Rare driver mutations in head and neck squamous cell carcinomas converge on NOTCH signaling

See allHide authors and affiliations

Science  13 Mar 2020:
Vol. 367, Issue 6483, pp. 1264-1269
DOI: 10.1126/science.aax0902

Cancer drivers converge on NOTCH

Cancer genome–sequencing projects have emphasized the handful of genes mutated at high frequency in patients. Less attention has been directed to the hundreds of genes mutated in only a few patients—the so-called “long tail” mutations. Although rare, these mutations may nonetheless inform patient care. Loganathan et al. developed a reverse genetic CRISPR screen that allowed them to functionally assess in mice nearly 500 long tail gene mutations that occur in human head and neck squamous cell carcinoma (HNSCC). They identified 15 tumor-suppressor genes with activities that converged on the NOTCH signaling pathway. Given that NOTCH itself is mutated at high frequency in HNSCC, these results suggest that the growth of these tumors is largely driven by NOTCH inactivation.

Science, this issue p. 1264


In most human cancers, only a few genes are mutated at high frequencies; most are mutated at low frequencies. The functional consequences of these recurrent but infrequent “long tail” mutations are often unknown. We focused on 484 long tail genes in head and neck squamous cell carcinoma (HNSCC) and used in vivo CRISPR to screen for genes that, upon mutation, trigger tumor development in mice. Of the 15 tumor-suppressor genes identified, ADAM10 and AJUBA suppressed HNSCC in a haploinsufficient manner by promoting NOTCH receptor signaling. ADAM10 and AJUBA mutations or monoallelic loss occur in 28% of human HNSCC cases and are mutually exclusive with NOTCH receptor mutations. Our results show that oncogenic mutations in 67% of human HNSCC cases converge onto the NOTCH signaling pathway, making NOTCH inactivation a hallmark of HNSCC.

Head and neck squamous cell carcinoma (HNSCC) is the sixth most common human cancer, and the 5-year survival rate is <50% (1). HNSCC arises in the mucosal lining of the upper aerodigestive tract and is tightly linked to tobacco use, alcohol consumption, and human papillomavirus (HPV) infection. The most common genetic alterations in HNSCC affect p53 (71%), FAT1 (23%), CDKN2A (22%), PIK3CA (18%), NOTCH1 (17%), and HRAS (6%), followed by a “long tail” of hundreds of individually rare mutations, most of which lack biological or clinical validation (2, 3) (fig. S1, A and B).

The long tails of recurrent but rare mutations remain enigmatic in cancers (4, 5). It is difficult to reconcile their apparent positive selection with their low frequency and immense diversity. The long tail might reflect the stochasticity of cancer evolution, where each mutation confers only a small fitness advantage, but several such low-penetrant mutations might cooperate and substantially promote tumor progression (6, 7). Some long tail mutations may be highly penetrant but simply affect genes that are rarely mutated. Several mutations may also converge on the same cellular signaling pathway, which could explain some of the diversity and low frequency, highlighting the importance of a pathway-centric view of the cancer landscape (8, 9). Given that long tail driver mutations can be the genetic basis for exceptional responses to therapy (10, 11), it is crucial to identify clinically relevant mutations and define their oncogenic mechanisms.

To functionally assess HNSCC long tail genes, we developed a CRISPR screen to identify genes that, upon mutation, predispose mice to HNSCC development. We constructed lentiviruses that coexpress a single-guide RNA (sgRNA) and Cre recombinase and used ultrasound-guided in utero microinjections to deliver these lentiviruses to the single-layered surface ectoderm of live mouse embryos (12) (Fig. 1A). The surface ectoderm generates several structures, including the skin epithelium and oral mucosa. We used multicolor Lox-Stop-Lox-Confetti mice, which, upon Cre-mediated excision of the Lox-Stop-Lox (LSL) cassette, stochastically switch on one of four fluorescent proteins, and determined the viral titer required to generate thousands of discrete clones within the oral cavity and epidermis (Fig. 1B).

Fig. 1 Direct in vivo CRISPR screen for HNSCC tumor suppressors.

(A) Experimental workflow for pooled in vivo CRISPR screen. A lentiviral sgRNA library targeting mouse homologs of human HNSCC long tail genes is introduced into the surface epithelium of mice, and tumors are analyzed by next-generation sequencing (NGS). (B) Representative images of whole body, oral cavity, tongue, and palate of newborn Cre-reporter LSL-Confetti mice transduced with Cre lentivirus. Scale bars, 500 μm. (C) Tumor-free survival of mice transduced with an sgRNA library targeting putative HNSCC genes or with a control sgRNA library (n > 20 per group; P < 0.0001, log-rank test). (D) Representative images of Pik3caH1047R;Cas9 mice transduced with the HNSCC sgRNA library showing multiple, progressively growing tumors in the oral cavity and tongue. Scale bars, 500 μm. (E) Pie chart showing tumor-suppressor genes with enriched sgRNAs in tumor DNA obtained from the four different HNSCC mouse models.

Next, we tested the efficiency of CRISPR/Cas9-mediated in vivo mutagenesis (fig. S2A). We found that LSL-Cas9-GFP;LSL-tdTomato mice transduced with a Cre lentivirus encoding a scrambled sgRNA (LV-sgScr-Cre) coexpressed green fluorescent protein (GFP) and tdTomato. By contrast, mice transduced with sgRNA-Cre lentivirus targeting GFP (LV-sgGFP-Cre) displayed tdTomato+ cells lacking GFP, demonstrating a knockout efficiency of 85 ± 5% (fig. S2B). Next, we targeted the heme biosynthesis gene Urod, the loss of which leads to the accumulation of unprocessed, fluorescent porphyrins (13), which resulted in bright red fluorescence in the oral mucosa and skin (fig. S2C), demonstrating efficient targeting of an endogenous gene.

To determine whether this approach can reveal genetic interactions, we recapitulated cooperation between oncogenic phosphatidylinositol 3-kinase (PI3K)/Akt signaling and loss of tumor suppressors such as transforming growth factor-β receptor II (TgfβrII) or p53 (14). Cas9-mediated ablation of TgfβrII or p53 in conditional Pik3caH1047R mice (LSL-Pik3caH1047R;LSL-Cas9) triggered rapid formation of HNSCC tumors, whereas littermates transduced with scrambled control sgRNAs remained asymptomatic (fig. S2, D and E). Because of the transduction method used, these mice simultaneously developed HNSCC as well as cutaneous SCC (cSCC). The latter tumors are genetically, histologically, and pathologically related to HNSCC (3).

Next, we generated a lentiviral sgRNA library to mutate the mouse homologs of 484 recurrent but infrequent human HNSCC long tail genes and control libraries containing 418 nontargeting sgRNAs or 414 randomly picked genes (four sgRNAs per gene) (figs. S1 and S2F and tables S1 and S2) (2). To identify tumor-suppressor genes that cooperate with known HNSCC oncogenic driver mutations, we transduced these sgRNA libraries into LSL-Cas9 mice that harbor (i) a conditional Pik3ca oncogene (LSL-Pik3caH1047R); (ii) a conditional HRas oncogene (LSL-HRasG12V); (iii) a conditional, dominant-negative p53 mutation (LSL-p53R270H); or (iv) epithelium-specific expression of the HPV16-E6/E7 oncogenes (K14-HPV16). Within a year, none of the mice transduced with the control libraries developed tumors, highlighting that Pik3caH1047R, HRasG12V, p53R270H, and HPV16-E6/E7 on their own are insufficient to initiate SCC. By contrast, all mice transduced with the long tail sgRNA library developed multiple HNSCC and cSCC tumors within weeks (Fig. 1, C and D, and fig. S3, A and B). Pik3caH1047R;Cas9 mice transduced with an sgRNA library targeting 215 breast cancer long tail genes did not develop tumors over a 4-month observation period (fig. S2G and table S2), indicating the existence of strong, HNSCC-specific long tail tumor-suppressor genes.

To identify these tumor-suppressor genes, we determined sgRNA representation in 205 mouse tumors. Most tumors showed strong enrichment for a single sgRNA compared with tumor-adjacent, phenotypically normal epithelium (fig. S3C). Fifteen genes showed enrichment of two or more independent sgRNAs in multiple tumors, with Adam10, Ripk4, and Ajuba being the most prevalent hits, followed by Notch2 and Notch3 (Fig. 1E, fig. S3D, and table S3). Although most genes scored in all oncogenic backgrounds, Ripk4, for example, did not surface in HPV-mutant mice. For validation of our top hits, we injected Pik3caH1047R;Cas9 and HRasG12V;Cas9 mouse embryos with Adam10, Ajuba, or Ripk4 sgRNAs that were not present in the initial library. These mice rapidly developed invasive SCC tumors but were more prone to developing HNSCC than cSCC compared with mice with p53 or TgfβrII loss (fig. S4, A to E). Sequencing of the sgRNA target sites in tumor DNA confirmed frame-shift mutations in all tested tumors (fig. S5). Inducible activation of the Pik3caH1047R oncogene and Cas9-mediated mutagenesis of Adam10 or Ripk4 in adult mice also led to rapid HNSCC development (fig. S6, C and D). Loss of Adam10 or Ripk4 triggered tumor development even in wild-type mice with very long latency (fig. S6, A and B). Thus, Adam10, Ripk4, and Ajuba are strong HNSCC suppressors.

The membrane-anchored metalloproteinase ADAM10 controls diverse cellular processes through regulated intramembrane proteolysis of numerous proteins (1517). To confirm its role in suppressing HNSCC, we generated conditional Adam10fl/fl;Pik3caH1047R compound mutant mice and transduced embryos with lentiviral Cre. Homozygous Adam10 deletion rapidly induced HNSCC tumors, recapitulating our CRISPR/Cas9 findings (Fig. 2A and fig. S7, A to C). Even heterozygous Adam10fl/+;Pik3caH1047R mice developed HNSCC, although with longer latency (Fig. 2A). To assess whether tumor development was due to Adam10 loss of heterozygosity, we used fluorescence-activated cell sorting to isolate tumor cells from Adam10-homozygous and -heterozygous Pik3caH1047R tumors and from control sgp53;Pik3caH1047R tumors. Transcriptional profiling and Western blot analysis revealed wild-type Adam10 expression in Adam10-heterozygous tumors, albeit at a reduced level compared with control tumors (Fig. 2B and fig. S7D), indicating that Adam10 functions as a haploinsufficient tumor suppressor in HNSCC.

Fig. 2 Adam10 suppresses HNSCC development by regulating NOTCH.

(A) Tumor-free survival for Adam10fl/fl, Adam10fl/+, and Adam10+/+;Pik3caH1047R mice transduced with Cre lentivirus. (B) Adam10 mRNA expression levels in tumors of the indicated genotype. (C) GSEA reveals down-regulation of the NOTCH pathway in Adam10fl/fl;Pik3caH1047R tumors. (D) RT-PCR analysis of NOTCH target gene expression in keratinocytes isolated from postnatal day 4 pups of the indicated genotype. Data are shown as mean ± SEM (n = 3). *P < 0.05. (E) Tumor-free survival for Pik3caH1047R mice transduced with lentiviral Cre-sgRNA targeting the indicated NOTCH pathway components. (F) NOTCHIC expression rescues loss of Adam10. Tumor-free survival of Adam10fl/fl;Pik3caH1047R mice transduced with Cre lentiviral constructs that allow for doxycycline-inducible expression of NOTCHIC. Inducible expression of tdTomato served as a control.

Gene set enrichment analysis (GSEA) of these tumor cell transcriptomes revealed differentially expressed gene sets specifically associated with “Hallmarks of G2M checkpoint” and “E2F targets” as well as “NOTCH signaling” (Fig. 2C and fig. S8A). NOTCH receptors are transmembrane proteins that are proteolytically cleaved upon binding to JAG1/2 or DLL1/3/4 ligands. The cleaved NOTCH intracellular domain (NOTCHIC) enters the nucleus, binds the DNA-binding protein RBPJ, and regulates gene expression. Adam10-knockout tumor cells showed reduced expression of NOTCH target genes such as Hes1 and Heyl (fig. S8E), consistent with ligand-dependent activation of NOTCH by ADAM10-mediated cleavage (14, 17). Down-regulated NOTCH pathway activity was also seen in primary keratinocytes isolated from Adam10-deficient Pik3caH1047R mice 4 days after birth, before any overt phenotypic changes could be observed (Fig. 2D and fig. S8, B to D), indicating that ADAM10 directly activates NOTCH.

Given the prominent role of NOTCH signaling in suppressing mouse and human HNSCC development (18), we tested whether genetic ablation of canonical NOTCH signaling components in Pik3caH1047R mice phenocopies Adam10 loss. Indeed, sgRNA/Cas9-mediated mutagenesis of Notch1, Jag1, and Jag2 or the obligate transcriptional NOTCH effector Rbpj, as well as conditional deletion of Notch1 and Notch2, caused HNSCC development. Targeting the NOTCH ligand Dll1 or Adam17, a metalloproteinase implicated in ligand-independent NOTCH activation that is required for epidermal barrier function (19, 20), did not trigger tumor development (Fig. 2E; fig. S9, A and B; and fig. S5D). Finally, we tested whether restoring NOTCH signaling in Adam10-deficient mice would block tumor development. Inducible expression of ectopic NOTCHIC before tumor appearance or in established tumors resulted in effective tumor suppression and delayed HNSCC development (Fig. 2F and fig. S9, C to F). Thus, loss of Adam10 promotes HNSCC tumorigenesis by impairing NOTCH signaling.

Our second hit, AJUBA, is mutated in 7% of HNSCC, 18% of cutaneous SCC, and 2 to 7% of esophageal SCC cases, but not in other human cancers (3, 21). AJUBA is a scaffold protein and has been shown to regulate the Hippo (22), Wnt (23), and Aurora-A signaling pathways (24) and cell adhesion molecules (25), but its role in SCC pathogenesis is unclear (1). Ajuba-null mice were embryonically lethal. However, we found that loss of one Ajuba allele triggered HNSCC development in Pik3caH1047R, HRasG12V, or K14-HPV16 mice but retained expression of the wild-type Ajuba allele (Fig. 3A and fig. S10, A to D). These data suggest that Ajuba also functions as a haploinsufficient tumor suppressor.

Fig. 3 Ajuba is a new NOTCH regulator.

(A) Tumor-free survival of Ajuba+/− and Ajuba+/+ mice in Pik3caH1047R, HRasG12V, or K14-HPV16 mouse backgrounds. (B) RT-PCR results showing effects of CRISPR/Cas9-mediated ablation of Ajuba on EDTA activation of the canonical NOTCH targets Hes1 and Hey1 in primary mouse keratinocytes. Data are shown as means ± SEM (n = 3). *P < 0.05. (C) Co-IP of HA-Ajuba with endogenous Numb, full-length Notch1 (300 kDa), and NOTCH1IC (120 kDa) in primary keratinocytes, which changes upon NOTCH pathway activation (30 min EDTA; 15 min recovery). (D) Co-IP of endogenous Notch1 with Numb and Itch1 in wild-type (WT), Ajuba-knockout (KO), and Ajuba-overexpressing (KO+OE) primary keratinocytes. (E) Genetic ablation of Ajuba impairs nuclear accumulation of NOTCHIC. Immunofluorescence of Ajuba WT and KO as well as Ajuba-overexpressing (OE) WT and KO primary keratinocytes is shown. Cells were treated with EDTA (30 min) to stimulate NOTCH receptor activation, allowed to recover for 15 min (EDTA rec), and stained for NOTCH1. Scale bar, 50 μm.

Transcriptional profiling and GSEA of keratinocytes isolated from sgAjuba;Pik3caH1047R mice 4 days after birth showed no difference in Hippo or Wnt pathway activation but significant down-regulation of “NOTCH signaling” (fig. S11, A to D). To corroborate this finding, we investigated the effect of Ajuba loss on NOTCH pathway activation in primary keratinocytes. We used either the canonical NOTCH ligand JAG1 or EDTA treatment, which forces dissociation of the heterodimeric NOTCH receptor, thereby allowing release of NOTCHIC. Loss of Ajuba impaired ligand- and EDTA-induced up-regulation of NOTCH target genes such as Hes1 and Hey1 (Fig. 3B and fig. S11E). To test whether these changes are directly mediated by NOTCHIC, we performed Rbpj CUT&RUN (cleavage under targets and release using nuclease) sequencing (26). Most Rbpj-binding sites in primary keratinocytes are in enhancers and promoters, where Rbpj is dynamically recruited upon NOTCH activation (fig. S12, A and B) (27). NOTCH pathway inhibition by pharmacological ADAM10 inhibition or loss of Ajuba markedly impaired Rbpj recruitment to and the expression of canonical NOTCH target genes such as Hes1, Heyl, and Nrarp and other genes such as Id1, Egfr, Tgfβ1, Krt10, and JunB (fig. S12, A to D, and table S4). Ten of the 15 primary screening hits are themselves NOTCH target genes, including Ajuba, Adam10, Ripk4, Notch1/3, Jag1/2, and Irf6, indicating the existence of a feed-forward loop to enhance NOTCH signaling and thus tumor suppression (figs. S12C and S17 and table S3).

To investigate Ajuba’s function, we identified vicinal proteins by proximity-dependent biotinylation coupled to mass spectrometry (BioID) in human 293 cells. AJUBA BioID enriched for Hippo components (LATS1/2, AMOT, and PTPN14) and several new proximity interactors including NOTCH1, NOTCH2, and NUMB (fig. S13A). NUMB recruits the E3 ubiquitin ligase ITCH to NOTCH, thereby facilitating NOTCH ubiquitination and degradation (28). In accordance with the BioID results, FLAG-tagged Ajuba coimmunoprecipitated (co-IP) endogenous Lats1, Notch1, Notch2, and Numb from primary mouse keratinocytes (fig. S13B and Fig. 3C). The reciprocal co-IP also showed that endogenous Notch1 and Notch2 interacted with endogenous Ajuba (fig. S13C).

Next, we explored how Ajuba regulates NOTCH signaling. NOTCH activation increased the interaction between Ajuba and Numb but reduced binding between AJUBA and cleaved NOTCH1IC (Fig. 3C), implicating a dynamically regulated process. To test functional relevance, we genetically ablated Ajuba in primary keratinocytes and analyzed Notch-Numb-Itch complex formation. Loss of Ajuba resulted in (i) increased association between Numb and Notch1, (ii) aberrant recruitment of Itch to Notch1, and (iii) concomitant increased Notch1 ubiquitination, all of which was reversed by reexpressing Ajuba (Fig. 3D and fig. S13, D and E). Consistent with these findings, nuclear accumulation of NOTCH1IC and NOTCH2IC after receptor activation was markedly reduced in Ajuba-knockout cells, which was reversed upon reexpressing Ajuba (Fig. 3E and fig. S14, A to D). Last, we tested whether Numb, which is amplified in ~37% of HNSCC patients, could promote HNSCC development. Indeed, overexpression of Numb in Pik3caH1047R mice triggered rapid HNSCC and cSCC development (fig. S14E). These data suggest that Ajuba binds and sequesters Numb, thereby promoting NOTCH signaling and HNSCC tumor suppression.

Our third hit, Ripk4 (Receptor Interacting Protein Kinase 4), was reported to be a NOTCH target gene (29), further implicating NOTCH signaling in HNSCC. Indeed, NOTCH pathway activation led to increased Rbpj CUT&RUN footprints in promoter and enhancer regions of Ripk4 and transcriptional up-regulation of Ripk4 (figs. S11E and S12, B and C). Conditional Ripk4fl/fl;Pik3caH1047R but not heterozygous Ripk4fl/+;Pik3caH1047R compound mutant mice formed tumors upon lentiviral Cre transduction, corroborating our CRISPR results (fig. S15A). Ripk4 promotes the differentiation of oral and epidermal keratinocytes by phosphorylating and activating the Irf6 (Interferon Regulatory Factor 6) transcription factor (30). Irf6 also scored in our screen and constitutes another direct NOTCH target gene (table S3 and fig. S12C). These data indicate that the Ripk4–Irf6 axis is an important downstream effector of NOTCH.

To identify additional downstream NOTCH targets that suppress HNSCC, we generated a sgRNA library targeting 80 genes down-regulated upon Adam10 ablation (fig. S15, B and C, and table S6). Transduced Pik3caH1047R;Cas9 mice developed tumors enriched for sgRNAs targeting the integrin subunit Itgb5, the Hippo component Angiomotin-Like 2 (Amotl2), Endonuclease Domain Containing 1 (Endod1), and Sushi Domain Containing 2 (Susd2) (Fig. 4, A and B, and fig. S15D). Independent sgRNAs targeting Itgb5 or Amotl2 confirmed our findings and induced highly aggressive SCC in the Pik3caH1047R mice (Fig. 4C and figs. S5E and S15E). CUT&RUN sequencing and real-time polymerase chain reaction (RT-PCR) confirmed that Itgb5 or Amotl2 are direct NOTCH target genes (Fig. 2D and fig. S12C) (27).

Fig. 4 NOTCH target genes and NOTCH pathway mutations in human tumors.

(A) Tumor-free survival for Pik3caH1047R mice transduced with a lentiviral Cre-sgRNA library targeting genes down-regulated in Adam10-deficient tumors (P < 0.0001, log-rank test). (B) Pie chart showing top-scoring sgRNAs enriched in tumor DNA from (A). (C) Tumor-free survival for Pik3caH1047R;Cas9 mice transduced with sgRNAs targeting Itgb5 and Amotl2 or scrambled control sgRNAs (P < 0.0001, log-rank test). (D) Patterns of monoallelic minimal deleted regions (MDRs) in 504 HNSCC TCGA samples. (E) Expression of ADAM10 or AJUBA in tumors carrying ADAM10 or AJUBA mutations or copy number alterations, respectively. (F) Genetic alterations of NOTCH receptors 1 to 3, ADAM10, AJUBA, RIPK4, and NUMB in human HNSCC samples (n = 504) from TCGA presented as a cBioPortal OncoPrint displaying a trend toward mutual exclusivity among mutations.

To extend our findings from mouse to human cancers, we analyzed 504 HNSCC cases from The Cancer Genome Atlas (2). The ADAM10, AJUBA, and RIPK4 genes were mutated in 0.8%, 7.5%, and 3.0% of tumors, respectively, a frequency expected for long tail mutations. However, an additional 12.7% and 11.1% of human HNSCCs showed heterozygous ADAM10 and AJUBA loss, respectively (fig. S16A). Analysis of overlapping allelic loss segments showed that chromosome 14 exhibited focal, monoallelic losses encompassing AJUBA, whereas ADAM10 exhibited preferential monoallelic loss of broad chromosome segments (Fig. 4D). We also found allelic imbalance of heterozygous single nucleotide polymorphisms in the chromosomal region of AJUBA, further confirming heterozygous loss of AJUBA (fig. S16, B and C). Mutations and allelic copy number loss coincided with reduced expression of ADAM10 and AJUBA (Fig. 4E). In addition, mutations and allelic loss of ADAM10, AJUBA, or RIPK4 showed a trend toward mutual exclusivity with mutations in NOTCH1/2/3 receptors (Fig. 4F, fig. S16D, and table S7). Altogether, ~27% of HNSCC patients carry inactivating NOTCH1/2/3 receptor mutations, and an additional ~40% of HNSCC samples show inactivating alterations of ADAM10 or AJUBA or amplification of NUMB (Fig. 4F). Although mutual exclusivity is neither a necessary nor a sufficient condition for genes involved in the same pathway, our functional studies suggest that these alterations converge on inactivating NOTCH signaling in NOTCH receptor wild-type HNSCC patients (fig. S17). Thus, NOTCH is one of the most commonly dysregulated pathways in HNSCC.

As in the case of HNSCC, the identification of long tail genes that drive tumorigenesis in other cancer types may show convergence on specific pathways and reveal new regulators of those pathways and could inform on cancer biology and tumor evolution. This information may even provide translational opportunities and increase the number of patients who benefit from a pathway-specific treatment. In addition, our work indicates that many long tail genes function in a haploinsufficient manner. Because most algorithms used to delineate driver from passenger mutations are based on statistical enrichment of somatic point mutations, amino acid conservation, or homozygous deletions and amplification and never take heterozygous deletions into account, many haploinsufficient tumor suppressors might have been overlooked. This could be of special interest in the postgenomic era, in which most, if not all, mutations and copy number alterations have been identified, but their functional annotation lags far behind. In summary, this study shows the power of integrating cancer genomics with mouse modeling using in vivo CRISPR screens to uncover tumor-suppressive pathways in the long tail of cancer-associated mutations.

Supplementary Materials

Materials and Methods

Figs. S1 to S19

Tables S1 to S7

References (3151)

References and Notes

Acknowledgments: We thank all members of our laboratories for helpful comments; Y. Q. Lu, D. Dervovic, and G. Mbamalu for assistance; Z. Y. Lin for mass spectrometry assistance; C. Go and J. D. R. Knight for access to the cell-map resource; The Centre for Phenogenomics and Network Biology Collaborative Centre at LTRI; and D. Durocher, J. Wrana, L. Pelletier, R. Bremner, J. McGlade, and J. Woodgett for critically reading the manuscript. Funding: This work was supported by a project grant to D.S. from the Canadian Institute of Health Research (CIHR 365252) and the Krembil Foundation. D.S. is the recipient of a career development award from HFSP (CDA00080/2015). S.K.L. is the recipient of a Canadian Cancer Society fellowship (BC-F-16#31919). A.C.G. was supported by a Terry Fox Research Institute program grant. Author contributions: S.K.L. performed all experiments. K.S. performed Ajuba mouse experiments, immunohistochemistry, and, together with K.T., helped in immunofluorescence experiments. E.L. helped to prepare the viral library. R.T. performed quantitative RT-PCR and CUT&RUN experiments. R.H.O. helped with the random genes library. A.M. performed all bioinformatic analysis. B.R. and A.-C.G. performed and analyzed the mass spectrometry experiments. P.S.-T. helped with the design of the mass spectrometry experiments. R.Q. and T.J.P. performed bioinformatics analysis on human TCGA data. D.S. coordinated the project and, together with S.K.L., designed the experiments and wrote the manuscript. Competing interests: The authors declare no competing interests. Data and materials availability: All mass spectrometry data have been deposited in the MassIVE repository (MSV000083405 and PXD012600). Data for the 192 different BioID baits targeting the major subcellular compartments of a human cell used to control for AJUBA interactome specificity are available at All code and the manifests used to analyze the human HNSCC TCGA data are available under the R package SchramekLOH v1.0.0 ( All RNA-sequencing and CUT&RUN data are available at the Gene Expression Omnibus (GEO) (GSE140495, GSE140496, and GSE140497). All other data are available in the main text or the supplementary materials.

Stay Connected to Science

Navigate This Article