Search

Pool composition of in vivo screen.

Genes and alleles selected for the project.

Annotation of 1163 ORFs.

Comparison to in silico methods.

Supplementary Figure S1. Distribution of barcode read representation in pre-expansion and pre-injection samples. Supplementary Figure S2. Tumor composition of in vivo pooled screen, excluding the pools shown in Figure 2. Supplementary Figure S3. Gene expression differentiates functional alleles. Supplementary Figure S4. Validation of rare oncogenic alleles, excluding the ones shown in Figure 4. Supplementary Figure S5. Gene expression signatures of NFE2L2 wild type and gain-offunction mutants are correlated. Supplementary Figure S6. Comparison to in silico methods.

L1000 gene expression data of 1036 ORFs.

Composition of cells and tumors from the in vivo screen.

Background: A multi-analyte blood test has the potential to maximize performance for early detection across different cancer stages and types. Improvements in early-stage cancer detection might be achieved using multi-component tests with high sensitivities and specificities. We recently performed a large feasibility study to assess the performance of 4 biomarkers (aneuploidy, methylation, mutation, and protein) for the detection of cancers from up to 15 organ sites. Specifically, a training and validation set was tested for 3 biomarkers (aneuploidy, methylation, and protein) and the performance was subsequently confirmed in an independent testing set. Methods: We have now further improved the performance of a 4-marker cancer detection blood test by fine-tuning the respective marker calling models and thresholds, exploring prostate-specific antigen (PSA) for prostate cancer detection, and developing an overarching Machine Learning (ML) cancer classifier. To improve the mutation detection, we tested (in triplicate) 200 plasma and buffy samples from young, non-cancer subjects and mutant DNA from cell lines to develop an ML-based mutation calling algorithm. This caller was validated on 186 samples and tested on an independent set of 1388 cancer and non-cancer samples. The calling of cancer-associated DNA methylation events was refined by performing training, validation, and testing across different studies. We also explored models for methylation detection based solely on distribution of methylation signal observed in non-cancer samples. Free and total PSA were investigated as markers for prostate cancer detection by including clinically relevant Gleason scores in the development of the protein-based cancer calling algorithm. Results: In the previous analysis the combination of mutation, aneuploidy, methylation, and protein biomarkers resulted in an overall sensitivity of 61.0% (95% CI: 56.9%-65.0) at a specificity of 98.2% (95% CI: 97.1 – 99.4%). We will present the added performance benefit of ML-based mutation variant calling. PSA derived features were evaluated with the goal of increasing the detectability of high-grade prostate cancers while minimizing the detection of indolent cancers. Lastly, we compared the Boolean logic-based 4-biomarker combination algorithm used in the previous analysis with an ML-based cancer classifier. The results of the modeling, applied to the testing set, will be shared. Conclusions: In summary, improvements in cancer detection performance may be achieved by optimizing each biomarker calling algorithm as well as overarching cancer classifier. When combining these improvements, we believe that a single blood test will provide robust sensitivity for the detection of several cancer types, particularly for earlier-stage disease in real world settings. Citation Format: Vladimir Gianullin, Leonardo Hagmann, Kevin Arvai, Amira Djebbari, Christopher L. Nobles, Larson Hogstrom, Mael Manesse, Vuna Fa, Fanglei Zhuang, Xi Chen, Viatcheslav E. Katerov, Jorge Garces, Hatim T. Allawi, Abigail McElhinny, Frank Diehl, Gustavo C Cerqueira. Improved sensitivity of a multi-analyte early detection test based on mutation, methylation, aneuploidy, and protein biomarkers. [abstract]. In: Proceedings of the AACR Special Conference: Precision Prevention, Early Detection, and Interception of Cancer; 2022 Nov 17-19; Austin, TX. Philadelphia (PA): AACR; Can Prev Res 2023;16(1 Suppl): Abstract nr P041.

Background: A multi-analyte blood test has the potential to maximize performance for early detection across different cancer stages and types. Improvements in early-stage cancer detection might be achieved using multi-component tests with high sensitivities and specificities. We recently performed a large feasibility study to assess the performance of 4 biomarkers (aneuploidy, methylation, mutation, and protein) for the detection of cancers from up to 15 organ sites. Specifically, a training and validation set was tested for 3 biomarkers (aneuploidy, methylation, and protein) and the performance was subsequently confirmed in an independent testing set. Methods: We have now further improved the performance of a 4-marker cancer detection blood test by fine-tuning the respective marker calling models and thresholds, exploring prostate-specific antigen (PSA) for prostate cancer detection, and developing an overarching Machine Learning (ML) cancer classifier. To improve the mutation detection, we tested (in triplicate) 200 plasma and buffy samples from young, non-cancer subjects and mutant DNA from cell lines to develop an ML-based mutation calling algorithm. This caller was validated on 186 samples and tested on an independent set of 1388 cancer and non-cancer samples. The calling of cancer-associated DNA methylation events was refined by performing training, validation, and testing across different studies. We also explored models for methylation detection based solely on distribution of methylation signal observed in non-cancer samples. Free and total PSA were investigated as markers for prostate cancer detection by including clinically relevant Gleason scores in the development of the protein-based cancer calling algorithm. Results: In the previous analysis the combination of mutation, aneuploidy, methylation, and protein biomarkers resulted in an overall sensitivity of 61.0% (95% CI: 56.9%-65.0) at a specificity of 98.2% (95% CI: 97.1 – 99.4%). We will present the added performance benefit of ML-based mutation variant calling. PSA derived features were evaluated with the goal of increasing the detectability of high-grade prostate cancers while minimizing the detection of indolent cancers. Lastly, we compared the Boolean logic-based 4-biomarker combination algorithm used in the previous analysis with an ML-based cancer classifier. The results of the modeling, applied to the testing set, will be shared. Conclusions: In summary, improvements in cancer detection performance may be achieved by optimizing each biomarker calling algorithm as well as overarching cancer classifier. When combining these improvements, we believe that a single blood test will provide robust sensitivity for the detection of several cancer types, particularly for earlier-stage disease in real world settings. Citation Format: Vladimir Gianullin, Leonardo Hagmann, Kevin Arvai, Amira Djebbari, Christopher L. Nobles, Larson Hogstrom, Mael Manesse, Vuna Fa, Fanglei Zhuang, Xi Chen, Viatcheslav E. Katerov, Jorge Garces, Hatim T. Allawi, Abigail McElhinny, Frank Diehl, Gustavo C Cerqueira. Improved sensitivity of a multi-analyte early detection test based on mutation, methylation, aneuploidy, and protein biomarkers. [abstract]. In: Proceedings of the AACR Special Conference: Precision Prevention, Early Detection, and Interception of Cancer; 2022 Nov 17-19; Austin, TX. Philadelphia (PA): AACR; Can Prev Res 2023;16(1 Suppl): Abstract nr IA023.

Summary The intestinal microbiome is a key determinant of responses to biologic therapy in inflammatory bowel disease (IBD). However, diverse therapeutics and variable responses among IBD patients have posed challenges in predicting clinical therapeutic success. In this prospective study, we profiled baseline stool and blood in patients with moderate-to-severe Crohn's disease or ulcerative colitis initiating anti-cytokine therapy (anti-TNF or -IL12/23) or anti-integrin therapy. Patients were assessed at 14 weeks for clinical remission and 52 weeks for clinical and endoscopic remission. Baseline microbial richness indicated preferential responses to anti-cytokine therapy and correlated with the abundance of microbial species capable of 7α/β-dehydroxylation of primary to secondary bile acids. Serum signatures of immune proteins reflecting microbial diversity identified patients more likely to achieve remission with anti-cytokine therapy. Remission-associated multi-omic profiles were unique to each therapeutic class. These profiles may facilitate a priori determination of optimal therapeutics for patients and serve as targets for newer therapies.

Delivery mode is the variable with the greatest influence on the infant gut microbiome composition in the first few months of life. Children born by Cesarean section (C-section) lack species from the Bacteroides genus in their gut microbial community, and this difference can be detectable until 6-18 months of age. One hypothesis is that these differences stem from lack of exposure to the maternal vaginal microbiome, as children born by C-section do not pass through the birth canal; however, Bacteroides species are not common members of the vaginal microbiome, thus this explanation seems inadequate. Here, we set out to re-evaluate this hypothesis by collecting rectal and vaginal samples before delivery from 73 mothers with paired stool from their infants in the first two weeks of life. We compared microbial profiles of infants born by planned, pre-labor C-section to those born by emergent, post-labor surgery (where the child was in the birth canal, but eventually delivered through an abdominal incision), and found no significant differences in the microbiome between these two groups. Both groups showed the characteristic signature lack of Bacteroides species, despite their difference in exposure to the birth canal. Surprisingly, this signature was only evident in samples from week two of life, but not in the first week. Children born by C-section often had high abundance of Bacteroides in their first few days of life, but these were not stable colonizers of the infant gut, as they were not detectable by week two. Finally, we used metagenomic sequencing to compare microbial strains in maternal vaginal and rectal samples and samples from their infants; we found evidence for mother-to-child transmission of rectal rather than vaginal strains. These results challenge birth canal exposure as the dominant factor in infant gut microbiome establishment and implicate colonization efficiency rather than exposure as a dictating factor of the newborn gut microbiome composition.

Summary Mode of delivery strongly influences the early infant gut microbiome. Children born by cesarean section (C-section) lack Bacteroides species until 6–18 months of age. One hypothesis is that these differences stem from lack of exposure to the maternal vaginal microbiome. Here, we re-evaluate this hypothesis by comparing the microbial profiles of 75 infants born vaginally or by planned versus emergent C-section. Multiple children born by C-section have a high abundance of Bacteroides in their first few days of life, but at 2 weeks, both C-section groups lack Bacteroides (primarily according to 16S sequencing), despite their difference in exposure to the birth canal. Finally, a comparison of microbial strain profiles between infants and maternal vaginal or rectal samples finds evidence for mother-to-child transmission of rectal rather than vaginal strains. These results suggest differences in colonization stability as an important factor in infant gut microbiome composition rather than birth canal exposure., Graphical Abstract, Highlights Week 1 gut microbiota does not differ between infants born vaginally versus C-section Week 2 gut microbiota of C-section infants lacks Bacteroides Microbiota of infants born by C-section after labor resembles scheduled C-section Bacterial strains in infants match maternal rectal rather than vaginal strains, Mitchell et al. compare early-life infant gut microbiota by delivery mode, suggesting early colonization by Bacteroides regardless of delivery mode, but loss of Bacteroides by 2 weeks in C-section-delivered infants, whether or not exposed to the vagina in labor. Infant strains matched maternal rectal rather vaginal strains.

A typo in the 'Reviewer information' section of this Letter was corrected online.

Bacterial community acquisition in the infant gut impacts immune education and disease susceptibility. We compared bacterial strains across and within families in a prospective birth cohort of 44 infants and their mothers, sampled longitudinally in the first months of each child's life. We identified mother-to-child bacterial transmission events and describe the incidence of family-specific antibiotic resistance genes. We observed two inheritance patterns across multiple species, where often the mother's dominant strain is transmitted to the child, but occasionally her secondary strains colonize the infant gut. In families where the secondary strain of B. uniformis was inherited, a starch utilization gene cluster that was absent in the mother's dominant strain was identified in the child, suggesting the selective advantage of a mother's secondary strain in the infant gut. Our findings reveal mother-to-child bacterial transmission events at high resolution and give insights into early colonization of the infant gut. Using longitudinal metagenomic sequencing from 44 mother/child pairs, Yassour et al. characterized mother-to-child strain transmission patterns. While mothers' dominant strains were often inherited, nondominant secondary strain transmissions were also observed. Microbial functional analysis reveals that inherited maternal secondary strains may have a selective advantage to colonize infant guts., National Institutes of Health (Grant 1DP3DK094338–01)

The human genome contains thousands of long non-coding RNAs1, but specific biological functions and biochemical mechanisms have been discovered for only about a dozen2-7. A specific long non-coding RNA-non-coding RNA activated by DNA damage (NORAD)-has recently been shown to be required for maintaining genomic stability8, but its molecular mechanism is unknown. Here we combine RNA antisense purification and quantitative mass spectrometry to identify proteins that directly interact with NORAD in living cells. We show that NORAD interacts with proteins involved in DNA replication and repair in steady-state cells and localizes to the nucleus upon stimulation with replication stress or DNA damage. In particular, NORAD interacts with RBMX, a component of the DNA-damage response, and contains the strongest RBMX-binding site in the transcriptome. We demonstrate that NORAD controls the ability of RBMX to assemble a ribonucleoprotein complex-which we term NORAD-activated ribonucleoprotein complex 1 (NARC1)-that contains the known suppressors of genomic instability topoisomerase I (TOP1), ALYREF and the PRPF19-CDC5L complex. Cells depleted for NORAD or RBMX display an increased frequency of chromosome segregation defects, reduced replication-fork velocity and altered cell-cycle progression-which represent phenotypes that are mechanistically linked to TOP1 and PRPF19-CDC5L function. Expression of NORAD in trans can rescue defects caused by NORAD depletion, but rescue is significantly impaired when the RBMX-binding site in NORAD is deleted. Our results demonstrate that the interaction between NORAD and RBMX is important for NORAD function, and that NORAD is required for the assembly of the previously unknown topoisomerase complex NARC1, which contributes to maintaining genomic stability. In addition, we uncover a previously unknown function for long non-coding RNAs in modulating the ability of an RNA-binding protein to assemble a higher-order ribonucleoprotein complex.

SUMMARYWe previously piloted the concept of a Connectivity Map (CMap), whereby genes, drugs and disease states are connected by virtue of common gene-expression signatures. Here, we report more than a 1,000-fold scale-up of the CMap as part of the NIH LINCS Consortium, made possible by a new, low-cost, high throughput reduced representation expression profiling method that we term L1000. We show that L1000 is highly reproducible, comparable to RNA sequencing, and suitable for computational inference of the expression levels of 81% of non-measured transcripts. We further show that the expanded CMap can be used to discover mechanism of action of small molecules, functionally annotate genetic variants of disease genes, and inform clinical trials. The 1.3 million L1000 profiles described here, as well as tools for their analysis, are available at https://clue.io.HIGHLIGHTSA new gene expression profiling method, L1000, dramatically lowers costThe Connectivity Map database now includes 1.3 million publicly accessible L1000 perturbational profilesThis expanded Connectivity Map facilitates discovery of small molecule mechanism of action and functional annotation of genetic variantsThe work establishes feasibility and utility of a truly comprehensive Connectivity Map

We previously piloted the concept of a Connectivity Map (CMap), whereby genes, drugs, and disease states are connected by virtue of common gene-expression signatures. Here, we report more than a 1,000-fold scale-up of the CMap as part of the NIH LINCS Consortium, made possible by a new, low-cost, high-throughput reduced representation expression profiling method that we term L1000. We show that L1000 is highly reproducible, comparable to RNA sequencing, and suitable for computational inference of the expression levels of 81% of non-measured transcripts. We further show that the expanded CMapcan be used to discover mechanism of action of small molecules, functionally annotate genetic variants ofdisease genes, and inform clinical trials. The 1.3 million L1000 profiles described here, as well as tools for their analysis, are available at https://clue.io.

Cancer genome characterization efforts now provide an initial view of the somatic alterations in primary tumors. However, most point mutations occur at low frequency, and the function of these alleles remains undefined. We have developed a scalable systematic approach to interrogate the function of cancer-associated gene variants. We subjected 474 mutant alleles curated from 5,338 tumors to pooled in vivo tumor formation assays and gene expression profiling. We identified 12 transforming alleles, including two in genes (PIK3CB, POT1) that have not been shown to be tumorigenic. One rare KRAS allele, D33E, displayed tumorigenicity and constitutive activation of known RAS effector pathways. By comparing gene expression changes induced upon expression of wild-type and mutant alleles, we inferred the activity of specific alleles. Because alleles found to be mutated only once in 5,338 tumors rendered cells tumorigenic, these observations underscore the value of integrating genomic information with functional studies. Significance: Experimentally inferring the functional status of cancer-associated mutations facilitates the interpretation of genomic information in cancer. Pooled in vivo screen and gene expression profiling identified functional variants and demonstrated that expression of rare variants induced tumorigenesis. Variant phenotyping through functional studies will facilitate defining key somatic events in cancer. Cancer Discov; 6(7); 714–26. ©2016 AACR. See related commentary by Cho and Collisson, p. 694. This article is highlighted in the In This Issue feature, p. 681

Recent genome sequencing efforts have identified millions of somatic mutations in cancer. However, the functional impact of most variants is poorly understood. Here we characterize 194 somatic mutations identified in primary lung adenocarcinomas. We present an expression-based variant-impact phenotyping (eVIP) method that uses gene expression changes to distinguish impactful from neutral somatic mutations. eVIP identified 69% of mutations analyzed as impactful and 31% as functionally neutral. A subset of the impactful mutations induces xenograft tumor formation in mice and/or confers resistance to cellular EGFR inhibition. Among these impactful variants are rare somatic, clinically actionable variants including EGFR S645C, ARAF S214C and S214F, ERBB2 S418T, and multiple BRAF variants, demonstrating that rare mutations can be functionally important in cancer.

Recent cancer genome sequencing and analysis has identified millions of somatic mutations in cancer. However, the functional impact of most variants is poorly understood, limiting the use of this genetic knowledge for clinical decision-making. Here we describe a new high-throughput approach, expression-based variant impact phenotyping (eVIP), which uses gene expression changes to infer somatic mutation impact. We generated a lentiviral expression library representing 53 genes and 194 somatic mutations identified in primary lung adenocarcinomas. Next, we introduced this library into A549 lung adenocarcinoma cells and 96 hours later performed gene expression profiling using Luminex-based L1000 profiling. We built a computational pipeline, eVIP, to compare mutant and wild-type expression signatures to infer whether variants were gain-of-function, change-of-function, loss-of-function, or neutral. Overall, eVIP identified 69% of mutations as impactful whereas 31% appeared functionally neutral. A very high rate, 92%, of missense mutations in the KEAP1 and STK11 tumor suppressor genes were found to inactivate or diminish protein function. As a complementary approach, we assessed which mutations are epistatic to EGFR or capable of initiating xenograft tumor formation in vivo. A subset of the impactful mutations identified by eVIP could induce xenograft tumor formation in mice and/or confer resistance to cellular EGFR inhibition. Among these mutations were 20 rare or non-canonical somatic variants in clinically-actionable or -relevant oncogenes including EGFR S645C, ARAF S214C and S214F, ERBB2 S418T, and PIK3CA E600K. eVIP can, in principle, characterize any genetic variant, independent of prior knowledge of gene function. Further application of eVIP should significantly advance the pace of functional characterization of mutations identified from genome sequencing. Citation Format: Alice H. Berger, Angela N. Brooks, Xiaoyun Wu, Yashaswi Shrestha, Candace Chouinard, Federica Piccioni, Mukta Bagul, Atanas Kamburov, Marcin Imielinski, Larson Hogstrom, Cong Zhu, Xiaoping Yang, Sasha Pantel, Ryo Sakai, Nathan Kaplan, David Root, Rajiv Narayan, Ted Natoli, David Lahr, Itay Tirosh, Pablo Tamayo, Gad Getz, Bang Wong, John Doench, Aravind Subramanian, Todd R. Golub, Matthew Meyerson, Jesse S. Boehm. High-throughput phenotyping of lung cancer somatic mutations. [abstract]. In: Proceedings of the 107th Annual Meeting of the American Association for Cancer Research; 2016 Apr 16-20; New Orleans, LA. Philadelphia (PA): AACR; Cancer Res 2016;76(14 Suppl):Abstract nr 4368.

Recently, the decline in the cost of genome sequencing has led to the rapid identification of thousands of cancer-associated somatic mutations. However, progress in characterization of these genetic events has lagged significantly behind. Understanding mutation function is critical not only for research purposes but also for determining targeted treatment strategies based on individual tumor genetic profiles, yet determination of mutation impact remains a significant bottleneck. Here we describe a high-throughput approach to classify somatic mutations that is robust, scalable, and requires no prior information of gene function. We generated a lentiviral cDNA expression library of ~550 mutated and wild-type alleles of genes mutated in lung adenocarcinoma and introduced these alleles into four human lung cell lines. 96 hours post-infection, gene expression profiles were generated using Luminex-based L1000 profiling. In total, more than 2000 gene expression signatures were generated. We discovered that gain-of-function mutants induce expression signatures with a greater signal strength or different identity than the corresponding wild-type gene signature. In contrast, loss-of-function mutants could be identified by their incapability to induce strong signatures. Based on these features of signature strength and signature identity, we developed a decision-tree approach to classify mutations as either dominant, loss-of-function, or likely inert. An orthogonal functional approach, an EGFR inhibitor resistance screen, was used as validation. The gene expression approach correctly classified known gain-of-function mutations in KRAS (13/13), EGFR (6/7), and ARAF (2/2) and identified dozens of never-characterized gain-of-function and loss-of-function missense mutations. In addition to rare, dominant mutations in clinically-actionable oncogenes such as PIK3CA and AKT1, we identified unexpected dominant mutations in the transcription factor MAX and the phosphatase subunit PPP2R1A, among others. We also observed a substantial enrichment of loss-of-function mutations in tumor suppressor genes such as STK11, KEAP1, FBXW7, and CASP8 as well as in genes not previously connected to lung adenocarcinoma, including GPR137B and MAPK7. Most genes assayed also harbored variants that are likely inert, further underscoring the importance of characterizing individual variant alleles. The method developed here can, in principle, characterize any genetic variant, independent of prior knowledge of gene function, and should significantly advance the pace of functional characterization of mutations identified from genome sequencing. Citation Format: Alice Berger, Angela Brooks, Xiaoyun Wu, Larson Hogstrom, Itay Tirosh, Federica Piccioni, Mukta Bagul, Cong Zhu, Yashaswi Shretha, David Root, Pablo Tamayo, Ryo Sakai, Bang Wong, Aravind Subramanian, Todd Golub, Matthew Meyerson, Jesse Boehm. High-throughput gene expression profiling as a generalizable assay for determination of mutation impact on gene function. [abstract]. In: Proceedings of the AACR Special Conference on Computational and Systems Biology of Cancer; Feb 8-11 2015; San Francisco, CA. Philadelphia (PA): AACR; Cancer Res 2015;75(22 Suppl 2):Abstract nr PR04.

Recently, the decline in the cost of genome sequencing has led to the rapid identification of thousands of cancer-associated somatic mutations. However, progress in characterization of these genetic events has lagged significantly behind. Understanding mutation function is critical not only for research purposes but also for determining targeted treatment strategies based on individual tumor genetic profiles, yet determination of mutation impact remains a significant bottleneck. Here we describe a high-throughput approach to classify somatic mutations that is robust, scalable, and requires no prior information of gene function. We generated a lentiviral cDNA expression library of ~550 mutated and wild-type alleles of genes mutated in lung adenocarcinoma and introduced these alleles into four human lung cell lines. 96 hours post-infection, gene expression profiles were generated using Luminex-based L1000 profiling. In total, more than 2000 gene expression signatures were generated. We discovered that gain-of-function mutants induce expression signatures with a greater signal strength or different identity than the corresponding wild-type gene signature. In contrast, loss-of-function mutants could be identified by their incapability to induce strong signatures. Based on these features of signature strength and signature identity, we developed a decision-tree approach to classify mutations as either dominant, loss-of-function, or likely inert. An orthogonal functional approach, an EGFR inhibitor resistance screen, was used as validation. The gene expression approach correctly classified known gain-of-function mutations in KRAS (13/13), EGFR (6/7), and ARAF (2/2) and identified dozens of never-characterized gain-of-function and loss-of-function missense mutations. In addition to rare, dominant mutations in clinically-actionable oncogenes such as PIK3CA and AKT1, we identified unexpected dominant mutations in the transcription factor MAX and the phosphatase subunit PPP2R1A, among others. We also observed a substantial enrichment of loss-of-function mutations in tumor suppressor genes such as STK11, KEAP1, FBXW7, and CASP8 as well as in genes not previously connected to lung adenocarcinoma, including GPR137B and MAPK7. Most genes assayed also harbored variants that are likely inert, further underscoring the importance of characterizing individual variant alleles. The method developed here can, in principle, characterize any genetic variant, independent of prior knowledge of gene function, and should significantly advance the pace of functional characterization of mutations identified from genome sequencing. Citation Format: Alice Berger, Angela Brooks, Xiaoyun Wu, Larson Hogstrom, Itay Tirosh, Federica Piccioni, Mukta Bagul, Cong Zhu, Yashaswi Shretha, David Root, Pablo Tamayo, Ryo Sakai, Bang Wong, Aravind Subramanian, Todd Golub, Matthew Meyerson, Jesse Boehm. High-throughput gene expression profiling as a generalizable assay for determination of mutation impact on gene function. [abstract]. In: Proceedings of the AACR Special Conference on Translation of the Cancer Genome; Feb 7-9, 2015; San Francisco, CA. Philadelphia (PA): AACR; Cancer Res 2015;75(22 Suppl 1):Abstract nr PR12.