3Distributions Matter Are these groups of cells the same? But they've got the same mean fluorescence…?
4Engineering must consider variation Biological systems exhibit large variation in behavior due to many classes of cause, including:Inherent process stochasticitye.g., transcription, translation, replication, …Cell-to-cell differencese.g., size, cycle state, health, mutations, location, …Protocol stochasticitye.g., transfection variation, insertion site, …Protocol execution issuese.g., reagent variation, contamination, instrument drift, ...System engineering must handle each class differently!Largely uncorrelated effectson individual system elementsHighly correlated effectson individual system elementsPredictable distributions, affectedby choice of protocol parametersUnpredictable, must be detectedand appropriately compensated
5Example: Metabolic Process Composition Enzyme 1Enzyme 2ABCInputIntermediateProductWhat is the effect of variation in enzyme performance?Different classes of variation have different effects:uncorrelated per-cell variation decreased overall variationcorrelated per-cell varation increased variation, selection pressureprotocol variation batch-to-batch "fickleness"
6Example: Detecting Protocol Failure Bad TransfectionGood SamplesBad SamplesDistribution ModelBy modeling transfection distribution, can detect failure of batches or individual samples[Beal et al., 2012;Kiani et al., 2014;Davidsohn et al, submitted]
7Per-Cell Measurements Reveal Variation Example: RNA Replicon CotransfectionExample: Constitutive mKate in HEK293Cell VariationDose VariationQuantizationDose EfficiencySpectral BleedAutofluorescenceDose/Resource VariationTransfectionEfficiency[Beal et al., 2012][Beal et al., submitted]Quantifying variation components requires per-cell measurements of large populations of cells
9Output R2 R1 First, some metrology… Unit mismatch! [Output] [R2] [R1] Arbitrary RedArbitrary Blue[R2][R1][Output]R2R1Unit mismatch!
10Good Engineering Needs Absolute Units Metrology 101: precision measurement enables:Comparison of results across experiments and labsDeeper insight into the behavior of devicesEffective dissemination of materials and methodsTesting and validation of materials and systemsEstablishment of commercial & industrial servicesSafety assurance and traceability of responsibilityMany biological measurements are only relative!
11How Flow Cytometry Works Challenges:AutofluorescenceVariation in measurementsSpectral overlapTime ContaminationLots of data points!Different protein fluorescenceIndividual cells behave (very) differently
12Metrology vs. Flow Cytometry Flow cytometry great for per-cell measurements, but…Arbitrary unit output depends on…Instrument brand, configurationInterference from other colorsChoice of instrument settingsRun-to-run calibration driftThese are notphysical units!Blue (Arbitrary Units)Fortunately, this can be corrected…Yellow (Arbitrary Units)
13Calibrated Flow Cytometry Standardized Units (MEF)Autofluorescence Subtraction[Roederer, 2002;Wang et al., 2008;NIST/ISAC, 2012;Beal et al., 2012;Kiani et al., 2014;Davidsohn et al, submitted]Non-Distorted CompensationColorEquivalence MappingResult: replicable measurements in absolute units (MEFL),rather than non-replicable relative units as usual practice.
14Example: Predicting Repressor Cascades Precision dose-response measurement allows high-precision prediction with quantitative modelspCAGDoxT2ArtTA3VP16Gal4pTREEBFP2R1pUAS-Rep1pUAS-Rep2EYFPR2mkatePrediction of Repressor CascadeRange vs. Error for 6 CascadesTAL14 TAL21+: experimentalo: predictedEach line is a dose/response curve for a different relative number of circuit copies.Subpopulation identified by color on inset mKate histogram[Davidsohn et al., submitted]
15How much does calibration matter? 8x tighterrange just bycalibration!(2.8x better high errors,2.8x better low errors)[Davidsohn et al., submitted]
16Example: Engineering Replicon Expression Per-cell measurement of dose-response gives model allowing high-precision control of expressionmVenusnsP1-4SGPExample:Prediction of fluorescence vs. time for novel mixtures of 3 Sindbis RNA repliconsMix 1: 0.1Y, 0.1R, 0.1BMix 2: 0.3Y, 0.3R, 0.3BMix 3: 0.1Y, 0.5R, 0.4BMix 4: 0.2Y, 0.2R, 0.6BMix 5: 0.01Y, 0.1R, 0.5BMix 6: 0.4Y, 0.02R, 0.02BmKatensP1-4SGPEBFP2nsP1-4SGPExample Prediction of 3-RNA Replicon Mix:Range vs. Error for 6 MixturesMix Number[Beal et al., 2014]
17Example: Device Engineering Absolute comparisons, segmentation vs. constitutive fluorescence allow better design and debugging of devicesExample: CRISPR Device DesignDistribution analysis identifiesintrons as a problem areaOriginal intronic design:Modified design tofix intron issues:Device ig-bNormal TREinductionExpected high-copy leakDevice ig-cMax ~5xrepressionMax ~20xrepressionno response &no leaky expression intron failure[Kiani et al., 2014]
19TASBE Public Web Interface https://synbiotools.bbn.com/On first use, you will have to terms of serviceYour data is secure, and can’t be shared on site.FireFox recommended; Chrome has an image-display bug.Register: individual accounts or group account? Anonymous access also available (but not private)
24Fluorescent Beads Absolute Units SpheroTech RCP-30-5ATool currently only supports SpheroTech RCP-30-5ARun beads every time: flow cytometers drift up to 20 percent! Also can detect instrument problems, mistakes in settings
25Compensating for Autofluorescence Use your negative control for this
26Compensating for Spectral Overlap Use a strong positive control for each colorNote: only linear when autofluorescence subtracted[cf. Roederer 2002]
27Translating Fluorescence to MEFL Only FITC channel (e.g. GFP) goes directlyOthers obtained from triple/dual constitutive controlsMust have exact same constitutive promoter!Must have a FITC control protein!This is a new control that you need to add
28Creating a Color Model (1/5) When you pick the machine, wait a moment and available channels will fill inPick allchannelsused
29Creating a Color Model (2/5) Negative controlCalibration beadsSimilar entries for single positive controls
30Creating a Color Model (3/5) Must have at least FITC-X for each other fluorescence XIf you only have two colors, repeat one here
31Creating a Color Model (4/5) The same control set can be run with different parametersMake sure you pick this right!Leave these alone unless there are model problemsFree choice of labels
32Creating a Color Model (5/5) Results:Lots of graphs for you to check result validityOnline model for one-variable experimentsMatlab .mat file that you can use for calibration by hand
33Bead problems? Usually means either: you didn’t run enough beads, oryour low peaks are being affected by sensor noise, oryou've got a serious flow cytometer hardware problemFirst step: adjust peak & minimum thresholds
35One-Variable Characterization Rapid, high-precision device characterization with reproducible absolute unitsOutputDoxR1pCAGDoxT2ArtTA3VP16Gal4pTREEBFP2R1pUAS-Rep1EYFPmkateTransfer curve for TAL 14Transfer curve for TAL 21Each line is a dose/response curve for a different relative number of circuit copies.Subpopulation identified by color on inset mKate histogramR1 = TAL14R1 = TAL21
36Experiment Assumptions Each functional unit on its own plasmid(but you can work with combined systems too)3 colors: input, output and constitutive(but you can kludge your way down to two colors)Significant variation in circuit copy number(but you can get population statistics too)
37TASBE Characterization Method OutputDoxR1pCAGDoxT2ArtTA3VP16Gal4pTREEBFP2R1pUAS-Rep1EYFPmkateTransient cotransfection of 5 plasmids Calibrated flow cytometry Analysis by copy-count subpopulations
38Multi-plasmid cotransfection!?! Avoids all problems with adjacency, plasmid size, sequence validationsVariation appears to be independent
39One-Variable Analysis (1/5) Make sure you'll be able to understand these several months from now
40One-Variable Analysis (2/5) If you only have 2 FPs, make Input = Output.If you only have 1 FP, make all the same
41One-Variable Analysis (3/5) Each column is a replicateEach row is an independent variable valueAdjust rows, columnsUnits ignored - you can do time, voltage, whatever
42One-Variable Analysis (4/5) Expected range of fluorescence(log10)Generally leave these alone
43One-Variable Analysis (5/5) Results: Matlab .mat file, CSV file, lots of graphsThree key graphs, others are preliminary:Input InductionConstitutiveHistogramInput/OutputPopulation graphs too, but they are delicate… better to compute from .mat
44SummaryCalibrated flow cytometry enables order-of-magnitude increase in accuracyTASBE "color model" tool calibrated unitsTASBE "characterization" tool does single-variable experimentsBest with 3-color mammalian cotransfectionsAlso works with other organisms, less colors, etc.Online tools produce graphs and Matlab files