130 Program for Research in Science and Engineering Evaluating Aerosol-Type Effects on MethaneSAT’s CO Retrieval Accuracy 2

Authors:

Leonardo Di Tommaso, Eleanor Walker, Steven Wofsy

Date Created:

2025-01-01

Course Title:

Professor:

Not specified

About Paper:

Carbon dioxide (CO ) emissions are a major driver of global from the NASA Deep Blue Algorithm. For desert dust scenes, warming, and improved monitoring is essential for informing we found a weak but statistically significant negative correlation climate policy. While MethaneSAT is a satellite primarily (slope: ▯1:84, R 2 = 0:006, p ▯ 0:035). In contrast, scenes designed to detect methane (CH )4 its spectrometers also capture with smoke from agricultural burning showed a stronger positive CO and oxygen (O ) signals, presenting an opportunity to relationship (slope: 2:90, R = 0:03, p < 1e-6), and war-related 2 2 estimate CO 2luxes with high spatial resolution. The effect of smoke scenes exhibited the strongest correlation (slope: 29:30, aerosol scattering on these retrievals, if any, remains unclear. R = 0:13, p < 10 ▯5). These results suggest that the impact of This study investigates the relationship between atmospheric aerosols on the O2-proxy CO r2trievals from MethaneSAT varies aerosols and MethaneSAT’s CO 2 product. We conducted a by aeros2l type, likely due to differences in scattering behavior. geospatial comparison between products of MethaneSAT (i.e. O - 2 Low R values may reflect differing overpass times and inherent proxy-corrected CO ,2delta pressure (dP), albedo) and products noise in retrievals. Future work will explore correlation between of the Visible Infrared Imaging Radiometer Suite (VIIRS) (i.e. MethaneSATproductsandAODattheirretrievalwavelengths(i.e. Aerosol Optical Depth (AOD), Angstrom exponent (AE), surface 1286 nm, 1610 nm) calculated using the Angstrom-power law. reflectance). Scenes having various aerosol types, including dust, This work contributes to understanding MethaneSAT’s capacity to smog, and fire smoke, were analyzed using robust regressions and generate reliable CO 2roducts and highlights the need for future density distribution plots in R. Preliminary results indicate scene-odels to incorporate aerosol-type-specific corrections. specificrelationshipsbetweendP(hPa)andAOD(550nm)derived Investigating Unique Somatic Mutations in Prenatal Brains Exposed to Cytomegalovirus Emerson Utgaard, Benjamin Finander, Christopher Walsh Harvard College | Winthrop House | Neuroscience | 2027 Cytomegalovirus (CMV) is the pathogen most commonly mutations with low DNA input. Because DNA from FFPE brain transferred from a pregnant individual to the developing fetus. tissue is already degraded from the fixation process, we must also One in two hundred babies are born with congenital CMV, which optimize the META-CS protocol for our specific samples. Once is associated with birth defects, hearing and vision problems, and we have obtained DNA from META-CS, we plan to compare an increased risk for psychiatric disorders. One of the known the somatic mutation burden between our CMV samples and age- molecular consequences of cytomegalovirus activity in cultured matched controls. Finding a unique signature of somatic mutations human cells is a heightened DNA Damage Response, but it is in the CMV samples would support our hypothesis that a maternal currently unknown whether any form of DNA damage occurs in CMV infection causes somatic mutations in the developing brain, the developing brain after CMV infection. Studies in individuals offering a possible connection between congenital CMV and with diagnosed neuropsychiatric disorders like schizophrenia psychiatric disorders. Discovering increased expression of DNA and autism show somatic Single Nucleotide Variants (sSNVs) damage genes in the RNA-seq data would also support our that suggest a somatic mutational process active during the hypothesis. Because CMV is the most common infectious risk developmentofthebrain.Therefore,ourprojectaimstofindunique factor for birth defects in the United States, it is essential to gain somatic mutations in prenatal brains exposed to cytomegalovirus. a better understanding of how maternal CMV infection affects the To investigate this, we extracted DNA and RNA from FFPE developing brain. prenatal brain tissue. We then performed bulk RNA-Seq and META-CS, a strand-tagging PCR technique used to call somatic Characterizing StitchR for High-throughput Protein-Protein Interaction Screening Assays Tanya Vidhun, Nicolas Rey, Richard Sherwood Harvard College | Quincy House | Human Developmental and Regenerative Biology | 2028 PPI-seq, a high-throughput assay for detecting and quantifying constructs. To assess whether RNA ligation efficiency could be protein-protein interactions (PPIs) in living cells, links interactionanced, we compared conditions with and without RtcB, an events to RNA readouts. However, its current design faces some RNA ligase that promotes StitchR-mediated fusion. We used key limitations. Both proteins of interest must be expressed from aunstitched N-t and C-t RNAs as controls. Reverse transcription single, large plasmid that carries multiple coding sequences and qPCR (RT-qPCR) assays were designed to specifically detect strong promoters. These large constructs are difficult to clone, stitched products. We optimized both primer design and reverse incompatiblewithlentiviraldelivery, andlimitscalabilitytomany- transcription (RT) conditions, identifying Random Primer RT as versus-many pooled interaction screens. Therefore, we draw on optimal for amplifying ligated RNA. Initial qPCR data confirmed StitchR, a novel synthetic RNA system that utilizes ribozyme that there is successful StitchR-mediated stitching of N-t and C-t machinery within the cell to catalyze “stitching” of multiple RNA RNA fragments, validating the system in human cells. Primer and fragments. This mechanism enables RNA fragments from two RT optimization improved sensitivity and specificity of stitched separate plasmids to be “stitched” together inside the cell and product detection. We are now developing a quantitative method preservesthenecessarybarcodeandeditinginformationonasingle to measure stitching efficiency across conditions. Small-scale RNA molecule. To test the suitability and efficiency of StitchR characterizationofStitchRinthecontextofourPPI-seqsystemwill for PPI-seq, we transduced human HCT116 cells with lentivirus guide future iterations of PPI-seq and help enhance the scope and carrying separate N-terminal (N-t) and C-terminal (C-t) StitchR quality of potential binary protein interactions that can be studied. Comparing the Effects of Potency Differences Between GDF11 and GDF8 on Cardiac Remodeling Romi Wagner, Hannah Fandl, Richard T. Lee Harvard College | Winthrop House | Human Developmental and Regenerative Biology | 2028 Heart disease remains one of the foremost biomedical challenges uses mice with CRISPR-Cas9 mediated substitutions of the gene of our time, accounting for 1 in 5 deaths in the United States. sequences responsible for the differential potency of GDF11 and Improved therapies must be developed to resolve the hypertrophy GDF8,diminishingGDF11potencyandincreasingGDF8potency. and fibrosis that occur as a result of pressure overload and To chemically induce cardiac stress, we deliver angiotensin-II stress. Growth differentiation factor 11 (GDF11) is a circulating via osmotic mini-pumps. After 28 days, we harvest tissues factor which has been reported to reverse pathologic remodeling and assess for several markers of pathologic remodeling. We in several models of cardiac disease. Myostatin (GDF8) is a characterizehypertrophyviaanalysisofheartweightnormalizedto negative regulator of muscle growth, and is expressed primarily tibia length, body weight, and by measuring cardiomyocyte cross- in skeletal muscle. Initial studies supposed that GDF11 was sectionalarea. Wealsoevaluatetheextentoffibrosisbymeasuring identical to its homolog GDF8 due to the ligands’ high similarity fibrotic area and quantifying collagen deposition. Currently, more and shared signaling pathways. However, more recent studies experiments are necessary to draw conclusive outcomes from have shown that amino acid differences at the type I receptor our study. These ongoing results will clarify the importance interface enable GDF11 to induce SMAD2/3 signaling more of signaling potency on GDF11’s cardioprotective properties, potently. We investigate the effect of potency differences between informing future decisions about the therapeutic relevance of GDF11 and GDF8 on in vivo progression of cardiac hypertrophy GDF11. and fibrosis during angiotensin-II treatment. Our experiment 132 Program for Research in Science and Engineering CXR-Osteo: Using Deep Learning on Chest Radiographs to Predict Osteoporosis Risk Jessie Wang, Lauren Cooke, Vineet Raghu Harvard College | Eliot House | Human Developmental and Regenerative Biology | 2028 Osteoporosis is a major public health concern that often remains cleaned and merged dataset of PLCO participants (N ≈ 200,000) undiagnosed until after a fracture. Chest radiographs (CXR) are by integrating imaging filenames, metadata, and self-reported among the most common imaging modalities in medicine and are osteoporotic records. We fine-tuned the model using the MIMIC- frequently acquired for unrelated clinical indications. As a result,XR and PLCO datasets to output a predicted probability of they offer a unique opportunity for opportunistic osteoporosis osteoporotic fracture risk, which we recorded for each participant. screening, especially forpatientswhomightnototherwiseundergo dedicatedbonedensityassessment. Buildingonpreviousworkthat Usingthisdataset,weevaluatedtheCXR-Osteomodel’spredictive performance on the entire cohort and found that it achieved an demonstrated promising performance in preliminary datasets, our area under the ROC curve (AUC) of 0.8116, indicating excellent objective is to develop and test a deep learning model, CXR-Osteo, discrimination between individuals who would and would not go for predicting long-term risk of osteoporotic fracture directly from CXRs. We will then evaluate the model’s performance with a on to experience an osteoporotic fracture. Sex-stratified insights large, real-world dataset to determine its potential for clinical may guide thresholding adjustment and are ongoing. integration and stratify its performance by biological sex. This Future directions include testing the CXR-Osteo model in a cohort assessment helps verify whether CXR-Osteo can reliably identify of approximately 31,000 patients who underwent outpatient chest high-risk patients during routine imaging. radiographs in 2013-2014 at Massachusetts General Brigham. This evaluation will determine whether CXR-Osteo predictions Our methods leverage the Prostate, Lung, Colorectal, and Ovarian correspond to actual long-term osteoporotic fracture risk. (PLCO) Cancer Screening Trial dataset, enriched with CXR images and linked osteoporosis outcomes. We constructed a

Abstract:

Carbon dioxide (CO ) emissions are a major driver of global from the NASA Deep Blue Algorithm. For desert dust scenes, warming, and improved monitoring is essential for informing we found a weak but statistically significant negative correlation climate policy. While MethaneSAT is a satellite primarily (slope: ▯1:84, R 2 = 0:006, p ▯ 0:035). In contrast, scenes designed to detect methane (CH )4 its spectrometers also capture with smoke from agricultural burning showed a stronger positive CO and oxygen (O ) signals, presenting an opportunity to relationship (slope: 2:90, R = 0:03, p < 1e-6), and war-related 2 2 estimate CO 2luxes with high spatial resolution. The effect of smoke scenes exhibited the strongest correlation (slope: 29:30, aerosol scattering on these retrievals, if any, remains unclear. R = 0:13, p < 10 ▯5). These results suggest that the impact of This study investigates the relationship between atmospheric aerosols on the O2-proxy CO r2trievals from MethaneSAT varies aerosols and MethaneSAT’s CO 2 product. We conducted a by aeros2l type, likely due to differences in scattering behavior. geospatial comparison between products of MethaneSAT (i.e. O - 2 Low R values may reflect differing overpass times and inherent proxy-corrected CO ,2delta pressure (dP), albedo) and products noise in retrievals. Future work will explore correlation between of the Visible Infrared Imaging Radiometer Suite (VIIRS) (i.e. MethaneSATproductsandAODattheirretrievalwavelengths(i.e. Aerosol Optical Depth (AOD), Angstrom exponent (AE), surface 1286 nm, 1610 nm) calculated using the Angstrom-power law. reflectance). Scenes having various aerosol types, including dust, This work contributes to understanding MethaneSAT’s capacity to smog, and fire smoke, were analyzed using robust regressions and generate reliable CO 2roducts and highlights the need for future density distribution plots in R. Preliminary results indicate scene-odels to incorporate aerosol-type-specific corrections. specificrelationshipsbetweendP(hPa)andAOD(550nm)derived Investigating Unique Somatic Mutations in Prenatal Brains Exposed to Cytomegalovirus Emerson Utgaard, Benjamin Finander, Christopher Walsh Harvard College | Winthrop House | Neuroscience | 2027 Cytomegalovirus (CMV) is the pathogen most commonly mutations with low DNA input. Because DNA from FFPE brain transferred from a pregnant individual to the developing fetus. tissue is already degraded from the fixation process, we must also One in two hundred babies are born with congenital CMV, which optimize the META-CS protocol for our specific samples. Once is associated with birth defects, hearing and vision problems, and we have obtained DNA from META-CS, we plan to compare an increased risk for psychiatric disorders. One of the known the somatic mutation burden between our CMV samples and age- molecular consequences of cytomegalovirus activity in cultured matched controls. Finding a unique signature of somatic mutations human cells is a heightened DNA Damage Response, but it is in the CMV samples would support our hypothesis that a maternal currently unknown whether any form of DNA damage occurs in CMV infection causes somatic mutations in the developing brain, the developing brain after CMV infection. Studies in individuals offering a possible connection between congenital CMV and with diagnosed neuropsychiatric disorders like schizophrenia psychiatric disorders. Discovering increased expression of DNA and autism show somatic Single Nucleotide Variants (sSNVs) damage genes in the RNA-seq data would also support our that suggest a somatic mutational process active during the hypothesis. Because CMV is the most common infectious risk developmentofthebrain.Therefore,ourprojectaimstofindunique factor for birth defects in the United States, it is essential to gain somatic mutations in prenatal brains exposed to cytomegalovirus. a better understanding of how maternal CMV infection affects the To investigate this, we extracted DNA and RNA from FFPE developing brain. prenatal brain tissue. We then performed bulk RNA-Seq and META-CS, a strand-tagging PCR technique used to call somatic Characterizing StitchR for High-throughput Protein-Protein Interaction Screening Assays Tanya Vidhun, Nicolas Rey, Richard Sherwood Harvard College | Quincy House | Human Developmental and Regenerative Biology | 2028 PPI-seq, a high-throughput assay for detecting and quantifying constructs. To assess whether RNA ligation efficiency could be protein-protein interactions (PPIs) in living cells, links interactionanced, we compared conditions with and without RtcB, an events to RNA readouts. However, its current design faces some RNA ligase that promotes StitchR-mediated fusion. We used key limitations. Both proteins of interest must be expressed from aunstitched N-t and C-t RNAs as controls. Reverse transcription single, large plasmid that carries multiple coding sequences and qPCR (RT-qPCR) assays were designed to specifically detect strong promoters. These large constructs are difficult to clone, stitched products. We optimized both primer design and reverse incompatiblewithlentiviraldelivery, andlimitscalabilitytomany- transcription (RT) conditions, identifying Random Primer RT as versus-many pooled interaction screens. Therefore, we draw on optimal for amplifying ligated RNA. Initial qPCR data confirmed StitchR, a novel synthetic RNA system that utilizes ribozyme that there is successful StitchR-mediated stitching of N-t and C-t machinery within the cell to catalyze “stitching” of multiple RNA RNA fragments, validating the system in human cells. Primer and fragments. This mechanism enables RNA fragments from two RT optimization improved sensitivity and specificity of stitched separate plasmids to be “stitched” together inside the cell and product detection. We are now developing a quantitative method preservesthenecessarybarcodeandeditinginformationonasingle to measure stitching efficiency across conditions. Small-scale RNA molecule. To test the suitability and efficiency of StitchR characterizationofStitchRinthecontextofourPPI-seqsystemwill for PPI-seq, we transduced human HCT116 cells with lentivirus guide future iterations of PPI-seq and help enhance the scope and carrying separate N-terminal (N-t) and C-terminal (C-t) StitchR quality of potential binary protein interactions that can be studied. Comparing the Effects of Potency Differences Between GDF11 and GDF8 on Cardiac Remodeling Romi Wagner, Hannah Fandl, Richard T. Lee Harvard College | Winthrop House | Human Developmental and Regenerative Biology | 2028 Heart disease remains one of the foremost biomedical challenges uses mice with CRISPR-Cas9 mediated substitutions of the gene of our time, accounting for 1 in 5 deaths in the United States. sequences responsible for the differential potency of GDF11 and Improved therapies must be developed to resolve the hypertrophy GDF8,diminishingGDF11potencyandincreasingGDF8potency. and fibrosis that occur as a result of pressure overload and To chemically induce cardiac stress, we deliver angiotensin-II stress. Growth differentiation factor 11 (GDF11) is a circulating via osmotic mini-pumps. After 28 days, we harvest tissues factor which has been reported to reverse pathologic remodeling and assess for several markers of pathologic remodeling. We in several models of cardiac disease. Myostatin (GDF8) is a characterizehypertrophyviaanalysisofheartweightnormalizedto negative regulator of muscle growth, and is expressed primarily tibia length, body weight, and by measuring cardiomyocyte cross- in skeletal muscle. Initial studies supposed that GDF11 was sectionalarea. Wealsoevaluatetheextentoffibrosisbymeasuring identical to its homolog GDF8 due to the ligands’ high similarity fibrotic area and quantifying collagen deposition. Currently, more and shared signaling pathways. However, more recent studies experiments are necessary to draw conclusive outcomes from have shown that amino acid differences at the type I receptor our study. These ongoing results will clarify the importance interface enable GDF11 to induce SMAD2/3 signaling more of signaling potency on GDF11’s cardioprotective properties, potently. We investigate the effect of potency differences between informing future decisions about the therapeutic relevance of GDF11 and GDF8 on in vivo progression of cardiac hypertrophy GDF11. and fibrosis during angiotensin-II treatment. Our experiment 132 Program for Research in Science and Engineering CXR-Osteo: Using Deep Learning on Chest Radiographs to Predict Osteoporosis Risk Jessie Wang, Lauren Cooke, Vineet Raghu Harvard College | Eliot House | Human Developmental and Regenerative Biology | 2028 Osteoporosis is a major public health concern that often remains cleaned and merged dataset of PLCO participants (N ≈ 200,000) undiagnosed until after a fracture. Chest radiographs (CXR) are by integrating imaging filenames, metadata, and self-reported among the most common imaging modalities in medicine and are osteoporotic records. We fine-tuned the model using the MIMIC- frequently acquired for unrelated clinical indications. As a result,XR and PLCO datasets to output a predicted probability of they offer a unique opportunity for opportunistic osteoporosis osteoporotic fracture risk, which we recorded for each participant. screening, especially forpatientswhomightnototherwiseundergo dedicatedbonedensityassessment. Buildingonpreviousworkthat Usingthisdataset,weevaluatedtheCXR-Osteomodel’spredictive performance on the entire cohort and found that it achieved an demonstrated promising performance in preliminary datasets, our area under the ROC curve (AUC) of 0.8116, indicating excellent objective is to develop and test a deep learning model, CXR-Osteo, discrimination between individuals who would and would not go for predicting long-term risk of osteoporotic fracture directly from CXRs. We will then evaluate the model’s performance with a on to experience an osteoporotic fracture. Sex-stratified insights large, real-world dataset to determine its potential for clinical may guide thresholding adjustment and are ongoing. integration and stratify its performance by biological sex. This Future directions include testing the CXR-Osteo model in a cohort assessment helps verify whether CXR-Osteo can reliably identify of approximately 31,000 patients who underwent outpatient chest high-risk patients during routine imaging. radiographs in 2013-2014 at Massachusetts General Brigham. This evaluation will determine whether CXR-Osteo predictions Our methods leverage the Prostate, Lung, Colorectal, and Ovarian correspond to actual long-term osteoporotic fracture risk. (PLCO) Cancer Screening Trial dataset, enriched with CXR images and linked osteoporosis outcomes. We constructed a

Source:

Harvard / Albert Tang, Eric Michael Moult, Adam Ezra Cohen / 2025

Topics:

rna, cmv, somatic, gdf11, aerosol, brain, dna, cxr, methanesat, mutation, risk, human

Co-authors:

@leonardoditommaso376 , @eleanorwalker377 , @stevenwofsy378