Coupla notes & thoughts on some things I read/listened to this month :)
Articles
Intellectual humility: the importance of knowing you might be wrong
- Intellectual humility = characteristic allowing for the admission of wrongness; so important now with replication crisis in science, and overall in the information-heavy, super-connected culture we live in (that makes spreading false info so easy)
- Three primary challenges: 1) Need to be aware of cognitive blind spots, ignorance, etc. 2) Need to be able to admit when notice this and were wrong— and need for society to celebrate, not punish, this. 3) It’s impossible to achieve perfect intellectual humility.
- A method of thinking: being curious about your blind spots, open to the possibility you might be wrong and willing to seriously explore and learn about alternatives… “a process of monitoring your own confidence.”
- Concept of statistical equivalence, underteremination by of theory by data —> different models can have equivalent observational consequences ; similar to how human perception is subjective
- Intellectual humility requires institutaiional and peer support… can’t just come from within! Need to be more transparent about our knowledge, up front about what we do and don’t know, AND have celebration of failure and cultural acceptance of it.
- Still need to balance humility with strength of convictions— taking it too far is just giving up on all ideas
Philosophy Is a Public Service
- Jonathan Keats —> biomimicry & art & philosophy & environment, WOW
- Using bristlecone pine tree growth as a clock: grows at a relatively steady rate per year, but changes based on CO2, rain so varies with climate conditions —> built an apparatus (electronic dendrometer) around the tree that measures time in terms of climate change (on the scale of 500-1000 -10000 years); make that data available/accessible to all
- Other environmentally-calibrated clocks like one modulated by river flow in Alaska
- New camera with 100 year exposure time —> observe changes to urban environment on a new time scale: intergenerational surveillance cameras
- Also did a project to try to get people to better understand (“digest”) climate change by interpreting it as ice cream, with specific flavors that somehow stimulate the gut in a certain way that mimics the al bedo effect and such
- Looking to stromatolites (early microbial communities that formed the first terrestrial structures in tidal areas) as inspiration to “staying put” in the face of sea level rise: What if we treated our cities like this? Continue to grow higher, sacrificing lower layers… the rising water also very effectively moderates urban heat island effect!
- “Instead of searching for the Antrhopocene, can we rediscover the Holocene?”
- Paleobiomimicry —> design informed by what early life forms can teach us about how to live
NIH must support broadly focused basic research
- NIH —> big expansion in basic science research after WWII with Vannevar Bush system: expanded scope to include research on any organism to study how cells work, how organisms develop & operate: based on theory of unity of life, importance of model systems to understand complex human/mammalian life. Hallmark = RO1s were investigatory-initiated, not solicited.
- US led biological innovation and research: RNAi, cell cycle, regulation and development all from bacteria, insects, plants, etc. CRISPR
- BUT, recently NIH is narrowing vision to pre-selected subjects, mammalian models… because now we have the tools to do studies on mammalian systems more easily, so shouldn’t bother with simpler models that are “irrelevant”? Initiatives like 4D Nucleome, BRAIN Initiative, limited to mammals, which ignores conservation across metazoans.
- Can’t just focus on “disease-relevance” in research with a very narrow, managed vision… US will lose leadership position in medical sciences and bio.
Systematically Improving Espresso: Insights from Mathematical Modeling and Experiment
- So much math for a perfect espresso shot. This is amazing.
History’s Largest Mining Operation Is About to Begin
- Atlantic article about the rise of deep sea mining, the little we do know & the lot we don’t know about our ocean floors, the pioneering historical expeditions that began to explore them, and current scientific efforts (like deep sea robots!) to map & profile their land & life before capitalism rolls on through.
An Existential Crisis in Neuroscience
- Asking if we will ever understand the human brain is like asking if you can understand NYC: not a matter of needing more data, just so complex that it’s not applicable
- Instead, we can ask: Can we describe the brain? Still a lot we don’t understand, and this is what much of science is about.
- Biologists often try to “bend the world to their ideas…”
“It’s much better—easier, actually—to start with what the world is, and then make your idea conform to it,” he said. Instead of a hypothesis-testing approach, we might be better served by following a descriptive, or hypothesis-generating methodology. Otherwise we end up chasing our own tails. “In this age, the wealth of information is an enemy to the simple idea of understanding,” Lichtman said.
- For something as complex as the brain (like the stock market), linear pattern of language is not to right tool to use (starting with fundamental concepts). Rather, must deal with the data, how to use and interpret it.
- Use of deep learning by Lichtman, Google to annotate images = helpful; construct DNN identical to human brain? Still, cannot lead to holistic understanding. DNN structure is too simple to model brain; we don’t know what details are important to include.
- Science is ultimately subjective: we choose what experiments to do, interpret and make sense of data.
Research
Single cell epigenomic atlas of the developing human brain and organoids
- Notes deleted, but basically did scRNAseq and scATACseq and compared transcriptome and epigenome trajectories of developing brain from same samples
- Identified similar cell-type clusters, but each contributed unique info
- scATACseq identified enhancers and gene targets for each (compared to known databases, regions with accelerated evolution in humans that are thought to correlate to enhancers for neurodevelopment)
- Also correlated variants associated with developmental disorders to enhancers identified in specific cell-types
- Pseudotime trajectories from chromatin and transcriptome —> correlated, chromatin accessibility “foreshadowed” gene expression
- When looking at organoids, some patterns recapitulated, but not completely (in scATACseq data)
Dynamic lineage priming is driven via direct enhancer regulation by ERK
- Fibroblast growth factor-extracellular signal-regulated kinase (FGF -ERK) signalling drives differentiation of mouse ES cells -> endoderm; inhibiting ERK -> ES self-renewal
- Found that ERK reversibly regulates transcription by changing enhancer activity only, not requiring changes to TF binding! This causes changes to the mediator complex that brings in RNAP, but leaves repressors and activators in place so the cell is primed for pluripotent expression state that fluctuates with ERK activity, which maintains plasticity and ability to re-enter self-renewal when ERK is repressed; if ERK signalling is activated for long enough, TF binding decreases from protein turnover and differentiation becomes permanent
- Used 4-OHT system to test what activation of ERK does; found that it regulates RNAPII binding of TSS and TATA binding protein (TBP)
- Interesting/confusing stuff about TF binding at enhancers and super-enhancers, relation to differentiation
- Cool graphs: Chow–Ruskey diagrams to show overlap of repressed and activated enhancers with changes in TF binding
- Lots of profiling of nascent transcription to look at activation and repression of ERK targets
- This explanation is better than anything I could summarise: “Through shifting the transcriptional machinery and selected cofactors from pluripotency genes to lineage and mitogenic ones, ERK both primes differentiation and ensures that reactivation of the ES cell gene regulatory network can occur within a discrete time window. Thus, signalling triggers a molecular countdown clock, which can be rewound on inhibition and runs out when the pluripotency transcription factors drop below a functional level that supports reactivation. That transcription can be repressed while leaving the occupancy of enhancers by transcription factors intact suggests a general mode for signalling responses that enables cells to preserve developmental potency.“
The Genetic/Non-genetic Duality of Drug ‘Resistance’ in Cancer
- Evolutionary & ecological principles help resolve duality, may offer new therapeutic approaches to overcome resistance
- Resistance = inherited ability for organism to grow at high concentrations of drug, strong genetic underpinning; tolerance = ability (genetic or not) to transiently survive high drug concentrations; persistence = when subpopulation of clonal population to survive when the rest is killed off. However, esp in cancer, terms used interchangeably -> basically just “resistance.”
- Genetic basis for resistance: (Study in BRAFamp in lung cancer and melanoma) Drug(s) exerts pressure that select for and amplify (copy number?) a mutant allele allowing them to survive. Long-term treatment with a single drug enables parallel evolution, so cells maintain intratumor heterogeneity. Mutation(s) or copy number increases required to survive = fitness threshold, differs for each drug; thus sequential monotherapy may just select for progressively more resistance. However, intermittent tx with different drugs can prevent evolution of resistant alterations by constantly changing the fitness threshold.
- Non-genetic mechanism for resistance: Study looked at same gene & cancer as the one above, but different drug, only targets mutated protein -> eradicated almost all cancer cells, but small population persisted -> resistance. Considered a heritable model (permanent transition to resistance) and transient, non-heritable model (can switch between pre-resistant and resistant); used Luria-Delbrück fluctuations analysis (do mutations arise randomly, or are they induced by the selective pressure?) -> no heritable pre-resistant state (no mutations), but rather extreme transcriptional heterogeneity at sc level; v rare cells expressing high levels of resistance markers -> epigenetically reprogrammed in response to drug, becomes heritable (alternative: selective pressure from drug acting on epigenetic heterogeneity). Other studies show that envt conditions can induce epigenetic changes that prime cells for oncogenesis by single mutations (Lamarckian evolution).
- Reconciling the duality: Two mechanism ≠ mutually exclusive: transient changes can allow subpopulation to persist, then eventually acquire heritable genetic or non-genetic resistance by selection; stochastic epigenetic reprogramming (rewiring) -> plasticity to give rise to persisters. Understanding relationship -> can administer drugs in right sequence, combination, frequency to prevent adaptive resistance. Analogous to development of 4 wings by 2-winged flies; induced by mutation to Ubx gene OR environmental exposure to ether (non-mutation), which eventually breed true without exposure: genetic assimilation (trait initially induced by env’t, then selected for and becomes genotype) or canilization (ability of organism to produce same phenotype regardless of env’t or genotype)
- Ecology & evolution: Initially resistance viewed as binary & irreversible, so max dose treatment to eradicate ASAP without resistance or metastasis = best idea; no longer the case. Phenotypic heterogeneity -> resistant clones exist before tx, so max dose just selects for these; alternative approaches (using min possible dose to avoid symptoms, intermittent therapy so sensitive cells outcompete resistant ones) may do better; maybe treat with two drugs in series rather than together to avoid double-resistant clones
Metabolic signatures of cancer cells and stem caells
- Cancer cells and stem cells both retain ability to re-enter cell cycle and proliferate, involves changing metabolic state —> decisions around requiring metabolic pathways closely intertwined with cell fate (in cancer and dev)
- In multicellular organisms, growth factor signalling controls anabolic and catabolic pathways to control growth/death of different populations, organs; divided up constant nutrient supply (vs unicellular: more growth with more nutrients) —> growth factor signalling pathways important to development are often dysregulated in cancer
- Metabolites not only act as substrates for anabolic growth, but involved in signalling and regulation —> differentiation and cell fate
- Pluripotent stem cells (PSCs) — can give rise to all three germ layers and proliferate endlessly; ideal model system to study relship of proliferation metabolism, differentiation; lineage determined more by growth factors and nutrients in media than origin
- Key question: how to distinguish metabolic changes supporting proliferation from those involved in differentiation? Comparing metabolism of cancer cells (more the first) and stem cells (more the second) can be helpful.
- Requirements for proliferation: glucose and glutamine —> increase biomass and make biomolecules
- Aerobic glycolysis (Warburg effect) = hallmark of rapidly-proliferating cells (cancer and PSCs); discard glucose carbons as lactate despite sufficient oxygen to completely oxidise; interfering with glycolytic pathways reduces proliferation. Uncertain why this happens… bioenergetics, oxidation?
- Other pathways also implicated in cancer proliferation when you look in vivo, depending on micro environment and cancer type; both PSCs and cancer cells can catabolise amino acids in unconventional ways
- Regulation of anabolism: growth-factor mediated activation of RTK pathways like Ras, P13K, to activate effectors like MEK/ERK, mTOR, AKT —> all very common oncogenes. MYC also upregulates glycolysis; loss of p53 switches from nutrient catabolism to anabolism. HIF1 (from environmental and genetic factors) triggers glycolytic switch in tumors and primes PSCs for differentiation.
- Same pathways regulate and maintain PSCs, but much less studied in this context.
- Interesting difference: MEK inhibition —> maintains naive PSC state, but may impose metabolic limits —> can be rescued by also inhibiting GSKB (but cancer cells can not proliferate with MEK/ERK inhibition)
- Metabolic regulation of chromatin and differentiation: key metabolites = substrates and co/factors of epigenetic effector enzymes, esp. histone acetylation and DNA and histone methylation
- Acetylation: acetyl-CoA can’t leave mitochondria unless made into citrate; enzyme ACL cleaves it in cytosol or nucleus back into acetyl-CoA for use. Oncogenic signalling (P13K/AKT pathway) activates ACL, maintain necessary gene expression. In hypoxic conditions, acetyl-CoA is limiting, and enzyme ACSS2 is needed to synthesise it from free acetate; tumors consume lots of acetate. Also dependent on pH, could be a source of heterogeneity. In PSCs, decreased acetylation —> differentiation (also bc decreased glycolysis limits citrate availability), but in other multipotent progenitors it drives lineage-specific reprogramming.
- Methylation: repressive for DNA, either for histones. Hypermethylation of DNA = hallmark of cancer (silence tumor suppressors), also contributes to heterogeneity. PSCs have much more dynamic chromatic accessibility.
- Methyl-transferases get their methyl groups from SAM, which is generated from things related to folate cycle. Dysregulation of SAM metabolism is common in cancer; decrease of SAM availability also triggers differentiation.
- Removal of methyl groups catalysed by LSD, JHDM, TET which depend on lots of co-factors linked to metabolism things. Hypoxic conditions in tumors —> depletion of glutamine —> more repressive histone methylation, promotes de-differentiation and drug resistance. Much more context-dependent effects in PSCs.
- Mutations to IDH enzymes that convert substrates of these enzymes = oncogenic, lock cells in de-differentiated state. Mutations to other enzymes (SDH, FH) lead to accumulation of metabolites that stabilise HIF1 in a pseudo-hypoxic state, promote EMT
Nature Method of the Year: Single-cell Multimodal Omics
- Notes available here
Developmental barcoding of whole mouse via homing CRISPR
- Reza Kalhor @ JH; homing guide RNA (hgRNA) —> targets its own locus to diversify sequence, acts as barcode
- Developed mouse model with 60 hgRNA loci —> lineage trace from embryogenesis through adult development; looked at embryogenic brain patterning
- Previous barcoding work used single elements; using multiple components exponentially scales information capacity when they are independent: 1) no cross-talk between elements and 2) no interference between mutational outcomes
- Crossed MARC1 mice (with the 60 hgRNA loci) and Cas9 knock-in mice to activate system; could generate 10^23 barcodes with just 10 hgRNAs; had a range of activity levels, so could get resolution on different time-scales throughout development; could carry out bottom-up reconstruction of lineage from zygote onward
- When looking at brain, found that barcodes on L and R sides of same region were more similar than different regions of the same side, suggesting that commitment to anterior-posterior axis comes before lateral axis (similar barcodes = more closely related)
- (Hi Ravi) Use of E. coli with Cas1/Cas2 recording system to track HGT in microbial communities, which previously could only really be done for plasmids with some sort of phenotypic marker
- Prime system to preferentially uptake exogenous spacers -> expansion of array means HGT (or phage infection) occurred
- Amplify and deep-sequence spacers to see what the transfer was, bioinformatics to ID
- Able to detect rare and non-replicative events — tested in clinical samples; found that plasmids vary in transfer efficiency
- Limitations: will only detect transfers that can occur to E. coli; since so much E. coli naturally in human feces, interactions with recording strain may be less common; E. coli not naturally competent so will not detect free eDNA uptake; recordings in clinical samples performed aerobically > only certain donors active; only 2% of arrays expanded, so need to expansion rate to get better capacity
Exon-Mediated Activation of Transcription Starts
- (Just from abstract)
- Evolutionary gain of /internal exons/ associated with gain of new TSS nearby -> increased expression
- Suggests splicing of internal exons impacts promoter choice, expr level by recruiting transcription machinery; important in evolution of new promoters, gene expr in mammals
- Effect goes both ways: inhibiting splicing -> less expression; creation of new spliced exons -> transcription from “cryptic promoters”
- Strongest effect for weak promoters near & upstream of efficiently-spliced exons
Biodiversity Alters Strategies of Bacterial Evolution
- Why do some bacteria evolve CRISPR systems against phages, while others have a more permanent surface receptor defense? Depends on resource availability: CRISPR system is more risky, because need to “inoculate” and then recognize and respond, but also much less costly, because only takes up resources when active. Surface receptor (which evolves more in lab conditions) is a safer bet, but much more costly— always taking up energy and nutrients.
- Similarly, difference depends on other species & strains in community… even in resource-rich environments, more likely to evolve CRISPR system because can’t “afford” to go the surface receptor route, which would make a strain less fit for survival vs competing ones — esp. when the competitors already have a CRISPR system or are resistance in some other way.
Terminator-free template-independent enzymatic DNA synthesis for digital information storage
- New DNA synthesis strategy, uses template-independent polymerase, terminal deoxynucleotidyl transferase (TdT) to makes DNA with short repeated strings of nucleotides -> store info in transition between non-identical nucleotides
- To control reaction, use apyrase to degrade dNTPs to TdT-inactive form and monophosphate version -> limits polymerisation by competing with TdT for dNTPs; figured out reaction conditions so that at least one nucleotide of each base could be added. Then, you can add the two enzymes and the dNTP and be sure at least some has been added, wash, and add the next one to make custom strands. They may each have different numbers of each nucleotide, but identical transitions!
- To “read,” sequence raw strand and store transitions as “compressed” strand
- Lots of stuff about the “codec” and methods for readout that have math
- Use of enzymatic method can improve on cost, quality, quantity of current organic methods
- pgFARM (paired guide RNAs for alternative exon removal) = “CRISPR–Cas9-based method to manipulate isoforms independent of gene inactivation”, link mid-spliced isoforms to specific molecular pathologies. Basically, just deliver two gRNAs to cells with dox-inducible Cas9 to delete the section between them.
- Validated that this works, doesn’t cause inversions, screened for large deletions (long range gDNA PCR??— just use a “long range” Taq)
- Poison exons: alternative exons that disrupt their gene’s ORF to trigger “nonsense-mediated RNA decay”; likely play important role in development but really hard to study, not sure what most actually do. 481 ultra-conserved regions in human and mouse genome, many overlap poison exons.
- Looked at function of ultra-conserved poison exon in SMNDC1 (highly expressed in lung cancer); found it modulates intron retention, which happens a lot in cancers; validated in other databases (poison exon exclusion and higher gene expression happened more in cancer, associated with poor survival)
- Used pgFARM library to screen 500 poison exons (identified from 12k by looking at most conserved across species, association with NMD); also screened corresponding constitutive exons in each gene to compare effect vs total gene loss. Did validation of library, that it worked as a dropout screen, looked at off-target effects, DSB toxicity effect, and functionally validated several hits.
- Another dropout screen -> are poison exons important for cell fitness? Poison exons typically included were depleted vs those typically excluded; skipping of many poison exons associated with only moderately lower fitness costs than knocking out the whole gene -> conservation likely explained because they are very important to fitness
- In vivo experiments -> some poison exons = tumor suppressors (anti-tumorigenic), even though many others promote cell growth
- Laser-induced Cre + loxP-DsRed-loxP-GFP reporter, validated by studying hematopoeisis, proved they could label sc in brain, etc.
- Previous IR-LEGO technique uses IR-laser to generate precise heat shock that induces Cre expression, permanently labelling cells green; limitations in balancing cell viability with efficient labelling after heat shock, permanently tracking cells, getting single-cell precision, statistical analysis to lineage trace; other techniques also not sufficient
- Here, improved on IR-LEGO by measuring temp rise with two-photon thermometer -> precise labelling
- Lots of thermodynamics, optics on ability to control location and intensity of heat shock necessary in different conditions, then validation
- Inability to multiplex, but can visually track where the cell(s) go over time
CRISPhieRmix: a hierarchical mixture model for CRISPR pooled screens
- Method for better design of CRISPRi/CRISPRa screens, addressing problems of 1) variability in gRNA efficiency, 2) large rare off-target effects; makes it difficult to systematically determine whether a gene is a hit when different guides have different effects without prior knowledge of function.
- Instead of using normal hypothesis testing model with global FDR, use mixture deconvolutional approach to find local FDRs; hierarchical mixture model to account for variable guide efficiency; and broad-tailed distribution based on negative controls to account for rare off-targets -> lower FDR and higher true positive rates.
- Model finds more genes than other algorithms and performs same/better on other datasets; also suggests most screens don’t use enough guides for full discovery.
- Lots of (very clear) description of model development that I will read again in the actual paper
- Perspective (edited by Charlie Gersbach!): stochasticity and heterogeneity -> sc resolution & synthetic genes, manipulation needed to build predictive model of gene regulation; summary of recent techniques looking at chromatin & how to combine to make “function”/framework encompassing all mechanisms
- Inputs: methods to manipulate & measure chromatin modification, TFs & polymerases at enhancers & promoters, nuclear localization, chromatin architecture (lots of Cas9 applications, sc tech); Output: methods to measure mRNA and protein levels (fluorescent tags/reporters, all the FISH)
- When measuring mRNA, consider transcriptional bursting -> negative binomial distrib, not Poisson (high cell-cell variability)
- Use of all these methods (summarized in tables) reveals heterogeneity, stochasticity– some cells completely silenced, different epigenetic “memories” depending on time scale & cell type
- For building a model: ex. transcriptional bursting: two-state with k_on, k_off, or three-state with a reversible and irreversible off state… how do different inputs affect each rate? Similar models for chromatin accessibility and modifications
Dynamics of epigenetic regulation at the single-cell level
- Time-lapse microscopy to study effects of four chromatin silencers (EED, KRAB, DNMT3B, HDAC4) with sc resolution; targeted to artificial chromosome in dox-inducible system -> all acted all-or-nothing per cell, but different fractions, durations, rates -> different kinds of memory
- KRAB and HDAC4 -> silencing within 1 cell cycle, the others took longer; still, Toff (time to silence after induction) varied for same silencer between sister cells -> stochasticity. After removing inducer, reactivation was similarly stochastic for all silencers except DNMT3B, which had no reactivation -> memory; KRAB completely reversed, but the other two stabilized with semi-silenced populations (suggesting some cells were reversible silenced, others irreversibly). Fraction at stabilization also depended on time/strength of recruitment to chromatin.
- Modeled by three-state system (on, reversibly-off, irreversibly-off)
- Cool graphs of the three different states, study of dynamics
Self-assembling manifolds in single-cell RNA sequencing data
- Better analysis method for scRNA-seq data = (dimensionality reduction by feature selection, differentiating biological vs technical noise). How to select the genes (the signal) that matters to identify cell types and states?
- Self-assembling manifold = SAM:
- Randomly group cells (random kNN graph) & look for genes with different expression between groups (compute variation in neighborhood)— “averages out” individual differences
- Reclassify cells according to this info (compute distance matrix, reassign), and repeat until stability is reached
- Unsupervised “soft feature selection,” reweights genes that are more spatially separated, and uses these weights for next nearest neighbor assignment. Overcomes thresholding challenge bc rescales all genes, and isn’t as influenced by individual cell variability
- Show that it works better at identifying cell types than other methods; don’t just use benchmark datasets with very clear subsets, but ones where feature selection is difficult
- When tested on difficult dataset, could separate cell types when other algorithms couldn’t; only ~1% of genes contributed. Final outcome is independent of initial conditions (random or output of another algorithm); proved that it doesn’t find structure in actual random data.
- Doesn’t get “fooled” by highly-variable genes, since it uses “neighborhood” expression; also less sensitive to data corruption (?)
- Selected genes also biologically meaningful, its ranking be used with GSEA
- Hematopoietic system = good to start with because cells can be phenotypically sorted; transcriptional heterogeneity observed from scRNA-seq -> here, used scATAC-seq on 10 populations of multipotent and lineage restricted progenitors
- Found “continuous” regulatory landscape, evidence of “priming” for certain lineages
- Common myeloid progenitors (CMPs) and granulocyte-macrophage progenitors (GMPs); GMPs = heterogeneous in epigenome and transcriptome; strategy how to enrich for subpopulations of GMPs at different stages of differentiation
- Integrate with scRNA-seq data -> link TF expression with chromatin accessibility at cis-reg places, and changes in expression of nearby genes
- Method of dimension reduction: 1) find PCs of bulk ATAC-seq, then 2) score single cells on contribution of each PC, 3) cluster by Pearson coeffic of normalized PC contribution & other cells, 4) PCA on this matrix and visualize by 1st three dimensions
- Identified trajectories, TF dynamics across differentiation (pseuodotime)
- Data generation methods and computational workflows -> applicable to combining other omics datasets (not just having to do them IRL at the same time?)
Discovering the anticancer potential of non-oncology drugs by systematic viability profiling
- Pooled screen of several thousand non-cancer drugs on several hundred cancer cell lines (vs previous studies, small amount or one and tons of the other) -> many had potential anti cancer effects
- Used PRISM (profiling relative inhibition simultaneously in mixtures): barcode cell lines; relative mRNA abundance is proxy for survival from pooled samples
- First screened at one dose; then chose ~1.5k hits to screen at 8 doses; results slightly noisier than standards but generally validated
- Clustered viability profiles of drugs with UMAP -> clustered by function (including clusters of non-anticancer drugs!), suggesting mechanism of action captured by PRISM
- In general, chemotherapies killed most, then targeted cancer, then others; however, dose dependent: targeted therapies not as selective at high doses (similar to chemotherapies); analysis of bimodality coefficient of dose-wise viability curve (uniform distrib = 5/9, more bimodal = closer to 1) -> some of most selective compounds = normal drugs
- Built predictive model of cell-killing activity for each cell line based on genomic features, dependencies from other screenings (from random-forest based ATLANTIS algorithm) -> surprising number of non-cancer compounds had high predictability; mRNA expression = most important predictor (not mutation!)
- Plotted compounds by bimodal coefficient (selectivity) and predictability (compared to CRISPR screen data for certain gene expression) -> further explores compounds ranking highly in both that were associated with four potential predictive biomarkers with highest both of these.
Lineage tracing on transcriptional landscapes links state to fate during differentiation
- How to identify cell fate from transcriptional state of progenitor cells? Use DNA barcodes to lineage trace + track transcriptome -> study hematopoiesis; locate states of “primed fate potential” on transcriptional landscape; however, still cases where intrinsic fate choice cannot be detected & happens before algorithms can detect
- Strategy: DNA-barcode (28-mer in eGFP 3’ UTR = LARRY, Lineage And RNA RecoverY) heterogenous population, let divide, capture some to sequence, let the rest go and sequence later. Reveals three types of clonal relationships: 1) sister cells at earliest time point (state-state); 2) clones at both early and late time points (state-fate coupling); 3) differentiated cells later on (fate-fate clonal dynamics)
- State = transcriptome, sisters = clones
- If sister cells (1) are transcriptionally similar, clonal related pairs over time (2) should reveal changes in sc gene expression leading to differentiation
- Sampling a clone -> estimate how a single cell changes over time, but depends on sisters being similar at early time points. Most sister cells had correlated transcriptome (similar state), but some diverged. Identified clones -> one fate; other clones gave rise to multiple fates. Progenitors of diff fates = along continuum, not discrete groups; bipotent progenitors -> extended ‘fate boundary,’ vs traditional stepwise, hierarchical differentiation path; evidence for ‘priming’ in expression patterns (certain marker genes); in vivo, fate was even more heterogenous among clones, depended on cytokines/envt conditions
- How predictable is fate from scRNA-seq? Doesn’t capture envt, chromatin, stochasticity. Tried logistic regression and NN with several gene sets as features; several hundred highly variable genes = best predictors (even vs TFs); accuracy of ~50% in vivo
- Are there stable cellular properties (hidden variables) affecting fate that aren’t picked up by scRNA-seq? Compared prediction accuracy of ‘early’ vs ‘late’ clones -> YES, hidden variables exist; scRNA-seq cannot predict/perfectly capture cell fate (just like FACS couldn’t!)
- Clonal analysis -> different routes of differentiation that scRNA-seq alone does not reveal
- Benchmarked common approaches to monitoring cell state dynamics with clonal tracking data -> clonal analysis needed to select predictors; pseudo time analysis does capture progression
- Gastrulation = differentiation of germ layers, shaping of body plan during mammalian embryogenesis; extraembryonic layers sustain embryo.
- Can culture human “gastruloids,” from embryonic stem cells with 7 cell types and radial organization (system supports “primate-specific” features)
- When gastruloid is dissociated, single cells are motile, cluster with same/similar cell types; cross-species comparison -> sorting is evolutionarily conserved.
- A primer on embryogenesis: First major lineage split: blastocyst -> trophectoderm (TE), hypoblast (primitive endoderm), pluripotent epiblast. Epiblast -> all embryonic tissue; TE and hypoblast -> extraembryonic (ExE) cells, which provide signals for polarity, germ layer formation. TE = polar (near epiblast) & mural (distal).
- Murine & primate blastocytes similar pre-implantation, but differ post: murine implant using mural TE, primates with polar. In mice, germ layers arise from epiblast elithelium, vs in primates, epiblast cells near polar TE -> amnion, and those near hypoblast -> epithelium -> germ layers.
- Gastrulation = highly conserved in animals. Mice and primates: primitive streak (PS) forms at posterior epithelial epiblast; cells undergo EMT -> enter PS, migrate to form mesoderm & endoderm; remaining cells -> ectoderm. At same time, primordial germ cells (PGC) specified in epiblast for mouse, amnion for primates. In frogs, fish, cell sorting -> maintenance of germ layers; when gastrulation cells re-aggregated in vitro, segregate back to layers; unknown if this happens in mammals.
- Mouse ESCs cultured in vitro have been shown to recapitulate patterns of in vivo gastrulation. Human ESCs, cultured on ECM microdiscs & stimulated with BMP4, also do this (differentiate into radial layers, EMT), but complexity unknown.
- Here, validated method of hESC gastruloids; scRNA-seq -> 7 cell types; exhibited cell sorting and motility
Super-resolution labelling with Action-PAINT
- STORM, PALM, PAINT (point accumulation for imaging in nanoscale topography) enable imaging of single molecules @ 5 nm resolution, but does not allow for manipulation
- PAINT: population of fluor. affinity probes transiently & repeatedly bind to target -> “blinking” that can be synthesized to localize target
- Action-PAINT 1) monitors and localizes DNA-binding event with DNA-PAINT, then 2) when binding to desired location is visualized, photo-crosslinks DNA to attach molecular label
- With PAINT, advantage = usually only one probe is bound in diffraction limited area. Thus, if induced at the right time, can specifically “choose” that target to be labelled with flash of light
- Uses nucleoside (CNVK) for photoinducible-crosslinking
- Three steps: first, pre-acquisition imaging (with PAINT) to select target; second, administer light pulse to label target; third, post-acquisition confirmation
- Target molecule (ex. a microtubule) tagged with primary antibody; PAINT labels with secondary, then Action-PAINT adds the third fluorescent tag
Immuno-SABER enables highly multiplexed and amplified protein imaging in tissues
- Immuno-SABER = immunostaining with signal amplification by exchange reaction; amplification by DNA-barcoded Abs and orthogonal concatemers generated by primer exchange reaction (PER; cool synthesis method); works on lots of sample types, get 5-80-fold amplification on up to 10 targets at a time; can be combined with ExM
- Previous methods could do good signal with fewer Abs, or 10ish targets with DNA but bad signal; took a long time
- PER
- How it works: 1) Stain target with DNA-tagged Ab AND make concatemers with PER (one part matching to DNA tag, other part = tail for signal amplification); 2) Hybridize bound Ab and & concatemers; 3) Hybridize imager probes (DNA complementary to the tail with fluorophore) 4) If multiplexing, exchange imagers and repeat till done
- Can even do “iterative” Immuno-SABER to attach multiple rounds of concatemers and amplify signal more
Podcasts
- Honestly, not to many impressions here. Interesting discussions on what policy priorities should be for the next Democratic president — should we do something with immediate benefits and low cost, or vice versa? Ex. health care vs climate change?
Throughline: Resistance is Futile
- Technologies from coffee (!!) to tractors faced societal & government backlash, then became widely accepted. That’ll probably happen with AI and virtual assistants, too.
- Dave Eggers lives his own way, and is so okay with it. No wi-fi, no smart phone… what a guy.
- Being a “morally good” person is terrifyingly simple, and also extremely difficult— depends on context you’re considering: are you doing good relative to your peers? What about relative to the absolute most good you could do?
- The idea that expanding our /circle of partiality/— family, friends, even humans above other living beings that we treat preferentially— is a part of this
- The idea that donating half your income may actually make you happier than if you spent it on yourself & loved ones: would give your life a clear meaning/purpose, while buying consumer products only induces short term happiness
- How do non-useful things fit into life — even reading novels vs nonfiction, etc.? Can consider pleasure it provides you, but ultimately, the former have less value; don’t improve the well-being of sentient creatures.
- Power of social reasoning -> actually more effective to be public about giving, because it changes norms and can convince others to (vs conventional Jewish, Christian values of anonymity)
Books
The Wildlands — Abby Geni — 1/2/20
- A 75%-engaging story trying a little too hard to convey a not-so-engaging environmental message.
On worry: _ He was telling her not to dwell on the past or fret about the future, since every moment was followed by another, some wonderful, some terrible, all unpredictable and unknowable beforehand, all essential components of the complexity of a vast and marvelous world._
Dept. of Speculation — Jenny Offill — 1/3/20
- Authentically stream-of-consciousness; almost painfully honest.
On homes: _ The reason to have a home is to keep certain people in and everyone else out. A home has a perimeter._
On things: _ A thought experiment courtesy of the Stoics. If you are tired of everything you possess, imagine that you have lost all these things._
Atlantic — Simon Winchester — 1/3/20
- My impression: stormy, stormy sea. Just could not get into it any more than that.
The Odyssey — Homer, Emily Wilson (translator) — 1/8/20
- Beautiful, in its truth to both contemporary and classical times. I smiled every time I pictured rosy-fingered Dawn and the purple sea.
Winners Take All — Anand Giridharadas — 1/9/20
- This made me think hard about my life and my dreams to “change the world.” Gave me the language and framework to more carefully consider, analyze, and interpret society in a more complete way.
Kochland — Christopher Leonard — 1/17/20
- A sprawling, thorough, and fascinating portrait of Charles Koch’s life work. Learning the human aspect behind a worldview I disagree so profoundly with is complicated, but I think ultimately lays the ground for more constructive conversation.
The Life You Can Save — Peter Singer — 1/21/20
- Reframes what it means to be a good person— draws the difficult but logical analogy between walking away from a drowning child because you don’t want to get your shoes wet, and not donating to proven, effective charities that save lives when we have the means to do so.
- Offers an alternative perspective to private philanthropy than Winners Take All, suggesting that it really is the ethical responsibility of each person to do the most good they can, and that donating to non-profits is often the most effective way to do so. Singer cites the multitude of flaws with governmental aid programs (that we individually have little control over changing) and the lack of evidence for systemic changes that can actually address extreme global poverty. Regardless, both Singer and Giridharadas emphasise the same ultimate point: the rich can’t keep getting richer and still save lives.
Strangers Drowning — Larissa MacFarquhar — 1/24/20
- An exploration of the spectrum of morality, giving, and helping. MacFarquhar addresses the themes of my previous few reads in the most human way, through history, philosophy, and, most impactfully, the stories of a series of ordinary, extraordinary do-gooders.
The Darker the Night, the Brighter the Stars — Paul Brok — 1/27/20
- … A Neuropsychologist’s Odyssey Through Consciousness… a somewhat jarring, absurd collection of clinical cases, Greek myths, philosophical musings, and personal stories; forming somewhat of a web of ideas on the mind, self, consciousness, imagination, and reality. Unique, but not particularly engaging.
- Reality is what is left when we are gone: thus, what of self, of consciousness?
- Consciousness is an action, something we do, rather than a thing; made up by the actions of sensing: evolved, specialised, & privatised analogues of biochemically responding to a stimulus.