Skip to main content

Thank you for visiting You are using a browser version with limited support for CSS. To obtain the best experience, we recommend you use a more up to date browser (or turn off compatibility mode in Internet Explorer). In the meantime, to ensure continued support, we are displaying the site without styles and JavaScript.

  • Brief Communication
  • Published:

Mapping the in vivo fitness landscape of lung adenocarcinoma tumor suppression in mice


The functional impact of most genomic alterations found in cancer, alone or in combination, remains largely unknown. Here we integrate tumor barcoding, CRISPR/Cas9-mediated genome editing and ultra-deep barcode sequencing to interrogate pairwise combinations of tumor suppressor alterations in autochthonous mouse models of human lung adenocarcinoma. We map the tumor suppressive effects of 31 common lung adenocarcinoma genotypes and identify a landscape of context dependence and differential effect strengths.

This is a preview of subscription content, access via your institution

Access options

Buy this article

Prices may be subject to local taxes which are calculated during checkout

Fig. 1: Trp53 deficiency alters the growth effects of tumor suppression in KrasG12D-driven lung tumors in vivo.
Fig. 2: Attenuated effects of tumor suppressor inactivation in Lkb1-deficient tumors further highlights a rugged fitness landscape.

Similar content being viewed by others


  1. Cancer Genome Atlas Research Network. Nature 511, 543–550 (2014).

    Article  Google Scholar 

  2. Lawrence, M. S. et al. Nature 505, 495–501 (2014).

    Article  CAS  Google Scholar 

  3. Mina, M. et al. Cancer Cell 32, 155–168.e6 (2017).

    Article  CAS  Google Scholar 

  4. AACR Project GENIE Consortium. Cancer Discov. 7, 818–831 (2017).

    Article  Google Scholar 

  5. Kersten, K., de Visser, K. E., van Miltenburg, M. H. & Jonkers, J. EMBO Mol. Med. 9, 137–153 (2017).

    Article  CAS  Google Scholar 

  6. Frese, K. K. & Tuveson, D. A. Nat. Rev. Cancer 7, 645–658 (2007).

    Article  CAS  Google Scholar 

  7. Rogers, Z. N. et al. Nat. Methods 14, 737–742 (2017).

    Article  CAS  Google Scholar 

  8. Lenski, R. E. et al. Proc. Biol. Sci. 282, 20152292 (2015).

    Article  Google Scholar 

  9. de Visser, J. A. G. M. & Krug, J. Nat. Rev. Genet. 15, 480–490 (2014).

    Article  Google Scholar 

  10. Siegal, M. L. & Leu, J. Annu. Rev. Ecol. Evol. Syst. 45, 496–517 (2014).

    Article  Google Scholar 

  11. Keshelava, N. et al. Cancer Res. 61, 6185–6193 (2001).

    CAS  PubMed  Google Scholar 

  12. Sherr, C. J. & McCormick, F. Cancer Cell 2, 103–112 (2002).

    Article  CAS  Google Scholar 

  13. Canisius, S., Martens, J. W. M. & Wessels, L. F. A. Genome Biol. 17, 261 (2016).

    Article  Google Scholar 

  14. Ji, H. et al. Nature 448, 807–810 (2007).

    Article  CAS  Google Scholar 

  15. Chou, H.-H., Chiu, H.-C., Delaney, N. F., Segrè, D. & Marx, C. J. Science 332, 1190–1192 (2011).

    Article  CAS  Google Scholar 

  16. Wünsche, A. et al. Nat. Ecol. Evol. 1, 61 (2017).

    Article  Google Scholar 

  17. Burkhart, D. L. & Sage, J. Nat. Rev. Cancer 8, 671–682 (2008).

    Article  CAS  Google Scholar 

  18. Hardie, D. G. & Alessi, D. R. BMC Biol. 11, 36–47 (2013).

    Article  Google Scholar 

  19. Sánchez-Rivera, F. J. et al. Nature 516, 428–431 (2014).

    Article  Google Scholar 

  20. Walter, D. M. et al. Cancer. Res. 77, 1719–1729 (2017).

    Article  CAS  Google Scholar 

  21. Jackson, E. L. et al. Genes Dev. 15, 3243–3248 (2001).

    Article  CAS  Google Scholar 

  22. Madisen, L. et al. Nat. Neurosci. 13, 133–140 (2010).

    Article  CAS  Google Scholar 

  23. Jonkers, J. et al. Nat. Genet. 29, 418–425 (2001).

    Article  CAS  Google Scholar 

  24. Nakada, D., Saunders, T. L. & Morrison, S. J. Nature 468, 653–658 (2010).

    Article  CAS  Google Scholar 

  25. Chiou, S. H. et al. Genes Dev. 29, 1576–1585 (2015).

    Article  CAS  Google Scholar 

  26. Sage, J., Miller, A. L., Pérez-Mancera, P. A., Wysocki, J. M. & Jacks, T. Nature 424, 223–228 (2003).

    Article  CAS  Google Scholar 

  27. DuPage, M., Dooley, A. L. & Jacks, T. Nat. Protoc. 4, 1064–1072 (2009).

    Article  CAS  Google Scholar 

  28. Rogers, Z. Protoc. Exch. (2017).

    Article  Google Scholar 

  29. Rogers, Z. Protoc. Exch. (2017).

    Article  Google Scholar 

  30. Caunt, M. et al. Cancer Cell 13, 331–342 (2008).

    Article  CAS  Google Scholar 

  31. Curtis, C. et al. Nature 486, 346–352 (2012).

    Article  CAS  Google Scholar 

Download references


We thank P. Chu and R. Ma for technical support, A. Orantes for administrative support, members of the Petrov and Winslow laboratories for comments, and T. Jacks for initial support. Z.N.R. and I.P.W. were supported by the National Science Foundation Graduate Research Fellowship Program (GRFP). Z.N.R. was additionally supported by a Stanford Graduate Fellowship. I.P.W. was additionally supported by NIH F31CA210627 and NIH T32HG000044. C.D.M. was supported by NIH E25CA180993. J.J.B. was supported by NIH F32CA189659. J.A.S. was supported by Susan G. Komen for the Cure PDF16377256. D.P. is the Michelle and Kevin Douglas Professor of Biology. This work was supported by NIH R01CA175336 (to M.M.W.), R01CA207133 (to D.P. and M.M.W.), and in part by the Stanford Cancer Institute support grant (NIH P30CA124435). The authors would like to acknowledge the American Association for Cancer Research and its financial and material support in the development of the AACR Project GENIE registry, as well as members of the consortium for their commitment to data sharing. Interpretations are the responsibility of study authors.

Author information

Authors and Affiliations



Z.N.R. generated barcoded vectors, produced lentivirus and performed mouse analysis, indel analysis and analysis of single sgRNA tumor sizes. C.D.M. performed data analysis, including processing sequencing data, and all statistical analyses. I.P.W. generated the mouse models, selected tumor suppressors to investigate, designed sgRNAs, generated Lenti-sgRNA/Cre vectors, tested sgRNA cutting efficiency, produced lentivirus and performed indel analysis. J.A.S. and C.C. performed analysis of human cancer datasets. S.Y. and J.J.B. contributed to experiments and analyses. D.A.P. and M.M.W. oversaw the project. Z.N.R., C.D.M., D.A.P. and M.M.W. wrote the manuscript with comments from all authors.

Corresponding authors

Correspondence to Dmitri A. Petrov or Monte M. Winslow.

Ethics declarations

Competing interests

Stanford University has filed a patent (US Provisional Application 62/481,067) based on this work in which Z.N.R., I.P.W., C.M., D.A.P., and M.M.W. are coinventors.

Additional information

Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Integrated supplementary information

Supplementary Figure 1 The current state of genetically-engineered mouse models of lung cancer for the analysis of the putative tumor suppressor alterations in this study and the frequency of these genomic alterations in human lung adenocarcinoma.

a. Summary of data from published studies in which the putative tumor suppressor genes studied here were inactivated in the context of oncogenic Kras-driven lung cancer models, with or without inactivation of p53 or Lkb1. b. The percent of tumors with potentially inactivating alterations (frameshift or non-synonymous mutations, or genomic loss) in each tumor suppressor gene for all tumors (All) as well as for tumors with potentially inactivating alterations in TP53 (TP53mut) or LKB1 (LKB1mut). The percent of tumors with each type of alteration is indicated. Data is shown for two clinical cancer genomics studies: The Cancer Genome Atlas (TCGA, 2014)32, and the Genomics Evidence Neoplasia Information Exchange (GENIE, 2017) database4.

Supplementary Figure 2 Description of multiplexed lentiviral vectors, tumor initiation, and Tuba-seq pipeline to quantify tumor size distributions in vivo.

a. Lenti-sgTS-Pool/Cre contains four vectors with inert sgRNAs and eleven vectors with tumor suppressor gene targeting sgRNAs. Each sgRNA vector contains a unique sgID and a random barcode. NT = Non-Targeting. b. Schematic of the sgID-barcode region of the vectors in Lenti-sgTS-Pool/Cre. Lenti-sgTS-Pool/Cre contains vectors with fifteen different 8-nucleotide unique identifiers (sgIDs) which link a given sgID-barcode read to a specific sgRNA. These vectors also contain a 15-nucleotide random barcode element. This double barcode system allows identification of individual tumors, as well as the sgRNA in the vector that initiates each tumor. c. Transduction of lung epithelial cells with the barcoded Lenti-sgTS-Pool/Cre pool initiates lung tumors in genetically engineered mouse models with (1) a Cre-regulated oncogenic KrasG12D (KrasLSL-G12D/+) allele, (2) a Cre reporter allele (Rosa26LSL-Tomato), (3) a Cre-regulated Cas9 allele (H11LSL-Cas9), as well as (4) homozygous floxed alleles of either p53 or Lkb1. Lentiviral vectors stably integrate into the genome of the transduced cell. Tumors were initiated in KT;Cas9, KPT;Cas9, and KLT;Cas9 mice to generate 31 different genotypes of lung tumors. Mice were analyzed after 15 weeks of tumor growth. Genomic DNA was extracted from whole lungs, after the addition of barcoded “benchmark” cell lines, the sgID-barcode region was PCR amplified, deep-sequenced, and analyzed to determine the relative expansion of each uniquely barcoded tumor using the Tuba-seq pipeline.

Supplementary Figure 3 Tumor suppression in KrasG12D-driven lung adenocarcinoma in vivo.

a. Fold change in sgID representation (ΔsgID representation) in KT;Cas9 mice relative to KT mice, which lack Cas9 and therefore should not expand relative to sgInert. Several sgRNAs (sgIDs) increase in representation, reflecting the increased growth of tumors with inactivation of the targeted tumor suppressor genes. Means and 95% confidence intervals are shown. b,c. The ability to detect tumor suppressive effects is improved by analyzing individually-barcoded tumors compared to bulk sgRNA representation (ΔsgID representation). (b) Analysis of the relative size of the 95th percentile tumor with each sgRNA identifies somewhat similar estimates of relative tumor size as bulk ΔsgID representation, which exhibits wider confidence intervals and generally weaker effect sizes. (c) P-value of the Log-Normal mean (LN mean) measure of relative tumor size versus P-value ΔsgID representation. Because individual tumor sizes are measured and then properly normalized to eliminate exogenous sources of noise, both the 95th percentile and LN Mean metrics identify functional tumor suppressors with greater confidence and precision. p53 loss is an exception, as its growth effects are poorly described by a Log-Normal distribution. All P-values are two-sided and obtained via 2 × 106 Bootstrapping permutation tests and a Bonferroni-correction for the number of investigated tumor suppressors. d-f. Same as in a-c, except for growth effects in KPT;Cas9 mice. Fold change is relative to KT mice, while 95th percentile and LN Mean size estimates are relative to KPT;Cas9 internal sgInert controls. g-i. Same as in a-c, except for growth effects in KLT;Cas9 mice. No tumor suppressors would have been identified without Tuba-seq.

Supplementary Figure 4 Rb and p53 tumor suppressor cooperativity in lung adenocarcinoma identified by Tuba-seq, confirmed in a mouse model using Cre/lox regulated alleles, and supported by the co-occurrence of RB1 and TP53 mutations in human lung adenocarcinoma.

a. Relative LN Mean size of sgSetd2, sgLkb1 and sgRb1 tumors. Rb1 inactivation increase tumor size less that Setd2 or Lkb1 inactivation in the p53-proficient KT;Cas9 background. Conversely, Rb1 inactivation increases tumor size to a similar extent as Setd2 or Lkb1 inactivation in the p53-deficient KPT;Cas9 background. P-values test null hypothesis of similar LN Mean to sgRb1. P < 0.05 in bold. b. H&E staining of representative lung lobes from KP and KP;Rb1flox/flox mice with tumors initiated with Adeno-CMV/Cre. Mice were analyzed 12 weeks after tumor initiation. Scale bars = 500 μm. c. Representative ex vivo µCT images of the lungs from KP and KP;Rb1flox/flox mice are shown. Lung lobes are outlined with a dashed white line. d. Quantification of percent tumor area in K;Rb1wt/wt, K;Rb1flox/flox, KP;Rb1wt/wt, and KP;Rb1flox/flox mice. Histological quantification confirms that Rb1-deletion increases tumor burden more dramatically in p53-deficient tumors. * P-value <0.05, n.s. = not significant. Titer of Ad-Cre is indicated. e,f. Co-occurrence of RB1 and TP53 mutations in two human lung adenocarcinoma genomics datasets: (e) TCGA 2014 dataset, and (f) the GENIE consortium 2017. P-values were calculated using the DISCOVER statistical independence test for somatic alterations (Methods)13. Combined data from these datasets is shown in Fig. 1g.

Supplementary Figure 5 Deep sequencing of targeted genomic loci confirms creation of indels at all targeted loci and shows selective expansion of cancer cells with indels in the strongest tumor suppressor genes.

a. Indel abundance in each region targeted by sgRNAs, as determined by deep sequencing of total lung DNA from the targeted regions of four KPT;Cas9 mice. Indel abundance is normalized to the median abundance of sgNeo1, sgNeo2, and sgNeo3. Error bars denote range of abundances observed, while dots denote median. Indels were observed in all targeted regions. sgp53 is not shown, as its target site is deleted by Cre-mediated recombination of the p53floxed alleles. b. Indel abundance as described in (a) versus the 95th percentile tumor size determined by Tuba-seq (as described in Fig. 1d). Each dot represents a single sgRNA in an individual mouse and each mouse is represented by a unique shape. Indel abundance correlated with Tuba-seq size profiles (as expected), however indel abundance does not measure individual tumor sizes and exhibits greater statistical noise. The largest single tumor in this entire analysis, as determined by Tuba-seq, was an sgCdkn2a tumor that similarly appeared as an outlier in the indel analysis—further corroborating faithful analysis of genetic events by Tuba-seq.

Supplementary Figure 6 Validation of the redundancy between Setd2 and Lkb1 in mouse models and in human lung adenocarcinomas.

a. Fluorescence dissecting scope images (top) and H&E stained section (bottom) of lung lobes from KPT and KPT;Cas9 mice with Lenti-sgSetd2#1/Cre or Lenti-sgNeo2/Cre initiated tumors. Mice were analyzed after 9 weeks of tumor growth. Lung lobes are outlined with a white dashed line in fluorescence dissecting scope images. Top scale bars = 5 mm. Bottom scale bars = 4 mm. b. Quantification of percent tumor area in KPT;Cas9 mice with Lenti-sgSetd2#1/Cre or Lenti-sgNeo2/Cre initiated tumors, and KPT mice with Lenti-sgSetd2#1/Cre initiated tumors. Each dot represents a mouse and horizontal bars are the mean. There is an increase in tumor area between KPT;Cas9 and KPT mice with tumors initiated with the same virus, but no difference between KPT;Cas9 mice tumors initiated with Lenti-sgSetd2#1/Cre and those initiated with Lenti-sgNeo2/Cre, presumably due to high mouse-to-mouse variability. Because these lentiviral vectors were barcoded, we performed Tuba-seq analysis of these mice to quantify the size of induced tumors (shown in Fig. 2). sgSetd2 increased tumor sizes in KPT;Cas9 relative to sgNeo2. ** P < 0.01, n.s. is not significant. c,d. Same as a,b except for KLT;Cas9 mice with Lenti-sgSetd2#1/Cre or Lenti-sgNeo2/Cre initiated tumors. Mice were analyzed after 9 weeks of tumor growth. Top scale bars = 5 mm. Bottom scale bars = 4 mm. Tuba-seq analysis is shown in Fig. 2. e,f. The co-occurrence of SETD2 and LKB1 (HGNC name STK11) in two human lung adenocarcinoma genomics datasets: (e) TCGA 2014 dataset32 (N = 229 patients), and (f) the GENIE Consortium4 (N = 1563 patients). Two-sided P-values were calculated using the DISCOVER statistical independence test. Combined data from these datasets is shown in Fig. 2.

Supplementary Figure 7 Correspondence of Tuba-seq fitness measurements to human genomic patterns.

a. Relative fitness measurements and human co-occurrence rates of the nineteen pairwise interactions that we investigated. LN Mean Ratio is the ratio of relative LN Mean (sgTS/sgInert) within the background of interest divided by the mean relative LN mean of all three backgrounds. Background rate can be either an unweighted average of the three backgrounds (raw), or weighted by each background’s rate of occurrence in human lung adenocarcinoma (weighted). *OR = “Odds Ratio” of the co-occurrence rate of a gene pair within the human data. One-sided P-values of human co-occurrence rates (>0.5 suggest mutual exclusivity) were determined using the DISCOVER test. Combined P-values generated using Stouffer's Method (Methods). P < 0.025 and P > 0.975 are in bold. Fitness measurements and co-occurrence rates generally correspond (Spearman’s r = 0.50, P-value = 0.03 for weighted LN Mean Ratio; r = 0.4 for unweighted). b. Graphical summary of fitness measurements and co-occurrence rates from a. Human Genetics Cooperativity is defined as a Combined Odds Ratio >1 and Redundant if <1. Tuba-seq Cooperativity is defined as a LN Mean Ratio >1 and Redundant if <1. c. Number of statistically-significant genetic interactions suggested from a pan-cancer analysis of twenty-one tumor types3. Tumor types abbreviations are borrowed from TCGA. Lung adenocarcinoma (LUAD) is black and is predicted to contain a quantity of genetic interactions that is similar to the median, suggesting that the ruggedness of the fitness landscape studied here may be representative of cancer evolution in general.

Supplementary Figure 8 Power analysis of larger genetic surveys.

By assuming lognormal tumor size distributions, the statistical power of Tuba-seq to detect driver growth effects and non-additive driver interactions in larger genetic surveys can be projected. Future experiments could utilize larger mouse cohorts and larger pools of sgRNAs targeting putative tumor suppressors. In all hypothetical experiments, the Lenti-sgTS-Pool/Cre titers and fraction of the pool with inert sgRNAs (for normalization) were kept consistent with our original experiments. a. P-value contours for the confidence in detecting a weak driver (parameterized by the sgCdkn2a distribution in KT;Cas9 mice). Any experimental setups above a contour detects weak drivers with a confidence greater than or equal to the P-value of the contour. b,c. Same as in a, except for moderate and strong drivers respectively (parameterized by sgRb1 and sgLkb1 in KT;Cas9 mice). sgRNA pool size is extended to 500 targets (instead of 100 targets in a pool) because larger screens are possible when investigating genes with these effect strengths. d-f. Same as in a-c, except for driver interactions. Driver interactions (LN Mean Ratio) are defined as a ratio of driver growth rates (sgTS/sgInert in background #1)/(sgTS/sgInert in background #2) that were statistically different from the null hypothesis of one. (d) A weak driver interaction parameterized by Rbm10p53 (7% effect size). (e) A moderate driver interaction parameterized by Rb1p53 (13% effect size). (f) A strong driver interaction parameterized by Setd2Lkb1 (68% effect size).

Supplementary information

Supplementary Figures

Supplementary Figures 1–8

Life Sciences Reporting Summary

Supplementary Table 1

Predicted off-target exonic Cas9 endonuclease sites of sgRNAs used in this study. The specificity of each sgRNA was predicted from a statistical model of mammalian Cas9 endonuclease activity described in Supplementary Table 2. Unintended ‘off-target’ endonuclease activity is predicted to be <2% of intended activity for all regions of the mouse exome. Off-target genes were also compared to the 2016 COSMIC census of 603 putative driver (carcinogenic) genes. Only two off-target sites target putative driver genes, neither of which have literature evidence for a role in lung cancer

Supplementary Table 2

Predicted genome-wide Cas9 endonuclease off-target sites of sgRNAs used in this study. The specificity of each sgRNA was predicted from a statistical model trained on the endonuclease activities of 700 sgRNAs targeting Cas9 in mammalian genomes. For each sgRNA used in this study, a list is shown of off-target sequences and their associated PAM motifs within the mouse genome (mm9). Off-target sites are ranked in the order of their score from 100% to 0% (predicted cutting activity). On-target sequences are also shown at the top of each list. For off-targets within an annotated gene, the corresponding Ensembl gene ID is indicated

Rights and permissions

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Rogers, Z.N., McFarland, C.D., Winters, I.P. et al. Mapping the in vivo fitness landscape of lung adenocarcinoma tumor suppression in mice. Nat Genet 50, 483–486 (2018).

Download citation

  • Received:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI:

This article is cited by


Quick links

Nature Briefing: Translational Research

Sign up for the Nature Briefing: Translational Research newsletter — top stories in biotechnology, drug discovery and pharma.

Get what matters in translational research, free to your inbox weekly. Sign up for Nature Briefing: Translational Research