparison)."
• In the context of molecular discovery, xDL identifies chemical substructures ("rationales") with positive predictive value for a property of interest.
• "As applied to molecular discovery, explainable DL identifies patterns of chemical atoms and bonds—chemical substructures—that have positive predictive value for a property of interest."
• This enables generalization beyond individual compounds to classes of compounds sharing these substructures, leading to specific chemical insights and efficient narrowing of chemical spaces.
• "By enabling models to predict structural classes directly, explainable DL can produce specific chemical insights, efficiently narrow chemical spaces and lead researchers toward effective chemical scaffolds."
• Explainability and interpretability are driving research towards increased comprehension, trustworthiness, and accountability of AI/ML models (Box 2).

3. The Proposed Explainable DL Platform and Protocol

• The protocol introduces a general-purpose explainable DL platform for molecular discovery, using GNNs as the DL model architecture.
• "In this protocol, we present an explainable DL platform for molecular discovery (Fig. 1)."
• "The platform uses graph neural networks (GNNs) as a DL model architecture. GNNs model chemical structures as collections of nodes (chemical atoms) and edges (chemical bonds) and perform computations that pool together information from neighboring atoms and bonds16,39–42."
• The protocol provides a practical guide to Chemprop, a user-friendly software package implementing GNNs, specifically directed Message Passing Neural Networks (MPNNs).
• "We provide a practical introduction to Chemprop16,43–45, a software package implementing GNNs..."
• "Chemprop is a software package implementing a variant of MPNNs, in which messages are associated with directed edges in contrast to nodes16,45."
• The protocol covers experimental data generation (using high-throughput screening as an example), model implementation, model explainability (using MCTS), evaluation, and experimental validation.
• "This protocol includes both computational and experimental components, and it does not require coding proficiency or specialized hardware. Starting from data generation and ending in the testing of model predictions, this protocol can be executed in as little as 1–2 weeks, depending on dataset size and the time required for experiments."
• The protocol is exemplified through antibiotic discovery, but its applicability extends to discovering molecules with other biological or chemical properties.

4. Key Components of the Protocol

• Graph Neural Networks (GNNs) and Message Passing Neural Networks (MPNNs): GNNs process molecular structures represented as graphs (atoms as nodes, bonds as edges) through message passing between neighboring atoms and bonds. Chemprop implements directed MPNNs.
• "GNNs are a type of ANN designed to operate on input data represented as graphs. Graphs are mathematical objects that consist of nodes and edges. For molecular discovery, these nodes and edges represent chemical atoms and bonds, respectively."
• "To incorporate local connectivity information, message passing—a step where embeddings from a node or edge’s neighbors are pooled with its preexisting embeddings—is often incorporated into a GNN layer’s update function, resulting in an architecture called MPNNs."
• Chemprop: A user-friendly, constantly maintained software package implementing MPNNs with diverse functionalities and customization options. It was successfully used in the discovery of halicin as an antibacterial compound.
• "Key features of Chemprop are that it is user friendly, possesses diverse functionalities and customization options and, since its development in 2019, has been constantly maintained. Its application to antibiotic discovery, resulting in the identification of halicin as an antibacterial compound17, was reported by our laboratory in 2020."
• Monte Carlo Tree Search (MCTS) for Explainability: For molecules predicted to be active ("hits"), MCTS is used to identify the smallest substructure ("rationale") that still results in a high predicted activity. This provides an explanation for the model's prediction.
• "For each molecule with a suitably high prediction score (‘hit’), we aim to determine the smallest substructure resulting in the molecule being classified as active (the molecule’s ‘rationale’)."
• "The MCTS algorithm solves the discrete optimization problem of finding the rationale of a molecule... each state in the search tree is a substructure (su...

Briefing Document: Native Structure of Mosquito Salivary Protein Uncovers Domains Relevant to Pathogen Transmission

Dates: Received - 04 October 2022 | Accepted - 07 February 2023 | Published - 17 February 2023
Source: Liu, S., Xia, X., Calvo, E., & Zhou, Z. H. (2023). Native structure of mosquito salivary protein uncovers domains relevant to pathogen transmission. Nature Communications, 14(1), 1-20. https://doi.org/10.1038/s41467-023-36577-y

Executive Summary:

This study, published in Nature Communications, presents the first native structure of the Salivary Gland Surface protein 1 (SGS1) from the Aedes aegypti mosquito, a key vector for numerous human pathogens. Utilizing cryo-electron microscopy (cryo-EM), the researchers determined the structure of this high-molecular weight protein, revealing a complex architecture comprised of a Tc toxin-like Rhs/YD shell, four receptor domains, and a set of partially shielded C-terminal daisy-chained helices predicted to form transmembrane helices. The structural insights shed light on the potential mechanisms by which SGS1 facilitates pathogen invasion into salivary glands and manipulates the host's immune response during blood feeding, a process responsible for approximately one million deaths annually.

Main Themes and Important Ideas/Facts:

• SGS1: A Crucial Protein in Pathogen Transmission:
• Female mosquitoes inject saliva containing various proteins into vertebrate hosts during blood feeding, facilitating pathogen transmission. This process is responsible for significant human morbidity and mortality.
• SGS1 is identified as one of the most abundant and conserved proteins secreted by female salivary glands of key disease-vectoring mosquitoes like Aedes aegypti, Culex pipiens, and Anopheles gambiae.
• Previous research indicated that SGS1 is essential for the invasion of Aedes aegypti salivary glands by Plasmodium gallinaceum sporozoites and positively affects Zika virus transmission.
• Orthologs of SGS1 in other mosquito species are thought to enhance the pathogenicity of arboviruses and Plasmodium parasites by modulating the host's immune response.
• The protein is exclusively expressed in the salivary glands of adult female mosquitoes, suggesting a specific role in blood-feeding and pathogen transmission.
• Cryo-EM Reveals the Native Structure of SGS1:
• The study successfully determined the native structure of full-length, uncleaved SGS1 (3364 amino acids) using cryoID and cryo-EM at an overall resolution of 3.3 Å.
• The structure is "cocoon shaped, with dimensions of 210Å × 115 Å × 86 Å" and is organized into six domains: two β-propeller domains, a rearrangement hotspot/tyrosine-aspartate (Rhs/YD)-repeats domain, a carbohydrate-binding module (CBM), a lectin carbohydrate-recognition domain (lectin-CRD), and a wedge domain.
• The C-terminal region, predicted to form transmembrane (TM) helices, is surprisingly "almost fully buried within the chamber inside the cocoon shell and forms daisy-chained helices."
• The Rhs/YD Shell: A Protective and Potentially Activating Environment:
• The Rhs/YD-repeats domain forms a "hollow shell (i.e., Rhs/YD shell)" structurally similar to that found in bacterial Tc toxins and eukaryotic teneurins.
• This shell encapsulates the C-terminal moiety, suggesting a mechanism for maintaining these potentially membrane-interacting helices in a soluble environment.
• Sequence analysis reveals a conserved aspartyl auto-protease (AP) site within the Rhs core of SGS1, similar to Tc toxins. Mass spectrometry data indicates that "an SGS1 peptide covering this auto-cleavage site was present in the salivary gland extract but was not detected in the saliva of Aedes aegypti," suggesting that auto-cleavage occurs upon secretion into saliva.
• The authors propose that a conformational change in SGS1 is likely required to activate this catalytic activity during secretion.
• Partially Folded Transmembrane Helices within the Chamber:
• The ~230 amino acid long C-terminal moiety, predicted to form six TM helices, is found to be "almost fully embedded inside the SGS1 chamber."
• Cryo-EM reveals that these predicted TM helices are only "partially folded and entirely sheltered in the chamber," interacting extensively with the interior of the Rhs/YD-repeats and the β-propeller 2 domain via hydrophobic contacts.
• Despite not being fully folded in the observed structure, the high conservation of these sequences among different mosquito species suggests an "essential function."
• The authors hypothesize that these metastable, extended structures might facilitate their detachment from the inner shell and potential insertion into host cell membranes upon injection.
• Receptor Domains on the Shell Surface:
• The exterior of the SGS1 Rhs/YD shell is decorated with four potential receptor domains: β-propeller 1, β-propeller 2, a carbohydrate-binding module (CBM), and a lectin carbohydrate-recognition domain (lectin-CRD).
• These domains are proposed to "mediate protein-protein interactions and/or facilitate carbohydrate binding" during pathogen transmission.
• The presence of these receptor domains is conserved among SGS proteins of disease-transmitting mosquito species.
• Proposed Model for Cleavage and Secretion:
• The study proposes a model where the C-terminal Tox-SGS domain is initially cleaved by furin protease in the salivary gland. This is supported by the absence of density for Tox-SGS in the cryo-EM map and mass spectrometry results indicating no peptides after residue 3035 were recovered from salivary glands.
• The remaining SGS1 fragment, potentially with the help of its receptor domains, may facilitate sporozoite/arbovirus invasion of the salivary gland.
• Subsequently, the SGS1 fragment is secreted into saliva, where the aspartyl auto-protease has already catalyzed its cleavage.
• The putative TM helices remain embedded within the Rhs/YD shell in saliva, protected from the soluble environment.
• Upon injection into the host during blood feeding, SGS1 may interact with host cells via its receptor domains and potentially release its TM helices, which could modulate host immune responses to benefit pathogen transmission.

Key Quotes:

• "Among the most abundant and conserved proteins secreted by female salivary glands is a high-molecular weight protein called salivary gland surface protein 1 (SGS1) that facilitates pathogen trans-mission, but its mechanism remains elusive."
• "Here, we determine the native structure of SGS1 by the cryoID approach, showing that the 3364 amino-acid protein has a Tc toxin-like Rhs/YD shell, four receptor domains, and a set of C-terminal daisy-chained helices."
• "These helices are partially shielded inside the Rhs/YD shell and poised to transform into predicted transmembrane helices. This transformation, and the numerous receptor domains on the surface of SGS1, are likely key in facilitating sporozoite/arbovirus invasion into the sali-vary glands and manipulating the host’s immune response."
• "The SGS1 structure is cocoon shaped, with dimensions of 210Å × 115 Å × 86 Å. It is organized into 6 domains: two β-propeller domains, a rearrangement hotspot/tyrosine-aspartate (Rhs/ YD)-repeats domain, a carbohydrate-binding module (CBM), a lectin carbohydrate-recognition domain (lectin-CRD), and a wedge domain."
• "Sequence alignment indicate...