Growth Rate Inhibition Metrics Correct for Confounders in Measuring

© 2016 Nature America, Inc. All rights reserved. contributed equally to this work. Correspondence should be addressed to P.K.S. ( 1 undergo cells control if approach): standard (the experiment the of end the at made counts cell from estimated are they when IC as such metrics response understood poorly remain that from one study to vary data next the drug-response biomarkers drug-response discover to combined often are datasets genomic area under the dose–response curve (AUC) ( concentration drug at highest cells the viable at which the cell count is half the control (IC to ted compute to curve of a (i) the concentration drug sigmoidal are by controls fit for counts untreated divided of drug presence days later.in counts the cell comprising several Data is measured CTG) CellTiter-Glo, using assayed ATP as level such surrogates, over a and (or of the number cells range of viable concentrations, In the case of anticancer drugs, cells are typically exposed to drugs approaches biology chemical using processes biological other of action ery of therapeutic molecules, the investigation of their mechanisms The quantification of drug response is fundamental to the discov drugs effective against specificpatient-derived tumor cells. discovery of drug-response biomarkers and the identification of growth using small molecules and biologics and to facilitate the We expect G requires only modest changes in experimental protocols. molecule drugs in dividing cells. conventional metrics for assessing the effects of small assays. We show that G growth rate inhibition (G T drug-response metrics that are insensitive to division number. with biomarker discovery. We derive alternative small molecule while obscuring valuable biological insights and interfering artefactual correlations between genotype and drug sensitivity, assay. number of divisions taking place over the course of a response IC D Marc Hafner in measuring sensitivity to cancer drugs Growth rate inhibition metrics correct for confounders Received HMS HMS LINCS Center Laboratory of Systems Pharmacology, Department of Systems Biology, Harvard Medical School, Boston, Massachusetts, USA. hese are based on estimation of the magnitude of drug-induced rug rug sensitivity and resistance are conventionally quantified by 50 e hw ee ht fr iiig el, rdtoa drug- traditional cells, dividing for that, here show We or T E

he he dependency of 2 max 1–

Oct 3 , and the study of signal transduction, cell division, and 1– values, but these metrics are highly sensitive to the o R be 3 metrics to improve the study of cell signaling and , 8 r r 1 , 9 , 2015; 2 , but it has recently been found that large-scale large-scale that found been recently has it but , , , Mario Niepel

accepted R 50 R IC and G ) ) using endpoint or time-course 50

1 50 and A suffer from a fundamental flaw flaw fundamental a from suffer 11– p r R M il max 1 1 3 E oreover, adopting G

2016; , . max 2 are superior to , , Mirra Chung on division rate creates

published 6 , 50 7 . Dose–response and ), ), (ii) the fraction of E max o nline ), and (iii) the ), and (iii) 1 0 for reasons 1 R

& Peter K Sorger 2 metrics

May

2016; 4 , [email protected] 5 - - .

d o i :10.1038/n conditions may not predict sensitivity under slightly different different slightly under sensitivity predict not may conditions assay of set arbitrary) (potentially one under sensitivity predict that biomarkers Thus, biology. underlying the in changes any of IC experiment, an of duration the in changes or conditions, growth in assay variation rate, an proliferation of in differences course natural of the because during divisions of numbers different time time at value GR the computing by measurements drug-response on of panels across curves (compare fell time division malized to a drug-naïve control in which cell number increased as ( increased IC lines, cell faster-growing E cells did not double in an assay run over typically 3 days, and thus ( line cell dividing slowly the In lines cell to the lower quartile, median, and upper quartile for breast cancer sion times ( divi different and cells) kill not but does arrests that drug a (i.e., drug cytostatic a to sensitivity identical with lines cell idealized We used computer simulation to model the drug response of three D RESULTS assay. the of course the over divide cells control which in perturbations other and RNAi, drugs, to metrics replace IC these that propose we and procedures, experimental in changes modest with can be determined GR metrics Methods). in Online curve’,GR the over rate and (we of ‘areause assay duration division cell independent GR presence and absence of the Parameterization drug. of GR in data yields rates growth of comparison the on based is which (GR), inhibition rate param growth for normalized the response, method drug new eterizing a propose therefore We conditions. max efinition of normalized growth rate inhibition (G inhibition rate growth of normalized efinition We can compensate for of the rate effects confounding division 50, 50 was was

t , , GR E in the presence of drug at concentration at concentration drug of presence the in max 1 m 3 max and are similar to the division times of NCI-60 cells NCI-60 of times division the to similar are and eth.385 , , and AUC dramatically, vary will independently values ≥ Fig. 1 Fig. T 0.5, and IC and 0.5, d , , GR = 1.8, 2.4, or 3.9 d). These division times correspond n 3 a a AOC ) because cell number (or CTG value) was nor (or number CTG value) cell ) because 50 ture methods ). AOC and GR , , and 50 , rather than AUC for reasons discussed discussed reasons for AUC than rather , E ) ( was undefined. In the case of the two two the of case the In undefined. was t c max h , 50 GR values in assessing cellular response and − = (Hill slope), values that values (Hill slope), are largely

1 2 | T c k ADVANCE d , ( E = 3.9 d), the total number of number total the d), 3.9 = max k t )/ values fell as division rate as fell division values ) ( 0

ONLINE 2 c These These authors :

Articles PUBLICATION Fig. 1 Fig. R ) a ). | 1

4 - - - .

© 2016 Nature America, Inc. All rights reserved. 4 d and that were exposed to a drug that was partially cytostatic, cytostatic, partially was that drug a to exposed were that and d 4 had that cells for data synthetic created we scripts). to links with Methods Online in provided are calculations for all (equations level at a molecular response drug relate changes in drug-induced cell states to dynamic measures of ( etc. responses, varying kinetics of drug–target interaction, drug efflux, adaptive capture values GR Time-dependent points. time more or two at GR measurements count cell given time-dependent evaluated be can A value Methods). (Online number cell initial of the place in used and experiments parallel in conditions same the under measured be can cells untreated for time doubling the for GI ( number cell initial the given samples, untreated and treated in number of cell surement Methods). (Online concentrations GR and value, GR measured GR( at which concentration the is GR concentrations, for of a drug range values GR Given death. complete and cytostasis, it lies 0 between and −1 in the case of cell it 0 equals in of 1 inhibition, the case growth in of the case partial 0 it and to lies between phenotype: relates value response directly conditions normalized to a single cell division. The sign of the GR simply the ratio between growth rates under treated and untreated ( cells control untreated of rate growth where in values shows line vertical 1 4 ranging from to d; time division with cells for assay 3-day a from theoretical computed ( E IC overlap. curves GR all that note key below); see days; in (given time the the division longer 3 d. the The line,over darker assayed drug for a cytostatic lines) (purple and value GR lines) (green count cell relative showing data 2 × interval a time and after before count cell on based ( ( the at start number cell on either based approach interval ( rates growth on based ( value: GR evaluating ( units. arbitrary a.u., IC where show marks Black inhibition. growth 50% denote and red lines samples control untreated to correspond lines Black (right). line cell and fast-growing (middle), medium- (left), for a drug slow- range for a cytostatic a concentration across counts cell relative yields ( ( time division on metrics of dependence drug-response Figure Online in described are (models cytotoxic or cytostasic, fully f x a max

Articles ) IC ctrl 50 50 ) Simulation of a simple drug-response model drug-response of a simple ) Simulation | To compare GR dose–response curves to conventional curves, curves, to conventional curves To dose–response GR compare mea endpoint from estimated be can values GR practice, In ADVANCE and GR and AOC and GR and 50 Fig. 1 Fig. or 50 1 k is calculated by integrating the GR curve over a range of range a over curve GR the integrating by calculated is | E ( determination, see see determination, x S max c Modeling drug response and the response drug Modeling E ( upplementary Fig. Fig. upplementary , c max 50 max t ∆ )) and )) ( d ) ) is the growth rate of cells and drug-treated x are evaluated. n/a, not applicable. not applicable. n/a, are evaluated. are projected onto the onto are projected t

0 ). Introducing time as a variable makes it possible to it possible makes a variable as time Introducing ). (green) and GR ONLINE ( T are projected onto the onto are projected ) ) and end of the experiment d x , values given in days). days). in given , values ( c , T t d d

= 2.4 d and GR (AUC = 2.4 ± ) time-dependent value value ) time-dependent

b b PUBLICATION

– ∆ ) conceptual approach approach ) conceptual k d 0 t )). ( )). and ) Methods for ) Methods Fig. 1 Fig. 50 e 1 k h ) Simulated ) Simulated or GR ( Supplementary Note Supplementary c GR c ). ). )), ( )), is the slope of the sigmoidal fit; fit; sigmoidal the of slope the is c | ; this is related to the procedure procedure the to related is this ; max x

c -axis; -axis; y n ) ) fixed- -axis. -axis. c (purple) a

) = 0.5, GR ) = 0.5, ture methods AOC

Fig. 1 Fig.

T d b max ranging from 1 to to 1 from ranging ). The GR value is is value GR The ). a e b Cell count normalized ). ). Alternatively, is the maximal maximal the is Relative cell count Cell count normalized to t = 0 count 0.5 1.0 to t = 0 count 0.5 1.0 1.5 0 0 0 1 2 3 Concentration (a.u. 2 0 0 0.1 GR( k From growthrates IC (0) (0) is the Slow ( 5 c 0 ) Time (h Time (h 4 4 1.0 = T E d

T max 2 IC d k Treated Untreated ( =3.9) 5 c 50 =0.66 0 ) Cytostatic response 7 8 /k 10.0 =n/ ) ) (0) 1.0

) – 1

E a exp( exp(

max 72 with different concentrations of epidermal growth factor (EGF). (EGF). factor growth of epidermal concentrations different with supplemented medium serum-free in grown cells 10A MCF in unchanged. were division cell per etoposide of effects the because conditions these under 3a Fig. and IC estimated condi the tions these Under nM. 60 to 0 from increased concentration doxycycline the as three-fold time division increase to observed was promoter doxycycline-inducible a of control the under cells cells transformed non of growth the slow to known is overexpression Oncogene RPE in BRAF). of effect independent is action of mechanism whose (and cytostatic cells a a has that etoposide, inhibitor to II them topoisomerase exposed then and cells (RPE) epithelial IC affects BRAF division expressed cell changing how study To G time; GR division cell on dependence same the from suffers it but IC ( preferable are values GR time-dependent and rate, growth with GR state, cycle cell of independently and rapidly Fig. 1a was this ( GR metrics not of true but corresponding duration, assay and time division with correlated For IC Methods). drugs, all three 2 Supplementary Fig. 2 Fig. Supplementary R 0 0.25 IC 0.75 1.00

GR t t In a second experiment, we measured etoposide sensitivity sensitivity etoposide measured we experiment, second a In 50 metrics are robust to experimental variability experimental to robust are metrics –1.0 –0.5 5 × × AOC 1.4 0 0.5 1.0 k E or k 0 ( (0)) c max 0.5 1.0 1.5 2.0 Concentration (a.u. )) , 0.1 ). In contrast, GR In contrast, ). 0 corrects for this ( this for corrects b E GR( 2 0 ; ; and In Methods). Online the case of that drugs cells kill increased from 0.25 to from 0.25 0.6 ( increased max c GR atdrugconcentration GR Medium ( c 0 1 2 3 1.8 ) =2 0 5 1.0 data and is less sensitive to experimental noise experimental to sensitive less is and data 0 Time (h 4 4 From cellnumbers E lo max g IC 2 T ( V600E Time (h 10.0 x 5 d ( =0.43 0 =2.4) c 15 7 8 ) =3.1 ) 3.9 /x ) 0 , Treated Untreated

GR ) 1

max / lo 6 in hTERT-immortalized retinal pigment pigment retinal hTERT-immortalized in g ) ad xrsin f BRAF of expression and , 50 2 f 2 ( x and Online Methods). AUC combines combines AUC Methods). Online and ctrl 50 value for etoposide increased 100-fold 100-fold increased etoposide for value IC50 and GR50 (a.u.) 0 0.25 IC 0.75 1.00 Supplementary Fig. 1c Fig. Supplementary 0.32 /x 72 3.2 5 and GR and 0 10 ) 0 c 1 – 1 x x(c) x 1 0 1 2 3 0 ctrl 0 50 4 3 2 Fast ( and max T d 24 d GR( Time (h) IC Fig. 2 Fig. E values varied only slightly slightly only varied values 5 T ma 0 1 2 3 IC 0 Fig. Fig. 1e Cytostatic response d c,t E 0 , Time-dependent value =1.8) x 5 E =0.28 48 max 0 ) =2 ma =1.4 a x and Treated Untreated values were values strongly lo lo Time (h)

E and GR g g

max max 72 2 2 , –0.5 –1.0 ( ( f x x 0.5 1.0 (0 ( ; ; c,t 0 0.25 IC 0.75 1.00 2 0 ,t Supplementary Supplementary Supplementary Supplementary ). + + 5 × 0 max ∆ ∆

3 2 1 GR t V600E 50 ∆ t t ) ) /x /x t values, we we values, (0 ( 5

c,t Cell count still varies varies still 0 72 ,t , GR

– – relative to endpoint T ∆ ∆ in RPE RPE in t t d x x x x )) )) ma ( ( (0 (0 – c,t c,t 1 ,t ,t 0.2 2 20 x 1 – + – + , 6 ∆ ∆ ∆ ∆ , 1 t t t t 4 ) ) ) ) 1 Concentration (a.u.) - - , © 2016 Nature America, Inc. All rights reserved. to inhibitors of PI3K and mTOR (such as omipalisib), becoming becoming mTOR omipalisib), and as PI3K of (such inhibitors to sensitivity. drug of estimates reliable within one cell division is likely to be a real advantage in values obtaining GR of stabilization the example), for cells, tumor human E IC are than duration assay on dependent less substantially and BT-20; for h ~36 the and 10A by MCF for h (<20 stabilized division first metrics GR whereas h), (~60 divisions three IC Estimated microscope. cell number measured approximately 8 We h every over 3 d using an automated action. of mechanisms diverse for chosen drugs different five to counting) cell automated facilitate to mCherry H2B- (expressing cells BT-20 and 10A MCF exposed we screen, basis. biological a have IC in variation large culture to lead can cell conditions in differences unintended 3b Fig. GR GR than variable more substantially IC range, ml ng 1 at ml ng 0.02 at d >10 from The division time of MCF 10A cells varied replicates. biological of one three from derive shown data the panels, all In h. 90 at evaluated GR the shows line Dashed (right). GR time-dependent corresponding and (left) omipalisib with treated spheroids as grown cells 10A MCF for intervals 18-h ( of action. mechanisms different with drugs five ofone to exposed cells 10A of MCF imaging live-cell from estimated as right), and (middle points time different at metrics response GR the to close concentration a drug for (bottom) values GR ( 30 at set were ICfor values undefined and Large medium. serum-free in of EGF concentrations different at E ICfor Values (left). medium serum-free in of EGF concentration and between Relationship sensitivity. etoposide altering ( of illustration. purposes for ICfor values undefined and Large of DOX. concentrations different at E ICfor promoter. Values BRAF expressing cells to (left) concentration (DOX) between Relationship sensitivity. of etoposide metrics on cells ( time. division the and of assay the length the Figure less sensitive over time over sensitive less d c a max max max ) Evaluation of relative cell count (top) and (top) count cell of relative ) Evaluation ) Effect of altering of altering ) Effect ) Time-dependent GR values estimated over over estimated values GR ) Time-dependent When grown in 3D culture, MCF 10A cells When are grown in MCF 3D 10A to cells known culture, adapt To test the GR approach in a drug sensitivity typical small-scale 50 max and GR and GR and Supplementary Fig. 3c Fig. Supplementary and and V600E . In the case of very slow and uneven growth (by primary primary (by growth uneven and slow very of case the In . 2 values ( T | ). Thus, potentially arbitrary and and arbitrary potentially Thus, ). d under the control of a DOX-regulated of a DOX-regulated control the under GR values are independent of both of both independent are values GR of MCF 10A cells on metrics of metrics on cells 10A of MCF 50 E max max −1 max and and (right) were evaluated at 59 h 59 at evaluated were (right) h 48 at evaluated were (right) µ 50 G ( EGF M for purposes of illustration. of illustration. purposes M for Fig. 2 value (left) and computed computed and (left) value values that is unlikely to to unlikely is that values 50 E and GR and max T T d 50 50 b d of hTERT RPE-1 RPE-1 of hTERT i. 2 Fig. and doxycycline doxycycline and and GR and and were set at 30 30 at set were for etoposide were were etoposide for 1 7 50 . Endpoint measures of drug sensitivity sensitivity drug of measures Endpoint . b 50 Supplementary Supplementary −1 ) Effect of ) Effect (middle) and (middle) c and ). Across this this Across ). 50 ). This confirms that GR metrics are metrics GR that confirms This ). EGF to <2 d <2 to EGF 50 50 (middle) and (middle) value value values values E max

µ M 50 values converged values only after 50

or or d

a c b GR Relative cell count 1.0 1.0 0. 0. 0. 0. 0. 0. 0 0 4 8 2 4 6 8 12 12 Tanespimycin -HSP9

1 1 T Td d 2. 3. 0. 1. 1. 2. 10 12 14 24 24 0 2 4 6 8 0 5 0 5 0 5 0 0.02 MCF 10Acells Time ofassay(h) (0.1 2 2 EGF (ng/mL) Fig. 2 Fig. No. ofdivisions d 1 0 RPE-1 cells 36 36 DOX (nM) 50 0.11 µ and 3 6 M) 0.32 3 3 48 48 6 1

Omipalisib Time-dependent GR c 1.00 0.2 0.4 0.6 0.8 1.0 1.2

2 60 60 0 0 10 4 4 of anticancer drugs for 3 days. For many drugs in this dataset, IC bers for breast cancer cell lines before and after exposure to a panel sion coefficient of −0.54, Spearman’s −0.54, of coefficient sion correlates with division rate An exception is a study by Heiser them. estimate to one allow nor rates division cell report neither The majority of large-scale drug-response datasets published G using data to of datehigh-throughput Analysis mechanisms. adaptive of analysis further ness of in omipalisib MCF 10A and spheroids, enabling detection time-dependent GR data directly capture the decreasing effective GR endpoint ( 4 day by ~10-fold increased they and addition, palisib time-dependent GR and GR the that observed we imaging, live-cell we by spheroids when monitored However, responses. adaptive such on report not do 72 72 20 Etoposide -topiosomerase Omipalisib GRovertime IC and GR (µM) IC and GR ( M) 50 50 ( 50 50 µ 10.0 µ 0.1 1.0 10 30 M) 84 84 1 Time (h) 40 0.001 (3.2 Fold change from 72 h Fold change from 72 h 0.02 0 >30.0 >30.0 10.0 10.0 50 1.0 3.2 1.0 3.2 µ Etoposide sensitivity Etoposide sensitivity M) 0.01 60 12 12 EGF (ng/mL) 0.11 GR IC GR IC 16 50 DOX (nM) 1 1 50 50 70 50 50 0. values lay midway in this range ( range this in midway lay values 24 24 n 1 1 a 80 Time ofassay(h) 3 2 3 2 ture methods 0.32 No. ofdivisions 31 36 36 Omipalisib -panPI3K 48 Time-dependent GR (nM) 48 Change inIC 50 Change inGR 1.00 10 32 (0.2 62 3 10 60 60

µ 6 M) 4 4 Omipalisib GR 50 20 (e.g., for cell cycle inhibitors, regres 72 72 values were lowest ~20 h after omi E E 30 max max

50 90 hendpoin 15 hinterval 0.25 0.50 0.75 0.25 0.50 0.75 50 | et al. 84 84 ADVANCE Time (h) 40 Linsitinib -IGF1R 50 0.02 50 GR difference from 72 h E difference from 72 h 3 1 0 P max max –0.2 overtime (10 , , which recorded cell num 0.2 0.4 0.6 0.2 0.4 0.6 -value < 10 < -value 60 Etoposide sensitivity Etoposide sensitivity 0 0 t 12 12 µ 70

EGF (ng/mL) M) 0.11 GR E ONLINE 1 1 DOX (nM) ma 3 6 24 24 ma 80 x R x metrics Time ofassay(h) 2 2 No. ofdivisions 0.32 36 36

Articles 1 PLX4720 -BRAF PUBLICATION Fig. 2 Fig. −66 Change in Change inGR 3 3 48 48 (10 GR E 1.0 , , ma 62 N µ ma 60 60 x 0 M) d Fig. 2 Fig. = 2,956; 2,956; = x 4 4

). Thus, Thus, ). E –0. 0 0. –0. 0 0. 72 72 ma 5 5 5 5 ma x x | d 84 84

50 GR GR

); ); max max - - - -

© 2016 Nature America, Inc. All rights reserved. are more taxane-sensitive than tumors in other types of breast cancer breast of types other in tumors than taxane-sensitive more are HER2 that showing ( responses positive (HR receptor- hormone whereas response, cytotoxic a of indicative GR lines (TNBC) cancer breast (HER2 In and HER2-amplified subtypes. GR microtubules assembled for paclitaxel ( and nM 10 around centered range narrow a span actually alysis of data in Heiser mitosis in be to is it likely more the divides, because line a cell faster and the cells, on mitotic arise primarily acts paclitaxel to thought is and described been previously has tion (100-fold) in IC varia substantial as well as rate division and sensitivity between correlation negative strong a exhibits that drug one is otherapy, IC way the of artifact an is it that suggesting GR using measured was the response drug also, when absent was correlation spurious; is data experimental in found rate division suggests finding that the correlation and drug sensitivity between ( parameters random using responses drug simulating repeatedly by and rate division and sensitivity assume drug between not connection does biological any that model idealized an using by correlation 4a Fig. Supplementary subtypes. other all IC subtype. clinical by Heiser in drugs all for ( (NM). nonmalignant (HR positive receptor hormone (TNBC), cancer breast negative (HER2 HER2-amplified subtype: by grouped dataset the in ( coefficients. correlation 3 d; over line that for of divisions number and IC estimated between relation the show plots) of dose–response the right the to and (below distributions Marginal divisions. fewer with lines cell denote curves darker response; cytostatic Heiser from data assay (3-day lines cell cancer breast in paclitaxel for values GR ( Figure a MDAMB231 MDAMB231 a

Articles 50 , | Paclitaxel, Paclitaxel, a taxane microtubule inhibitor widely used in chem b ADVANCE MCF10A MCF10A or GR or max 50 ) Fitted dose–response curves for ( for curves dose–response ) Fitted HS578T HS578T SKBR3 SKBR3 Supplementary Fig. 4c Fig. Supplementary MCF7 MCF7 (Spearman’s (Spearman’s BT20 BT20 3 values for paclitaxel vary considerably across cell lines lines cell across considerably vary paclitaxel for values 2 | 1 50 . We propose that future attempts to find biomarkers biomarkers find to attempts future that propose We . Evaluation of GR metrics in a high-throughput dataset. dataset. a high-throughput in metrics of GR Evaluation values for NM cell ( cell NM for values et al. et Spearman's correlationwithseedingnumber Supplementary Fig. 4c Fig. Supplementary –0.4 –0.4

ONLINE + *** Supplementary Fig. 4b Fig. Supplementary *** ) and nonmalignant lines generally exhibit cytostatic Correlation Correlation * 3 ). Red denotes cytotoxic response and blue denotes denotes blue and response cytotoxic denotes Red ). IC GR P 0 0 50 50 -values were derived from a rank-sum test for test a rank-sum from derived were -values

d et al. et PUBLICATION , * P 50 * *** amp e *** -value = 0.31, 0.31, = -value ) Distribution of IC ) Distribution 0.4 0.4 across cell lines ( c et et al. ) Number of divisions over 3 d for cell lines lines 3 cell d for over of divisions ) Number 3 and TNBC (basal-like) human tumors tumors human (basal-like) TNBC and ( ). However, we could reproduce this this reproduce could we However, ). d MDAMB231 MDAMB231 ) or for ErbB2 inhibitors only ( only inhibitors ErbB2 for ) or 3 d shows that GR ) or for HER2 for ) or ), close to the estimated affinity of of affinity estimated the to close ), MCF10A MCF10A HS578T HS578T SKBR3 SKBR3 MCF7 MCF7 BT20 BT20 | max

n a 50 ) relative cell count and ( and count cell ) relative a ). This is consistent with data data with consistent is This ). values are negative, which is is which negative, are values ture methods or GR or and Online Methods). This This Methods). Online and *** *** *** –0.4 –0.4 upeetr Dt 1 Data Supplementary 50 Fig. Fig. 3 (green) and GR and (green) amp Correlation Correlation ρ amp * n vitro in 50 GR - E values are Spearman’s Spearman’s are values cell lines ( lines cell and and max 50 0 0 max 18 50 ) and triple-negative ) and triple-negative values for paclitaxel a , ). ). This relationship is calculated. is ** 1 9 E 0.4 0.4 . However,. rean max *** 2 0 amp or GR or . In contrast, contrast, In . e ), triple- ), b 50 ) versus ) versus e + ), and ), ) grouped ) grouped (purple) (purple)

No. of divisions in 3 d max b Fig. 3 Fig. 4 1 2 3

)

values values 156

), ), b BT20 MDAMB231 MCF10A - - - 313

Seeding number rdcie f altxl epne ou o vrain n GR in variation on focus response paclitaxel of predictive than other subtypes to EGFR/ErbB2 inhibitors inhibitors EGFR/ErbB2 to subtypes other than show of that subtype this breast is more cancer ~10-fold sensitive disease this for therapy frontline are inhibitors ErbB2 and drugs such to tive HER2 though even subtypes, cancer breast across similar were ErbB2 and EGFR of ( similar was sensitivity drug of range and mean the that showed GR whereas cells, tumor than drugs anticancer to tive by IC (median faster still divided cells cells TNBC faster divided (median HR among subtype: with differs clinical cells studied by Heiser IC in variation than rather genotype and drug sensitivity. drug and genotype between associations meaningful obscures also and exist none dependency of IC in IC confounder hidden a is rate growth slow relatively their because of HER2 sensitivity P 625 Fig. 3 Fig. c = 1.3 × 10 a The average division rate of breast cancer cell lines in culture culture in lines cell cancer breast of rate division average The HER2 No. of divisions in 3 d 0 1 2 3

+ No. of Relative cell count 0.2 0.4 0.6 0.8 1.0 and HER2 1,250 amp

50 divisions 50 1 3 0 Paclitaxel relativecellcount Division rateby 0.001 d calculation. From these data we conclude that artefactual artefactual that we data conclude From these calculation. TNBC values, nonmalignant cells were, on average, more sensi more on average, were, cells nonmalignant values, ). Focusing on HER2 on Focusing ). subtype 2,500 HR 0.01 SKBR3 MCF7 HS578T 24 −4 + IC , 0.1 ). ). The failure of IC 50 2 5,000 NM amp 5

( . . GR µ M) 1 50 � subtypes divided most slowly (median =–0.66 d on cell division rate creates associations where amp amp 10 50 IC50 and GR50 (µM) >10 0.01 data for HER2 data � 0.1 cell lines to EGFR/ErbB2 inhibitors arises arises inhibitors to lines cell EGFR/ErbB2 100 nine biological replicates. biological nine to of s.e.m. the five are bars 3 d. in Error of divisions number the and number seeding *** D and Methods (Online action of mechanisms diverse having drugs 11 with treated and densities a range at of six plated were cells which in experiment an from derive Data shown. lines cell cancer breast six the E ICfor values estimated between correlation ( sensitivity. drug and rate Figure =–0.63 10 human tumors are preferentially sensi preferentially are tumors human max 0 HER2 1 divisions ata ata 3 No. of

P amp Potency foralldrugs , , GR 50 < 0.001. ( < 0.001. 1 2 TNBC 22 ). Significance: * Significance: ). 4 amp P 50 | ,

=3.4×10 E 2 max , and GR , and Plating density affects division division affects density Plating 3 50 HR T . lines, IC lines,

+ P d values to values show the preferential =0.69 b = 1.8 d, T GR

No. of –0.8 –0.4 d b 0.4 0.8 amp = 2.3 d), and nonmalignant

NM 11 ) Relationship between between ) Relationship divisions 1 3 0 0.001 max cell lines in Heiser Heiser in lines cell IC and seeding number for number seeding and 50 e 0.01 50 Fig. Fig. 3 Paclitaxel GR P IC and GR (µM) values for inhibitors inhibitors for values 50 50 GR < 0.05, ** < 0.05, >100 0.01 a 0. 0. 10 50 ) Spearman’s ) Spearman’s 1 1 HER2 1 1 S

( in vitro in upplementary upplementary µ c ErbB2 inhibitorpotency GR amp M) ). ). As measured � =–0.13 50 P P =2.19×10 =0.17 TNBC 10 P T 100 < 0.01, < 0.01, � ( d 50 divisions =0.41 3 = 3.2 d), No. of Fig. 3 Fig. HR

values values +

4 et et al.

et et al.

50 NM max

, GR e

3 max - - , ;

© 2016 Nature America, Inc. All rights reserved. ( IC on quently conse and divisions, densities. of number all on density of across effect well the Thus, equally grew which cells, MCF7 for IC and density between Correlations medium growth of the of conditioning because presumably density, with increased rate division cells, ( SK-BR-3 and BT-20 effects other and components, medium essen of tial depletion inhibition, density, contact with of because decreased rate presumably division lines, cell three first the ( correlated cells SK-BR-3 they and BT-20 and for cells, negatively 578T Hs and MDA-MB-231, 10A, and addition drug ( of treatment h 72 and after time the at cells fixed counting and imaging IC action. rates, of Growth mechanisms diverse with drugs 11 to cells exposed plating, were after h 24 range. six 32-fold at a over subtypes densities different seeding of representative lines cell breast six resistance drug to promote reported been widely has density increasing density: seeding is variable such One rate. division affect that variables of face the in changes sensitivity drug how quantify to us allow metrics GR E GR 0.03 to exposure following densities cell and points time different at caspase-3 cleaved for positive cells 10A of MCF Fraction replicates. biological of three s.e.m. the are bars Error 3 d for (left). paclitaxel with treated and densities different at seeded cells 10A MCF for density. values GR cell high at apoptosis and density cell low at cytostasis ( replicates. biological of one three from are shown Data inhibitor. 10 of (right) presence or (middle) absence the in 20 to exposure following numbers seeding different at 10A MCF for values GR Time-dependent replicates. biological of three s.e.m. the are bars Error (left). densities cell different at linsitinib with treated 10A MCF 3 d for at curves dose–response GR inhibitor. IGF1R an linsitinib, to resistance adaptive and ( replicates. biological of one two from derive shown Data (right). of medium volumes different in but cells seeded of number a constant with methotrexate with treated 10A 3 MCF d for at curves response 30 at capped are values (middle); oligomycin or GR 0.1 to exposure following numbers seeding different at cells 10A MCF for values GR dependent Time- metabolism. to related drugs two for volume medium to number of of cell ratio in summarized data on based cells 10A MCF in values GR on effect a significant had density seeding which for drugs selected of three Evaluation resistance. and sensitivity of drug mechanisms diverse 5 Figure line–drug pairs exhibited a statistically significant association association significant statistically a exhibited pairs line–drug Supplementary Fig. 5a Fig. Supplementary ffects of cell density on drug sensitivity drug on density of cell ffects max 50 µ values for MCF 10A cells for methotrexate methotrexate for cells 10A MCF for values M batimastat, a matrix metalloproteinase metalloproteinase a matrix M batimastat, E values for paclitaxel in MCF 10A cells seeded at different densities (right). Data shown are from one of two biological replicates. biological of one two from are shown Data (right). densities different at seeded cells 10A MCF in paclitaxel for values µ µ max M for illustration purposes. GR dose– GR purposes. illustration M for M methotrexate (left). Time-dependent Time-dependent (left). M methotrexate | Time-dependent GR metrics reveal reveal metrics GR Time-dependent correlated positively with seeding number for MCF MCF for number seeding with positively correlated 50 and and Figure Figure 50 c b , , ) Paclitaxel-induced ) Paclitaxel-induced ) Medium conditioning conditioning ) Medium E E max max 4 . . ( ). However, only a small number of cell of cell number a small only However, ). Supplementary Data 2 Data Supplementary , varied dramatically among cell lines lines cell among dramatically varied , , and GR metrics were estimated by by estimated were metrics GR and , a 26– ) Influence ) Influence µ 2 M linsitinib M linsitinib 9 . To investigate this, we cultured we cultured . this, To investigate 50 and and P < 0.01, 0.01, < E c b a max

Fig. 4 Fig. GR –0.4 –0.2 0.2 0.4 0.6 0.8 1.0 1.2 ). Overall, IC Overall, ). 0.0001 GR Time-dependent GR were weakest weakest were 0 0.2 0.4 0.6 0.8 1.0 1.2 0.2 0.4 0.6 0.8 1.0 1.2 0 0 Fig. 4 Fig. Paclitaxel doseresponse b ) 0.1 Linsitinib doseresponse 0.001 Methotrexate, 0.10 30– 20 Paclitaxel 0.32 a Linsitinib 3 5,00 2,50 1,25 625 313 156 31 2 ). In In ). . In In . Time (h) , 0 0 0 3 0.01 50 40 3 1 Seeding number - - . ( 156

µ ( M) 3.2 µ

M) 625 clinical trials ( trials clinical II Phase in currently inhibitor IGF1R an linsitinib, for observed GR in variation dependent time-) (and Density- ent. GR of medium, volume a constant synthase; ATP of inhibitor an oligomycin, of true also was (this drug this to sensitivity for variable key the was se, per density cell not and volume, and medium number cell between ratio the that showed ( state transcriptional of of determinant primary the history was culture the the not and assay of time the at density cell the that ρ lated with the number of cells at the time of collection (Spearman’s component (which captured 32% of was variance) strongly corre the principal first analysis, was by analyzed principal-component responses ing wound and respiration, cellular genes processes, catabolic in for involved enrichment significant with density, cell also with cells varied these in expression gene that revealed RNA-seq ties; GR in variation of biology density. the by altered was which in response drug situations were these whether dered GR and density plating between was reduced by cotreatment with the metalloprotease inhibitor inhibitor metalloprotease the with cotreatment by reduced was 0.1

Supplementary Fig. 6b Fig. Supplementary 2,500 60 µ = 0.98) and less so with time in culture ( culture in time with so less and 0.98) = 313 156 5,000 2,500 1,250 625 10 In MCF 10A cells, six drugs were associated with significant significant with associated were drugs six cells, 10A MCF In M 32 Fraction of cCASP3- positive cells Time-dependent GR Time-dependent GR50 (µM) Fig. 5 Fig. 0.001 0.2 0.4 0.6 0.8 1.0 0.6 0.1 0.2 0.3 0.4 0.5 0.01 0.1 0 0 10 10 1 156 20 30– Fig. 5 Fig. n µ a 20 M linsitinib,nobatimastat Metabolic inhibitorGR 20 a 32 and and 50 313 Paclitaxel, 0.03 Seeding number ture methods Seeding density 156 , r GR or 3 Oligomyci Methotrexate 6 3 24 12 4 b 30

625 Time (h) ( 40 h h over tim Supplementary Fig. 7 Fig. Supplementary ). Variation in GR Variationin ). 625 Time (h) h h Supplementary Fig. 6a Fig. Supplementary 1,250 2,500 40 ). Follow-up studies with methotrexate methotrexate with studies Follow-up ). 60 n max µ 2,500 e M paclitaxel (middle). Time-dependent Time-dependent (middle). M paclitaxel µ 50 M o bt) cos edn densi seeding across both) (or

5,000 80

50 | ADVANCE 60 50 50 or GR or was therefore time depend time therefore was Time-dependent GR max Time-dependent GR GR –0.8 –0.6 –0.4 –0.2 0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8 1.0 0.2 0.4 50 0 0.001 Methotrexate doseresponse 20 0 10 0 10

with time and density density and time with ONLINE µ max Paclitaxel GR ρ M linsitinib,10 ); since cells grew in in grew cells since ); = 0.57), suggesting suggesting 0.57), = 20 0.01 Methotrexate 15 values. We won We values. ). When the data data the When ). Seeding number

Articles 30 Time (h) PUBLICATION Time (h) 156 20 0.1 Seeding number ma

50 156 40 625 µ x were also also were M batimastat overtim 25 (

625 µ 0 1 2,500 M) 2,500 50 120 100 80 60 40 20 30 e µ µ µ µ 60 | µ µ L L L L

L L - - - - - © 2016 Nature America, Inc. All rights reserved. GR grcalculator.org ( calculator GR online an and routines python and To facilitate the use of GR by metrics others we provide MATLAB ( interaction drug–target of kinetics the in variation or adaptation, drug response, delayed as such phenomena quantify to and values GR time-dependent compute to possible it make data Time-course quality). data ensure to control valuable a are number cell on data after and before that believe we (although needed is are number cell final the conditions only data, previous from culture known similar under rates division cell When ( time fixed a for perturbation other or drug of trations CTG concen (e.g., varying to culture a of surrogate exposure after and a before value) or number cell measuring after puted neglected often is that response and dose between relationship an important quantifies latter the and noise, experimental of face the in metric robust most the often h GR inhibition. partial denote values positive and stasis, 1 and −1, where negative values denote cell death, 0 denotes cyto on effect growth rate and from differs to GR equivalently. scored responses are drug biochemical similar with cells slow-growing and fast- that ensuring basis, per-division a on drug a of potency the GR environment). extracellular the in variation and overexpression, or depletion gene drugs, of study the (including IC replace and GR comparing growth rates in the presence and absence by of computed drug. are GR that metrics GR propose we alternative, an As and introduce unknown complications into biomarker discovery. action, drug of effects true the obscure data, in correlations cial IC apparent up can it change Such in while speeds variation others. increases, density as lines cell some in down slows rate division Cell ways. type, cell with medium composition, varies and seeding density, rate often drug-response in unpredictable division that existing show also We confounds metrics. seriously rate division in variation that experimentally and theoretically demonstrate we In paper this division. and cell on signaling cell studies damental fun many of and pharmacology, biology, cancer of cornerstone the is resistance and sensitivity drug of measurement Accurate DISCUSSION in dynamics tumors cell-killing xenograft high-density in and culture between low-density discrepancy the for reason one be may effect this unknown, is basis molecular its Though ( apoptosis elevated to corresponded GR negative that confirming caspase-3, cleaved con that tain cells taxol-treated of fraction the in increase an with GR ues at (cytotoxic) ( densities higher val negative to densities cell at low (cytostatic) ~0 from varying across plating densities, but GR response. drug in the microenvironment of conditioning autocrine for role a suggesting batimastat, ( online

GR Articles | In the simplest version of our method, GR metrics are com are metrics GR method, our of version simplest the In GR paclitaxel, of case the In ADVANCE 50 max values can also be calculated from GR curves; the former is is former the curves; GR from calculated be also can values and GR and reached its greatest negative value at 24 h, concomitant concomitant h, 24 at value negative greatest its reached max http://lincs 50 are rate robust in and to division cell should variation

50 ONLINE and and max 100-fold or more and therefore introduce artifi introduce therefore and more or 100-fold / ). ). Moreover, the data in paper,this the including values for the Heiser Heiser the for values E

.hms.harvard.edu PUBLICATION max in studies in which control cells divide divide cells control which in studies in max 50 |

values remained at ~5–10 nM nM ~5–10 at remained values was strongly density dependent, max n Fig. Fig. 5 a ture methods 6 captures the maximal drug drug maximal the captures / . Fig. 5 Fig. E ). et al. et max c , left). Time-dependent , Time-dependent left). in that it between falls dataset, are available available are dataset, c , middle and right). right). and middle , 50 http:// quantifies quantifies max Fig. 1 Fig. Fig. 1 Fig. AOC values values 3 www. 5 . and and d c 50 ). ). ). ). ------

suggested suggested that cancer therapy might bybe personalized screening been it recently has infections, for bacterial testing susceptibility antimicrobial with analogy By metrics. GR of use the from efit ben also and should in levels), EGF and changes overexpression oncogene by here illustrated (as sensitivity drug with relations cor spurious potentially to leading rate, division cell in changes in result often microenvironment the or genes of modification the involving studies biology Cell sensitivity. on drug effect their from rate division on have microenvironment or genotype that for drug sensitivity and resistance. GR metrics decouple any effect responsible processes biological and genes identify to ability our IC traditional of lieu in rics rates. division measuring then and conditions assay original the recreating require will this but facto, by post values data GR computing existing in this for correct to possible be might It rate. density, growth medium, and other factors that affect plating cell division in differences by confounded be might center) a within (or even centers across of datasets Wecomparison that speculate biomarkers of drug-response value the about concerns raising metrics 3. 2. 1. c R The authors declare no competing financial interests. COM the computational analyses. and M.H. performed the experiments; M.H. conceived GR metrics and performed M.H., M.N., and P.K.S. conceived this study and wrote the paper. M.N., M.C., A the manuscript. modified RPE-1 cells and A. Palmer, M. Eisenstein, and G. Berriz for help with Systems Biology, Harvard Medical School, Boston, Massachusetts, USA) for the to M.H. We thank M. Soumillon for expression profiling, J. Chen (Department of by a fellowship from the Swiss National Science Foundation (P300P3_147876) This work was funded by grants U54-HL127365 and P50-GM107618 to P.K.S. and Ackno online version of the pape Note: Any Information Supplementary and Source Data files are available in the number accession with database (GEO) Omnibus codes. Accession o version and are Methods references any in available the associated M therapy. patient optimizing for useful and reproducible differences such data for that using should create GR are drug-response metrics more Accounting variable. controlled poorly a number division making culture, in unevenly and slowly grow primary human tumor cells against panels of drugs om/reprints/index.ht eprints and permissions information is available online at UTHOR

ethods Based Based on the results in this paper, we believe that use of GR met response existing on based studies drug-response Large-scale P (2012). cancer.breast in compounds cells. cancer in sensitivity (2012). sensitivity.drug anticancer of modellingpredictive Heiser,L.M. M.J. Garnett, J. Barretina, ETIN w

led CONTRI 1 G G F , 2 f the pape the f , g 8 IN , ments 9 are discrepant for poorly understood reasons understood poorly for discrepant are A B et al. et NCI et al. et UTIONS et al. et A Data are deposited in the Gene Expression Expression Gene the in deposited are Data Subtype and pathway specific responses to anticancer to responses specific pathway and Subtype L m The Cancer Cell Line Encyclopedia enables Encyclopedia Line Cell Cancer The

r Systematic identification of genomic markers of drug of markers genomic identification of Systematic INTERESTS . r l . . Nature Proc. Natl. Acad. Sci. USA Sci. Acad. Natl. Proc. 50 , , E

4 max 8 3 , 570–575 (2012). 570–575 , , or AUC values will improve improve will AUC or values , Nature

http://www.nature. 1 09 36 GSE8029

4 , , 2724–2729 , 3 8 7 3 . . Such cells

, 603–607 , online online 5 , 11 7

, 1 1 0 2 - - - . ,

3269–3276 (2008). 3269–3276 kinesin-5. and microtubules target that drugs antimitotic chemotherapy.preoperative disassembly. cancer.ovarian with patients from cultured cells tumour of paclitaxel to sensitivity 1996). (McGraw-Hill, 51 Ch. L.) Limbird, Therapeutics of Basis Pharmacological cells. cancer matrix-attached naevi. human of arrest p16INK4a. and p53 of accumulation with associated senescence cell prematureprovokes ras agents.anticancer 123 of potency growth-inhibitory the with correlations and screen drug anticancer Institute Cancer National the of lines cell in pathway sets. data line cell cancer two between agreementPharmacogenomic Consortium. Cancer in Sensitivity at Preprintstudies. Nature research. biology cancer dataset. sensitivity small-molecule action. of mechanism reveals 1 drugs. cancer to responses in variation systematic reveal potency than other Metrics series. minireview chemistry discovery. drug and biology chemical in action of mechanism identification and Shi, J., Orth, J.D. & Mitchison, T. Cell type variation in responses to responses in variation type Cell T. Mitchison, & J.D. Orth, J., Shi, Rouzier,R. microtubule modulates taxol How R. Ruhlen, & J. Shanks, M., Caplow, B.C.Baguley, P.in Calabresi, & G.A. Curt, C.J., Allegra, Chabner,B.A., T.Muranen, C. Michaloglou, S.W.OncogenicLowe, & D. Beach, M.E., A.W.,McCurrach, Lin, M., Serrano, O’Connor,P.M. Drug of Genomics the & Consortium Encyclopedia Line Cell Cancer The Z. Safikhani, B.Haibe-Kains, T.M.Errington, Seashore-Ludlow,B. M.G. Rees, estimation. EC50/IC50 accurate for Guidelines J.L. Sebaugh, Sorger,P.K.Gray,J.W.& Heiser,L.M., S., Honarnejad, M., Fallahi-Sichani, biological meets biology Chemical J.M. Gottesfeld, B.F.& Cravatt, Danc Schenone,M., 0 , 128–134 (2011). 128–134 ,

50 4 Nat. Chem. Biol. Chem. Nat. et al. et , 389–393 (2013). 389–393 , et al. et et al. et J. Biol. Chem. Biol. J. et al. et et al. et Nat. Chem. Biol. Chem. Nat. et al. et et al. et Eur. J. Cancer J. Eur. et al. et et al. et Correlating chemical sensitivity and basal gene expressiongene basal and sensitivitychemical Correlating Breast cancer molecular subtypes respond differently to differently respond subtypes molecular cancer Breast Inhibition of PI3K/mTOR leads to adaptive resistance in resistance adaptive to leads PI3K/mTOR of Inhibition Revisiting inconsistency in large pharmacogenomiclarge in inconsistency Revisiting ˇ ík, V., Wagner, B.K. & Clemons, P.A.TargetClemons, & V.,Wagner,B.K. ík, Cancer Res. Cancer Resistance mechanisms determining the in vitro in the determining mechanisms Resistance http://dx.doi.o Cell Characterization of the p53 tumor suppressor tumor p53 the of Characterization et al. et An open investigation of the reproducibility ofreproducibility the of investigation open An Inconsistency in large pharmacogenomic studies.pharmacogenomic large in Inconsistency BRAFE600-associated senescence-like cell cycle cell senescence-like BRAFE600-associated

Nature 88 eLife Harnessing connectivity in a large-scale a in connectivity Harnessing

, 593–602 (1997). 593–602 , 9 26 Clin. Cancer Res. Cancer Clin. J. Biol. Chem. Biol. J. , 232–240 (2013). 232–240 ,

Cancer Cell Cancer Nat. Chem. Biol. Chem. Nat.

3 9 436 5 9 Nature , e04333 (2014). e04333 , , 23399–23402 (1994). 23399–23402 , 7 , 708–714 (2013). 708–714 , A , 4285–4300 (1997). 4285–4300 , , 230–237 (1995). 230–237 , , 720–724 (2005). 720–724 , rg/10.1101/02615 Cancer Discov. Cancer 9th edn. (eds. Hardman, J. & J. Hardman,(eds. edn. 9th

5 2

21 8 , 84–87 (2015). 84–87 ,

2 , 227–239 (2012). 227–239 ,

85 11

12 , 11031–11032 (2010). 11031–11032 , , 5678–5685 (2005). 5678–5685 ,

, 109–116 (2016). 109–116 , 5 , 1210–1223 (2015). 1210–1223 , 3 (2015). The Cancer Res. Cancer Pharm. Stat. Pharm.

6 8 ,

28. 27. 31. 30. 29. 26. 25. 24. 23. 37. 36. 35. 34. 33. 32.

11153 (2009). 11153 growth. epithelial of dynamics spatial the governs contact cell-cell and factor growth epidermal between (1996). 10963–10966 cultures. density cell high in phosphorylationtyrosine receptor factor growth platelet-derived and factor growth epidermal of (2007). activity.hypoxia-induciblefactor-1 requires cells carcinoma breast MDA-MB-231 human in doxorubicin (1995). 873–879 line. cell colon-cancermulti-drug-resistant human agents.anticancer agents. (2001). 7184–7188 cells. tumorHER2-overexpressing of growth the ( ZD1839 inhibitor cells. cancer breast trastuzumab-treated andHER-2-overexpressing against (GW572016) Cell Cancer drugs. antimitotic to exposureprolonged following variation interline (2014). cancer.for combinations drug effective identify (2012). papillomatosis. respiratory for Res. Cancer pharmacodynamics.single-cell and microscopy vivo in by tumors in 11 NF- of translocation system. autocrine an serum: of absence the in grow to cells enablesproduction 177 Nf2/Merlin. by signaling EGFR of Contact-dependentinhibition Garrido, C. Garrido, to resistance Kinetic B. Chauffert, & C. Garrido,M.T.,Dimanche-Boitrel, Kim, J.H., Kushiro, K., Graham, N.A. & Asthagiri, A.R. Tunable interplay Tunable A.R. Asthagiri, & N.A. Graham, K., Kushiro, J.H., Kim, decreasephosphatase-mediated Protein-tyrosine A. Ostman, & M. Sorby, to Confluence-dependent resistance C.H. Graham, & R. Sullivan, Y.,Fang, B. Chauffert, kinase tyrosine The N. Rosen, & S.D. Averbuch, A., Basso, Moasser,M.M., Konecny,G.E. and intra-profound display cells Cancer Taylor,S.S. & K.E. Gascoigne, Crystal, A.S. Crystal, H. Yuan, J.D. Orth, J.E. Sero, factor(s) Transforminggrowth B.Ozanne, & M. P.L.,Anderson, Kaplan, McClatchey,A.I. & C.H. Liu, D., Lallemand,B.K., Cole, M., Curto, , 790 (2015). 790 , , 893–903 (2007). 893–903 , Cytotechnology Proc. Natl. Acad. Sci. USA Sci. Acad. Natl. Proc. et al. et et al. et

et al. et

14 et al. et 71 et al. et Cancer Res. Cancer et al. et , 111–122 (2008). 111–122 , et al. et , 4608–4616 (2011). 4608–4616 , n Use of reprogrammed cells to identify therapyidentify to cells reprogrammedof Use Cell shape and the microenvironment regulate nuclear microenvironmentregulate the and shape Cell Analysis of mitosis and antimitotic drug responses drug antimitotic and mitosis of Analysis a Circumvention of confluence-dependent resistance in a in confluence-dependent resistance of Circumvention ture methods Patient-derived models of acquired resistance can resistance acquired of Patient-derivedmodels ″ New insights into the kinetic resistance to anticancer to resistance kinetic the into insights New Cytotechnology Iressa Activity of the dual kinase inhibitor lapatinib inhibitor kinase dual the of Activity κ B in breast epithelial and tumor cells. tumor and epithelial breast in B

66 ″ , 225–235 (1998). 225–235 , ) inhibits HER2-driven signaling and suppresses and signaling HER2-driven inhibits ) , 1630–1639 (2006). 1630–1639 , N. Engl. J. Med. J. Engl. N. Proc. Natl. Acad. Sci. USA Sci. Acad. Natl. Proc.

12 7 | Exp. Cell Res. Cell Exp. 9 ADVANCE , 347–356 (1993). 347–356 , , 485–489 (1982). 485–489 ,

367 ONLINE Cancer Res. Cancer Science

313 Int. J. Cancer J. Int. , 1220–1227 , , 867–877 , J. Biol. Chem. Biol. J.

Articles PUBLICATION 346

1 0

6 61 Mol. Syst. Biol. Syst. Mol. , 1480–1486 , , 11149– ,

, J. Cell Biol. Cell J.

271 |

© 2016 Nature America, Inc. All rights reserved. where the fitted parameters are: parameters fitted the where as follows ( curve to fitted a sigmoidal Curve fitting and estimating drug-response metrics. time. division cell a 2 puted with × equation: the on × (2 interval time a over ated GR time-dependent values. Calculating where ( number cell initial the of place in used and experiments independent in measured in exposure. drug to prior just grown measured and sample parallel a from count cell the of mean 50%-trimmed where formula: data. to the according is calculated rate inhibition drug-response growth Normalized endpoint using values GR Calculating three technical replicates (on three separate plates). to yield an average relative cell count. We typically collect data from mean of the count for control cells. Technical replicates theare presenceaveraged of drug at concentration we define the relative cell count as cline or batimastat. For each cell line, drug, and drug concentration, 10A cells), and the concentration of a second drug such as doxycy centration of exogenous growth factors (e.g., EGF in currentthe study,case ofrelevant MCFconditions include seeding density, the con controls grown on the same plate under the same DMSO-treatedconditions.to normalized areIn drug the of presence the in counts response. drug of Metrics ONLINE n a • GR values The in time-dependent the current paper were com Alternatively, the untreated division time ture methods

partial growth inhibition ( ( Fig.8a to cytotoxic responses (i.e., induction( of cell death, GR c SupplementaryFig.8b

→ T inf x GR

is the duration of the assay. the of duration the is METHODS (

 : the effect of the drug at infinite concentration (GR c ) and and ) )). GR ), a valuea), correspondsof 0 fullytoacytostatic response ) ( t c , GR GR ∆ = t inf = 12 to h about half to 18 corresponds h, which ) ( GR x ( 2 c lies between −1 and 1; negative values correspond c ctrl lo ) x lo + = 0 ( ( g = ( g are as described above, and and above, described as are c = = 2 2 GR ) 2 ( ( t x = c x x lo lo ctrl 0 0 in / g 2 ( g

) , ) , 0 2 1 x t f Determining relative cell counts. cellrelative Determining 0 2 ), and a positiveanda ),valuecorresponds to × 2 × − + − + ) ( Supplementary Fig. 8c ) ( + x x ∆ ∆ c x ∆ ∆ ∆ ctrl lo t x t 1 t) around any time point t based based t point time any around t) x − )/ / g ( / ( / + ( T 2 c x / ) ( ) ( )/ T t c T T c x c c d 1 ) , x ( ) , and / ): ): t t / ctrl Supplementary Fig. 8b Fig. Supplementary − GE ) − d GR , where x 1 t GR values can be evalu C ctrl x ) ( ) ctrl Fig. in 50 T − d f is the 50%-trimmed 1

− h = GR 1 ( x 1 ln(2)/ c ( , Fig. Fig. ), c Supplementary ) is the count in ). GR data are 1 k d (0) (0) can be x ). 0 inf is the the is

≡ Cell Cell GR ): ): - - - -

sponds to the concentration at which a drug is fully cytostatic: GR( cytostatic: fully is drug a which at concentration the to sponds GR example, For manner. similar a in defined be can + to GR the 0.5, above is tration of at drug GR( which metrics. drug-response Inferred ( 0 to set is P GR( (i.e., curve flat particularly sensitive to outlier values when directly obtained obtained directly data. from when values outlier to sensitive particularly GR fitting; curve from is derived metrics than noise fitting curve and GR measured, unreliable. are concentrations five of than fewer free where assays for advantage therefore useful is especially is and This the artifacts. fitting fitting has curve require not does concentrations it that determined) imentally GR of calculation The time. same the at range. concentration drug same the across evaluated responses compare to used be only should GR that note to important and GR example, for as, centrations GR The 0. of value a c GR( where curve: the over area is the it use to values, intuitive more negative have can which curves, GR of case the In over the range of concentrations. tested curve the dose–response integrating on based is which (AUC), curve response the under area the is data dose–response quantifying for metric (GR common curve the over and curve the under Area individually. point latter. the do often we from the fitted orcurve GR obtained directly from experimental data; response. cytotoxic a to corresponds value negative a and response, cytostatic fully a to 0 corresponds of value a 1; and −1 between lies it and concentration, tested i = 0.05, the response is considered flat and the parameter GEC . GR . The GR The a of that than better significantly not is curve the of fit the If • • The GR The time each at evaluated are metrics all data, course time For

 c to 10 to 10 about usually is this practice, (in range concentration of magnitude higher and lower than the experimentally tested artefacts in fitting curve we constrain GEC typically constrain we practice, In is. curve dose–response the steep h GEC c = GR = min GR ( AOC Supplementary Fig. 8c Fig. Supplementary : the Hill coefficient of the fitted curve, which reflects how are the highest and lowest tested concentrations. It is is It concentrations. tested lowest and highest the are 50 3 has the benefit that, in the case of no response, it has has it response, no of case the in that, benefit the has AOC max

Supplementary Fig. 8d Fig. Supplementary c µ 100 : the concentration at half-maximal effect. To avoid avoid To effect. half-maximal at concentration the : i M). GR ) are measured GR values at discrete concentrations concentrations at discrete values GR measured ) are is the maximum effect of the drug at the highest highest the at drug the of effect maximum the is ) = 0. = ) value captures variation in potency and efficacy efficacy and potency in variation captures value h AOC GR AO to a value between 0.1 and 5. and 0.1 between value a to C values are also more robust to experimental experimental to robust more also are values c ) ) AOC 50 − = ≡ value is not defined and is therefore set set therefore is and defined not is value ∫ GR 1 can be normalized to the range of con of range the to normalized be can AOC inf GR c = GR ) based on an an on based ) ( ). By extension, other thresholds thresholds other extension, By ). values (like conventional AUC) conventional (like values AOC d c ) The GR The ). 50 c /log ) ) = 0.5. If the value for GR − ≅ 1 10 AOC ( ∑ 50 c c max i max doi: value is the concen the is value F GR at discrete (exper discrete at -test with cutoff of of cutoff with -test can be estimated estimated be can 50 / 10.1038/nmeth.3853 c ) ( to be two orders min c AOC i max , ), where where ), ). values are values 100 Another Another corre c max inf −5 50 - - - -

© 2016 Nature America, Inc. All rights reserved. concentration of drug that produces half-maximal cell killing. killing. cell half-maximal produces that drug of concentration where cycle: cell the of death independent cell induce that can drugs for model account The to cycle. generalized cell be also the of phase specific a at death cell time. sion rate division h SC c where manner: cell-cycle-dependent a in cells or killing rate division the decreasing either drugs with exponential, considered be can growth cell approximation, first killing, we developed a theoretical model of drug response. To the metrics or about assumptions of cell the different degree under cytostasis drug-response conventional and GR on time division response. drug of model Theoretical manuscript. current the in data the of all including datasets, example and materials, explanatory and tutorials various guide, user a contains website at calculator online an vide at scripts and an under MATLABlicense open software source and python available code source updated provide we metrics GR of tation ( is provided GR metrics. Computing GR cases, such In concentration. measured highest the at plateau a reach GR( the if constrained properly GR Similarly, values. interpolated fitting than to artefacts subject more are curves fitted values the from because extrapolated concentration tested highest the above nitude GR any discarding We range. suggest concentration same the over evaluated if only caveat that GR the with drugs potent less for concentrations higher to or drugs This range can be shifted to lower concentrations for more potent 10 to nM 1 from magnitude of orders four spanning doses elsewhere has discussed been curves dose–response of design Optimal curves. dose–response steep of case the in especially estimates, precise more provides GEC for estimates reliable wide sufficiently a span range in values and intermediate order have to obtain sufficiently to need curves drug-response fitting range. concentration Drug doi: is the drug concentration, concentration, drug the is is the Hill coefficient. The growth rate rate growth The coefficient. Hill the is 50 10.1038/nmeth.3853 is the concentration at half-maximal effect of drug, and and drug, of effect half-maximal at concentration the is x k is the cell count, count, is cell the L is the maximal killing rate (per day) and LC and day) (per rate killing maximal the is AOC https://github. S x x  M × = Supplementary Software Supplementary k can be larger than 1 to account for drugs inducing inducing drugs for account to 1 than larger be can is the most reliable metric. reliable most the is 0 AOC as as k values for values should be drugs compared different k 50    = ln(2) × × ln(2) = x x  1 value that is more than an order of mag of order an than more is that value × = − SC com/sorgerlab/gr50_t c S k Source code for code computing GR Source metrics M is the untreated growth rate (per day), rate (per growth is untreated the k h 50 h 50    S × The drug concentrations used for for used concentrations drug The , , 7 1 + M k http://www.grcalculator.or . In practice, . we In using nine suggest practice, h − 0 c is the maximal inhibitory effect, effect, inhibitory maximal the is h GR c = ln(2)/ = SC ) dose–response curve does not not does curve dose–response ) c S    , and GR and , M × − h 50 h × x + ). ). To compu the facilitate To simulate the effect of of effect the Tosimulate h c    T LC d    , where where , c k k inf , L corresponds to the the to corresponds h 50 h . Denser sampling sampling Denser . × ool + h c s inf . We pro . also T value is not not is value    d is the divi the is , 50 g is the the is . This This . µ M. M. - - - -

This equation for GR( equation This is: value) (GR inhibition rate growth normalized the and where is: count cell relative the Thus, where at concentration count of assay an for equations these Integrating and and rate growth untreated ( cycle cell the to of independent t sampled within the following distribution: following the within sampled For follows: as plemental figures ( sup and main the in shown simulations numerical the in used length assay and time division (see both on depend still they but al. et Heiser in used values (GI) inhibition growth from derived rics GR for mula GR on small GR and, thus, the metrics GR metrics the thus, and, • • • • • • • • • • Model parameters panels. in figure for Parameters simulations IC

) ( Supplementary Figure 3b Figure Supplementary 3a ranging from 0.31 to 10 to 0.31 from ranging 0.14 of bound lower a and 0.46 of s.d. with day h h 5, = Half inhibition concentration concentration inhibition Half coefficient Hill rate Division cytotoxic: Complete response: Mixed response: toxic Partial Cytotoxic: response: Partial Cytostatic: 0 and 2 and 0 Maximum inhibition inhibition Maximum Supplementary Note Supplementary t c 3 ) ( = 1.6 = 1.6 = , c x t c , such as GI as such , , x ( , x x ) ( b 0 ctrl , the impact of of impact the , h =

, ≡ x t = 1.6 = = 2

x    ≡ c x × = inf ( max lo lo t = 0) is the cell number at the time of treatment. treatment. of time the at number cell the is 0) = : : GR / g ) , ( g S , ( 0 S 0 2 x t t 0 0 2 M Ti i as ilsrtd y h aayia for analytical the by illustrated also is This . . For cases where drug action is mainly related related mainly is action drug where cases For . M ) ( k c x k / x x t = 2.6, 2.6, = 50 L Fig. Fig. 1 = 1, 1, = ex ctrl : normal distribution around per 0.9 : divisions distribution normal inf ), = 0), GR values are also independent of the the of independent also are values GR 0), = h ctrl , are more robust than traditional metrics, metrics, traditional than robust more are , ) , x t p = 2^(1 – = 2^(1 : uniform distribution between 1.5 and 2.5 and 1.5 between distribution uniform : S c S     / k M ) is independent of the length of assay the of length the ) is independent c M k t S . As shown in in shown As . k = b : = 0.65, 0.65, = 50 − × S L = 0.65, 0.65, = ; ; S ex > 0 is minimal on GR on minimal is 0 > 50    ). = 1.5, 1.5, = Supplementary Figs. Supplementary 1 M 50    S − S = 2, 2, = p = 2.6, 2.6, = 1 M M , GR , 1    = 0.45, 0.45, = : uniform distribution between between distribution uniform : S = × − , the parameters were randomly randomly were parameters the , M SC k t S 2 T T c S – 50     max S M M M 1 S h 50 h 50 = 1.2, 1.2, = − = 0, 0, = k 50 S = 0, 0, = × SC = 1.2, 1.2, = L SC , GR , 50 + c S Supplementary Figures 2 Figures Supplementary : log-uniform distribution distribution log-uniform : c S / M M k S h = 1.2, 1.2, = c h 50 h 50 h ) – 1. Note that the met the Note ) – that 1. 50 h h × × × AUC    = 1.2, 1.2, = = 1.6 = T + + = 1.6 = − M t h c c h days yields the cell cell the yields days T h t = 0, 0, = , and and , M LC − T − = 0.1, 0.1, = c k k 1 50 M t n T L , h 50 h LC LC a 2 = 1, 1, = and relatively relatively and h M × k ture methods c k = 1.6 = , , and L h + 50 h = 0.05, 0.05, = L h 50 h GR h × c + c + T T are also also are h     c h 7 50 50 c h , ) ) were = 3, 3, =     = 3, 3, =    − T , 1 50 - - - .

© 2016 Nature America, Inc. All rights reserved. Evaluating drug-response metrics in MCF 10A and BT-20 BT-20 and 10A time. MCF over in metrics drug-response Evaluating addition. drug after h 72 evaluated was sensitivity drug and of treatment, at EGF time the starting Bioscience) (Essen imager in live-cell were ZOOM an imaged cells IncuCyte H2B-mCherry 10A- MCF h. 72 after and treatment drug of time the at analysis for fixed and stained were cells RPE-1 Dispenser. Digital D300 a using etoposide of h series 24 dilution a with After treated were cells the Dispenser. Digital D300 a using Peprotech) (EGF, factor growth epidermal human of doses indicated with treated using an EL406 Microplate Washer Dispenser (BioTek). Cells were penicillin– 1% and streptomycin. Medium changes and cell washing were performed albumin serum supple bovine medium 0.1% with DMEM/F12 mented with twice we cells 10A-H2B-mCherry serum-starved MCF in rate growth the modulate To of doxycycline using a D300 Digital Dispenser (Hewlett-Packard). of the BRAF expression induced we cells, RPE-1 in rate growth the modulate at 250 Toand respectively. well, per 500 cells Scientific) (Thermo plates in using the Multidrop 384-well Combi Dispenser Reagent drug on sensitivity. effects determine to rate growth cell Manipulating caspase-3 488 nM. at 200 used was NucView (Biotium) substrate respectively; nM, 100 and nM 250 at cells, YOYO-1 and TOTO-3 (Thermo Fisher were Scientific) used plates using a D300 Digital Dispenser (Hewlett-Packard). To stain multiwell into directly dispensed were dyes reporter and Drugs ( database collection drug tested for purity inhouse as in described detail in the HMS LINCS Drugs and dyes. analysis. to prior mycoplasma of free be to found and (Lonza) kit detection PLUS mycoplasma MycoAlert the with tested were repeat cells and all Institute, Cancer tandem at Dana-Farber the profiling (STR) short by confirmed was identity Cell 41394). # # 15269) BRAF the full-length inserting by created were J. Chen) from (gift FluoroBrite cells hTERTmodified RPE-1 The with imaging. for replaced Scientific) Fisher was (Thermo DMEM DMEM traditional that tion excep the with strain, as grown in parental the manner same the were cells 10A-H2B-mCherry MCF CRISPR/Cas9. using locus Jaenisch, R. Addgene plasmid # from 22072) (gift terminator polyA SV40 and promotor, 20972) # mCherry expression cassette (gift of R. H2B- Benezra, Addgene an plasmid inserting by modified were cells BT-20 experiments, and 10A MCF time-lapse For recommendations. ATCC accord to grown ing and ATCC the from obtained were BT-20 and sue culture. processing. data and methods Experimental cells were plated at 1,250 and 2,500 cells per well, respectively, in respectively, well, per cells 2,500 and at 1,250 plated were cells n a • • ture methods

ranging from 0.56 to 5.6 to 0.56 from ranging values of 50% other the for 0.5 and 0 between tribution Half inhibition concentration concentration inhibition Half effect toxic Maximum 4 3 0 8 driven by a tet-inducible promotor by driven plasmid (Addgene a tet-inducible MCF10 A-H2B-mCherry and BT-20-H2B-mCherry BT-20-H2B-mCherry and A-H2B-mCherry MCF10 that comprised AAVS1 homology arms, the hPGK hPGK the arms, homology AAVS1 comprised that MCF Hs 10A, 578T, MDA-MB-231, SK-BR-3, MCF7, RPE-1 or MCF 10A-H2B-mCherry cells were plated plated were cells 10A-H2B-mCherry MCF or RPE-1 V600E Drugs were obtained from commercial vendors and oncogene by treating cells with indicated doses doses by oncogene with indicated cells treating V600E http://lincs.hms expression cassette (Addgene plasmid plasmid (Addgene cassette expression 3 9 T into the AAVS1 safe harbor genomic M : 0 for 50% of values, uniform dis uniform values, of 50% for 0 : T 50 : log-uniform distribution distribution log-uniform : .harvard.edu/db/sm

Cell lines and tis and lines Cell / ). ). - - - - -

imaged imaged in an IncuCyte ZOOM live-cell imager (Essen Bioscience) of omipalisib using a series D300 Digital Dispenser (Hewlett-Packard) dilution and a with treated were cells h, 48 After medium. reduced growth the factor to (Corning) growth Matrigel matrix 2.5% membrane of basement addition the with plates 384-well (Corning) flat-bottom, attachment-coated, ultralow into well per cells 200 at plated were cells A-H2B-mCherry MCF10 time. over spheroids 10A MCF in sensitivity drug Evaluating drugs: following the used we experiments, (PerkinElmer) System with a equipped chamber over live-cell of a period 96 h. For these Imaging High-Content Operetta addition an drug in after Digital imaged D300 and a using (Hewlett-Packard) drugs Dispenser indicated the of series dilution a with treated were Cells h. 24 for grown and Scientific) (Thermo Dispenser Reagent Combi Multidrop the using plates 384-well plated in 20–120 20–120 in plated h. 72 of period a over (PerkinElmer) chamber live-cell a with System equipped Imaging High-Content Operetta addition an drug in after imaged and (Hewlett-Packard) Dispenser Digital D300 a using drugs of series dilution a with treated were and grown for 24 Scientific) h. (Thermo Cells Dispenser Reagent plates per well cells in using the 5,000 Multidrop 384-well Combi to 156 from ranged that densities at plated were cells mCherry effects. drug density-dependent Investigating drugs: following the after 72 h of incubation with drug. For and these experiments, we treatment used drug of time the at analysis for fixed and were stained Cells (Hewlett-Packard). Dispenser Digital D300 a drugs using indicated the of series dilution a with treated were Cells Combi Reagent Dispenser (Thermo Scientific) and grown for 24 h. Multidrop the using plates 384-well in well per cells to 5,000 156 MCF7, SK-BR-3, and BT-20 were plated at densities ranging from densities. seeding ferent at dif plated cells cancer breast in sensitivity drug Evaluating frame. time h 10–90 a over h 15 every values GR ing for an 96 by was h. evaluated additional sensitivity computDrug series of drug, and imaged for 72 h. 72 for imaged and drug, of series • • • • • In the case of methotrexate and oligomycin, 1,250 cells were were cells 1,250 oligomycin, and methotrexate of case the In • • • • • • • • • • •

Tanespimycin/17-AAG, HSP90 inhibitor HSP90 Tanespimycin/17-AAG, inhibitor B-RAF PLX4720, inhibitor panPI3K/mTOR Omipalisib/GSK2126458, inhibitor IGF1R Linsitinib, inhibitor topoisomerase Etoposide, Tanespimycin/17-AAG, HSP90 inhibitor HSP90 Tanespimycin/17-AAG, inhibitor ALK TAE684, inhibitor B-RAF PLX4720, inhibitor CDK4/6 Palbociclib, microtubules target Paclitaxel, inhibitor panPI3K/mTOR Omipalisib/GSK2126458, inhibitor reductase dihydrofolate Methotrexate, inhibitor IGF1R Linsitinib, inhibitor EGFR/ErbB2 Lapatinib, inhibitor topoisomerase Etoposide, inhibitor EGFR Erlotinib, µ L of medium per well, treated with a dilution dilution a with treated well, per medium of L MCF 10A, Hs 578T, MDA-MB-231, MDA-MB-231, 578T, Hs 10A, MCF doi: 10.1038/nmeth.3853 MCF 10A-H2B- MCF - -

© 2016 Nature America, Inc. All rights reserved. in an Operetta High-Content Imaging System (PerkinElmer) (PerkinElmer) System equipped with a live-cell chamber or an IncuCyte Imaging ZOOM live-cell High-Content Operetta treatment an drug in after timepoints indicated the at imaged were assays. course time Live-cell (PerkinElmer). system analysis and storage data image Columbus the using and analyzed microscope an using Operetta imaged were cells Fixed min. 30 for (Sigma-Aldrich) maldehyde Stain (Thermo Fisher Scientific) for 30 min and fixed with 3% for Cell Dead Red Far LIVE/DEAD and Scientific) Fisher (Thermo 2 with timepoints indicated the at stained assays. endpoint Fixed-cell (Thermo Fisher and Scientific), washed three times in PBS. 30 min with whole for cellstained stainPBS, with (Thermo once Fisherwashed PBS-T, in Scientific) times andtwo Hoechstwashed in 1:1,000 blocking buffer Odyssey for 60 min at room. wereCells Fluor 488-conjugated goat anti-rabbit antibody secondary diluted in for PBS-T times 5 Alexa three with washed min and incubated Odyssey blocking buffer and incubated for 16 h at 4 °C. Cells were active Caspase-3 antibody (BD Biosciences) was diluted 1:1,000 in 60 min with Odyssey blocking buffer (LI-COR Biosciences). Anti- PBS with 0.1% Tween 20 (Sigma-Aldrich; PBS-T), and blocked for in twice in washed PBS with 0.3% Triton (Sigma-Aldrich), X-100 min 30 for permeabilized formaldehyde, 3% in min 30 for fixed (Hewlett-Packard) and incubated for 3, 6, 12, and 24 h. Cells were treated with a of then dilution series using paclitaxel a Dispenser D300 Digital and h 24 for grown were cells experiments, cence immunofluores For h. 72 additional an for Bioscience) imager (Essen live-cell ZOOM IncuCyte an in drug after (Hewlett-Packard) imaged and Dispenser Digital D300 a using substrate caspase-3 (Biotium) 488 NucView of nM 200 and paclitaxel of h. 72 additional an for Bioscience) live-cell (Essen imager ZOOM IncuCyte an in imaged and Dispenser Digital of linsitinib either with or without 10 doi: In the of case cells were paclitaxel, treated with a dilution series series a with were dilution treated cells In of case the linsitinib, 10.1038/nmeth.3853 After drug treatment, cells were were cells treatment, drug After Cells expressing H2B-mCherry H2B-mCherry expressing Cells µ M M batimastat using a D300 µ M Hoechst 33342 33342 Hoechst M - -

the MSigDB v4.0 GO biological process set process with biological GO v4.0 Institute MSigDB the Broad the from v2.1.0 software performed GSEA the was using analysis enrichment set Gene scripts. and inhouse libraries standard using MATLAB in performed was data (ref. v.2.2.1 Cuffquant using pipeline computational BTL the by quantified were Transcripts ref. in described for SCRB-Seq protocol the following (BTL), Labs Technology Broad the by prepared were Libraries pooled. amounts of RNA, sufficient wells with a low number of were cells RNA was extracted using the RNeasy mini kit (Qiagen). To ensure and times, indicated at the harvested were plates 384-well in ties mRNA analysis. software. analysis IncuCyte using well per area spheroid the of sum the measuring the IncuCyte analysis software. or Spheroid growth was system estimated identified by analysis and were storage data lines image Columbus cell the using apoptotic or dead, Live, (Biotium). substrate caspase-3 488 NucView the with identified were cells apoptotic and Scientific), Fisher (Thermo TOTO-3 YOYO-1or with cells counterstaining by identified were cells Dead imager. 43. 42. 41. 40. 39. 38.

Acad. Sci. USA Sci. Acad. profiles.genome-wideexpression interpreting for approach (2012). Cufflinks. andTopHat with experiments RNA-seq earl at Preprint RNA-Seq. single-cell throughput high- by differentiationdirected of CharacterizationT.S. Mikkelsen, oncogene.cancer breast (2009). 851–857 nucleases.zinc-finger using iPSCs and ESCs human cells. stem neural adult Subramanian, A. Subramanian, C. Trapnell, & Oudenaarden,A. van S., Semrau, D., Cacchiarelli, M., Soumillon, J.S. Boehm, Hockemeyer,D. type B1 define expression Id1 of levels High R. Benezra, & H.S. Nam, y/2014/03/05/00323 et al. et et al. et

1 et al. et mRNA analysis. Cells plated at different densi at different plated Cells mRNA analysis. 0 et al. et Differential gene and transcript expression analysis of analysis expressiontranscript and geneDifferential 2 Integrative genomic approaches identify IKBKE as a as IKBKE identifyapproaches genomic Integrative , 15545–15550 (2005). 15545–15550 , Efficient targeting of expressed and silent genes in genes silent and expressed of targeting Efficient Gene set enrichment analysis: a knowledge-based a analysis: enrichment set Gene Cell Stem Cell Stem Cell 6 Cell (2014).

12 4 9 2 , 1065–1079 (2007). 1065–1079 , ). Analysis of the transcriptional transcriptional the of Analysis ).

5 , 515–526 (2009). 515–526 , http://bior 4 Nat. Protoc. Nat. 3 . Nat. Biotechnol. Nat. xiv.org/content/ n a ture methods Proc. Natl. Proc.

7 , 562–578 ,

27 , 4 1 - .