Mutations in the LKB1 (also known as STK11) tumor suppressor are the third most frequent genetic alteration in non–small cell lung cancer (NSCLC). LKB1 encodes a serine/threonine kinase that directly phosphorylates and activates 14 AMPK family kinases (“AMPKRs”). The function of many of the AMPKRs remains obscure, and which are most critical to the tumor-suppressive function of LKB1 remains unknown. Here, we combine CRISPR and genetic analysis of the AMPKR family in NSCLC cell lines and mouse models, revealing a surprising critical role for the SIK subfamily. Conditional genetic loss of Sik1 revealed increased tumor growth in mouse models of Kras-dependent lung cancer, which was further enhanced by loss of the related kinase Sik3. As most known substrates of the SIKs control transcription, gene-expression analysis was performed, revealing upregulation of AP1 and IL6 signaling in common between LKB1- and SIK1/3-deficient tumors. The SIK substrate CRTC2 was required for this effect, as well as for proliferation benefits from SIK loss.
The tumor suppressor LKB1/STK11 encodes a serine/threonine kinase frequently inactivated in NSCLC. LKB1 activates 14 downstream kinases in the AMPK family controlling growth and metabolism, although which kinases are critical for LKB1 tumor-suppressor function has remained an enigma. Here we unexpectedly found that two understudied kinases, SIK1 and SIK3, are critical targets in lung cancer.
This article is highlighted in the In This Issue feature, p. 1469
Twenty years ago, LKB1 (also known as STK11) was identified as the single gene responsible for the inherited cancer disorder Peutz–Jeghers syndrome (1) and soon after was discovered to be the third most frequently mutated gene in non–small cell lung cancer (NSCLC), the leading cause of cancer-related death in the United States and worldwide (2, 3). Mutations in LKB1 are found in 20% of patients with NSCLC and frequently coincide with activating Kras mutations, prompting intense study of genetically engineered mouse models of LKB1 loss coupled with Kras activation to understand how deregulation of these two genes drives tumor formation and growth, and response to therapy in lung cancer (4–9). Studies of human patients and genetically engineered mouse models have shown remarkable convergence in determining that Kras;LKB1 (“KL”) tumors show a unique tumor biology that is distinct and separable from other subtypes of Kras-driven NSCLC, such as tumors where Kras mutations are coincident with p53 loss (10). One defining feature of KL tumors from a therapeutic standpoint is that these tumors exhibit primary resistance to a number of targeted therapeutics in both preclinical models and human patients (4, 11), and very recent studies suggest they are also uniquely resistant to immune checkpoint inhibitors, thus making LKB1-mutant lung cancer a very common form of cancer for which treatment options are extremely limited (12–14).
LKB1 is a serine/threonine kinase that activates a family of 14 downstream kinases, including the AMP-activated protein kinase (AMPK), which collectively are known to control cell metabolism, cell polarity, and growth (15, 16). Despite advances in the understanding of what the targets and functions of some of the downstream effectors of these 14 kinases are in different tissues, it remains unknown which of the 14 LKB1-dependent kinases are most required for its tumor-suppressive activity in the lung. The best studied of these LKB1-dependent kinases is AMPK, which is a highly conserved master regulator of cell metabolism and growth that rewires metabolism and mTORC1 in response to low nutrient and energy conditions (16). AMPK was hypothesized to be a lung tumor suppressor because it can indirectly suppress mTORC1 oncogenic signaling through the phosphorylation of a well-characterized downstream tumor suppressor, TSC2, and furthermore restrains progrowth, protumorigenic pathways that include inhibiting fatty-acid synthesis needed for cell growth (8). However, several recent studies utilizing mouse models of AMPK loss have suggested that AMPK plays a context-dependent role in cancer, as in some studies AMPK loss cooperated with mutational driver alterations to promote tumor formation in B- and T-cell lymphoma models (17, 18), but in opposite fashion, loss of AMPK became detrimental to cancer cell survival in the context of several leukemia models (19–21). The requirement of AMPK for tumor-cell survival fits with a wide body of literature on AMPK being required for cell survival under metabolic stress conditions; however, whether AMPK is pro- or antitumorigenic in the lung or any genetically engineered solid epithelial mouse model of cancer was completely unknown. In this context, we recently examined whether AMPK would be the central target of LKB1 in lung cancer (22). Unexpectedly, we found that AMPK deletion in Kras-driven models of NSCLC did not mimic the effects of LKB1 loss, and instead suppressed tumorigenesis, being required for the outgrowth of Kras-dependent tumors. Therefore, the identity of the critical kinase(s) downstream of LKB1 responsible for its tumor-suppressive functions in lung cancer remains unknown.
Mutational data in The Cancer Genome Atlas (TCGA) database do not point to any of the 14 AMPK-related kinases (AMPKR) directly activated by LKB1 as being significantly mutated in lung cancer, and it has been assumed that redundancy among closely related kinases in this family could provide functional compensation, as nearly all of them contain one, two or even three very close homologs (see Fig. 1A). Although the degree of compensation among family members is not well characterized, one possibility is that compensation in tumor-suppressive functions among closely related AMPKR kinase family members masks their individual roles in tumor biology when analyzing their function in isolation, and thus may necessitate the disruption of multiple redundant family members to properly assess their role in tumor biology.
The salt-inducible kinases (SIK1, SIK2, SIK3) are three such redundant kinases among the 14 AMPKRs dependent on LKB1 for their kinase activity. The SIKs are best known to control gene expression through direct inhibitory phosphorylation of the transcriptional regulators class IIa histone deacetylases (HDAC) and the CREB-regulated transcriptional coactivators (CRTC), thus controlling the transcriptional programs executed by these factors. The roles of SIKs in cancer remain very understudied. Among the SIKs, SIK1 had previously been described as a regulator of anoikis and suppression of metastatic potential in human breast cancer cells (23). SIK2 was recently reported to play a prosurvival, oncogenic role in ovarian carcinoma cells (24) and to cooperate with SIK3 in promoting the survival of myeloid leukemia cells (25), indicating that, like AMPK, perhaps SIK kinases can act in a protumorigenic fashion and would not necessarily act as tumor suppressors; or at least exhibit context-dependent roles in different tissues. Most recently, the SIKs were shown to be critical targets inhibited by the oncoprotein GNAS through PKA-mediated phosphorylation. GNAS is mutationally activated in a subset of pancreatic adenocarcinomas, and suppression of the activity of the SIKs appeared to be necessary for tumor maintenance (26), thus indicating that in certain tissues they exert growth-restraining properties. Whether the SIKs have functions in mediating effects of LKB1 in lung cancer has not been explored. Using a combination of cellular systems and genetically engineered mouse models, we analyzed members of the AMPKR family and report here the unexpected finding that the AMPKR kinases SIK1 and SIK3 cooperatively mediate important tumor-suppressor functions of LKB1 in Kras-driven lung cancer in mice.
CRISPR Analysis of the AMPKR Family in A549 Cells
Which of the 14 kinases of the AMPKR family downstream of the LKB1 tumor suppressor are most critical for mediating its effects in NSCLC remains unknown, despite years of intensive study. Given the potential redundancies among different members within subfamilies of the AMPKRs, we sought a cellular system where the tumor-suppressive effects of LKB1 could first be modeled as a surrogate toward eventual genetic modeling of these kinases on lung cancer in the mouse. To begin this analysis, we chose the LKB1-null human lung adenocarcinoma cell line A549. A549 cells show no difference in their growth rate and doubling time in standard full-media growth conditions when LKB1 cDNA is stably reconstituted (Supplementary Fig. S1A), yet show dramatic differences in tumorigenicity when orthotopically introduced into the lung by tail-vein injection (Supplementary Fig. S1B). We examined the expression of the 14 members of the AMPKR family in A549 cells with or without stable LKB1 reconstitution, which revealed detectable mRNAs of all AMPKRs except MARK1 (Fig. 1A), indicating that many downstream kinases could be mediating the effects of LKB1 in these lung cancer cells. CRISPR/Cas9 was utilized to generate a panel of isogenic A549 cell lines, all of which bore stable retroviral reconstitution of wild-type LKB1 cDNA and were further genetically disrupted for different subfamilies of the 14 LKB1-dependent AMPKR kinases using single validated small guide RNAs (sgRNA) against each kinase via stable lentiviral expression. Lines were generated using distinct selectable markers to co-delete multiple members of each subfamily of AMPKRs. Immunoblotting confirmed loss of each AMPKR in its cognate cell line (Fig. 1B–D); however, SIK1 loss was confirmed via qPCR due to lack of sensitive antibodies able to robustly detect endogenous SIK1 under normal cellular conditions in human cells (Supplementary Fig. S1C). We additionally verified SIK1 disruption using Sanger sequencing and found a frequency of >80% in frameshift indels in the region surrounding the guide sequences in all cell lines targeted with Sik1 sgRNA guides (Supplementary Fig. S1D). Notably, despite not having an antibody to detect endogenous SIK1, we characterized a new antibody which recognizes the endogenous activation loop threonine phosphorylated by LKB1 in SIK1, SIK2, and SIK3 in human cells (Fig. 1E) and showed that this phosphorylation signal is correspondingly ablated in SIK knockout (KO) cells. In addition, as previously reported in other cell types (15), here in A549 cells lacking LKB1, there was no evidence of phosphorylation of the activation loop of any of the three SIK family members. CRISPR KO of SIK1 and SIK3 resulted in a modest increase in SIK2 pT175, suggestive of cellular attempted compensation, although phosphorylation of CRTC3 was only modestly further bandshifted with loss of SIK1/2/3 as compared with SIK1/3 (Fig. 1D and E).
Two features of epithelial cancer cells known to correlate well with in vivo tumorigenicity are the ability of cells to survive in suspension, indicative of anchorage-independent growth, and the ability of cells to form colonies when placed in 3-D matrix. Indeed, we observed an LKB1-dependent suppression of growth in A549 cells plated in nonadherent conditions, yet no LKB1-dependent effect over the same time frame in growth on standard adherent conditions (Supplementary Fig. S1E vs. S1F, first two columns). Analysis of cells reconstituted with LKB1 that were subsequently targeted to disrupt subsets of AMPKRs revealed that loss of AMPKα1/α2,MARK1/4, MARK2/3, NUAK1/2, BRSK1/2, or SNRK did not significantly alter the survival of A549 cells after 5 days in nonadherent conditions (Supplementary Fig. S1E). However, the combined loss of members of the SIK family showed a remarkable rescue of growth in nonadherent conditions, comparable with loss of LKB1 (Supplementary Fig. S1E). Notably, this was not observed in standard adherent conditions (Supplementary Fig. S1F).
We next examined colony formation in soft agar in A549 cells, which revealed a nearly 80% suppression of colony growth when LKB1 was reconstituted (Fig. 1F and G). Unexpectedly, loss of AMPKα1/α2, NUAK1/2, MARK1/4, MARK2/3, BRSK1/2, or SNRK did not afford LKB1-reconstituted cells the ability to grow in soft agar (Fig. 1F and G). Analysis of all AMPKR subfamilies revealed very modest effects, except a remarkable full rescue of soft-agar colony formation when the entire SIK subfamily was genetically disrupted. Further study with cells lacking each SIK family member individually or lacking pairs of the SIKs revealed that of the three SIK family members, SIK1 mediated the greatest single effect on soft-agar colony formation (Fig. 1F and G). Loss of SIK2 had a minor effect and loss of SIK3 in particular synergized with loss of SIK1 (Fig. 1F and G). The relative importance of SIK1 and SIK3 to growth in soft agar matched the impact of SIK1, SIK2, or SIK3 loss on phosphorylation of CRTC3, one of the CRTC family of transcriptional coactivators that are targeted by SIKs (Fig. 1D). Genetic disruption of SIK1 or SIK3 alone, but not SIK2 alone, resulted in an increased abundance of the faster-mobility CRTC3 species (lower band), corresponding to a loss of phosphorylation. The observed redundancy among the three SIK family members matches what has been observed in studies of their function in mouse liver, where loss of all three unmasks strong phenotypes that are compensated when any one of the SIKs remains functional (27, 28). Collectively, our findings in A549 cells suggest that the SIK subfamily, and particularly SIK1, may be playing an unappreciated role in lung cancer tumor biology.
Analysis of the Role of the SIK Family in Genetic KRAS-Dependent Lung Tumor Models Using CRISPR
On the basis of the findings in A549 cells, we sought to directly examine the role of SIKs in the well-studied genetically engineered mouse models of KrasG12D-dependent lung cancer (29). Given a lack of readily available conditional floxed alleles of the SIKs, to assess the contribution of Sik1 to tumor development we first turned to a recently described lentiviral vector allowing for coexpression of Cre recombinase with Cas9 and an sgRNA of choice, pSECC (30). pSECC lentiviruses can be delivered in vivo to the lungs of mice to induce CRISPR-mediated gene KO in combination with Cre-mediated activation of oncogenes and deletion of floxed alleles in a single cell (Fig. 2A). We began by generating pSECC vectors bearing three distinct validated RNA guides (sgRNAs) targeting Sik1 (Supplementary Fig. S2A). We first tested the effects of inactivating SIK1 via intratracheal delivery of functionally titered, equivalent amounts of pSECC viruses harboring sgRNAs against Sik1, along with sgRNAs targeting Lkb1 (Supplementary Fig. S2B) and control sgTom (nonspecific sgRNA) or sgLacZ (30, 31) sgRNAs, into KrasG12D mice containing a conditional Cre-activated reporter (ROSA26LSLluciferase), which allows for real-time, noninvasive tracking of tumorigenesis and tumor progression occurring in these mice.
We quantitated tumor development using bioluminescence imaging (BLI) to measure the photon flux emitted from the lungs of these mice over time. This revealed significant acceleration of lung tumor development over the course of 22 weeks in mice treated with sgLkb1 (Fig. 2B). All three cohorts of Kras mice treated with individual Sik1 sgRNAs displayed a modest, but consistent increase in tumor development over the control cohort by BLI over this time course. Of note, in a separate experiment, we additionally tested the effect of delivering pSECC viruses with two validated Mark4 sgRNAsin a parallel group of KRASG12D mice, and found that targeting Mark4 alone in this model did not result in changes in tumorigenesis or progression in a manner distinguishable from control mice (Supplementary Fig. S2C).
To specifically gauge the effects of pSECC-Sik1 viruses in these mice, we quantitated their tumor burden as a percent of tumor area over the total lung tissue area by analysis of hematoxylin and eosin (H&E)–stained lung sections of experimental and control cohorts (Fig. 2C). Although we observed a wide variability in the overall tumor burden per mouse within all groups (Supplementary Fig. S2D), when considering all animals in each cohort as individual sets, as expected, sgLkb1-treated animals clearly displayed a significant increase in tumor burden (20% and 16% for sgLkb1 guides A and B, 0.9% and 1.3% for control sgTom and sgLacZ animals, and 4%, 2%, and 2% for the Sik1 guides). Although the percent tumor burden on average was small in non-sgLkb1 animals at the 22-week endpoint, we corroborated the modest but consistent fold increase in the overall tumor burden of these mice with all three Sik1 sgRNA guides, ranging from 1.45 to 3.6 fold increase over the average of the control sgTom and sgLacZ cohorts.
To more closely examine the effects that disrupting Sik1 had on tumors present in these mice, we analyzed the number and size of individual tumors in each mouse. Whereas the number of tumors per mouse did not reach above an average of about 9, among the non-sgLkb1 mice, the average size of individual tumors in all three cohorts of pSECC-Sik1 mice was significantly higher compared with control tumors, with a positive trend for all three guides used (8-, 3-, and 1.75-fold higher than control tumors for sgSik1 guides A, B, and C respectively; Fig. 2D). These observations represented a skew toward a higher number of larger tumors in sgSik1-treated animals: 20% and 11% of tumors were larger than 1 mm2 in mice with sgSik1-A and sgSik1-B, respectively, vs. 0% of controls, and 60% (sgSik1-A), 82% (sgSik1-B), and 70% (sgSik1-C) of tumors were sized between 0.1 and 1 mm2 compared with 42% for control tumors (Fig. 2E). Thus, targeting SIK1 consistently yielded a larger tumor size on average in cooperation with KRASG12D activation.
Next, we aimed to assess the impact of deleting Sik1 with pSECC-CRISPR viruses in KRASG12D tumors that were also conditionally deleted for the gene encoding the tumor suppressor p53 (KP). Notably, genetic loss of LKB1 on the KP background (KPL) greatly accelerates the rate of lung tumorigenesis, tumor growth, progression, metastasis, and tumor-related morbidity in mice compared with genetic KL and KP models (22). We reasoned that loss of SIKs in a cell-autonomous p53-null setting may likewise provide a more permissive background to better reveal the potential tumor-suppressive effects of SIK1 in models of NSCLC.
In addition to targeting Sik1 or Lkb1 with pSECC in KP mice, we wanted to expand this analysis to include other SIK family members. We therefore developed sgRNAs to additionally target Sik2 and Sik3 (Supplementary Fig. S2B). BLI imaging across a 15-week time course revealed that in contrast to inactivation of Sik2 or Sik3, targeting Sik1 in KP mice resulted in an increased lung tumor burden compared with control mice (Fig. 2F, Sik1 mice only; Supplementary Fig. S3A and S3B, Sik1–3 mice). Indeed, tumor burden quantitation from H&E-stained slides from the lungs of KP mice treated with sgSik1 guides at study endpoint showed an average tumor burden of 23.7% (sgSik1-A), 16.1% (sgSik1-C), and 7.4% (sgSik1-B) compared with a burden of 4.6% and 7.1% of lung tumors induced by control sgTom or sgLacZ, respectively (Fig. 2G). The percent lung tumor burden in mice targeted for Sik2 and Sik3 inactivation on average did not differ from the burden seen in control animals. In contrast, the average of all three cohorts targeting Sik1 with pSECC-sgRNAs in the KP model showed the percent lung tumor burden average to be 18% in sgSik1 animals from an average of 7.2% in control animals (a 2.5-fold change; Supplementary Fig. S3C). At the individual level of KP mice treated with sgSik1 viruses, 5 of 10 animals showed tumor burden over 18%, which was the upper limit of burden in control KP mice, and two individuals presented with burden of 38% and 39%, falling inside the range of tumor burden of KP mice treated with sgLkb1 guides (26%–57%; Supplementary Fig. S3D). Quantitation of individual lung tumors revealed that mice treated with viruses harboring all three sgRNAs targeting Sik1 resulted in a significantly larger tumor size compared with control-treated mice (Fig. 2H). The range of tumors larger than 1 mm2 in mice treated with sgSik1 guides was 33%, 28%, and 15% depending on sgRNA guide, compared with 5% in control animals (Fig. 2I). Nonetheless, considerable variability in tumor size was observed in these models, making us examine whether some of the sgRNAs were more effective than others at inducing disruption of the locus in vivo, and/or whether a selection to observe out-of-frame indels in tumors at endpoint might be associated with specific guides. Using next-generation deep sequencing of barcoded and pooled primary tumors, we observed a greater percentage of frameshift indels with Sik1 guide A, as compared with Sik1 guides B and C (Supplementary Fig. S3E and S3F), which correlated directly with the extent of tumor burden associated with these guides (Supplementary Fig. S3A). Similarly, the more effective Lkb1 sgRNA, guide A, was also associated with a greater percentage of tumors bearing frameshift indels, indicative of a consistent selective advantage for tumors bearing Lkb1 guide A (Supplementary Fig. S3E and S3F vs. S3A).
Analysis of Genetic KRAS-Dependent Lung Tumor Models Using Floxed Sik1 and Sik2 Alleles
As a complementary and powerful alternative way to interrogate the function of SIKs in NSCLC, we turned to recently described conditional floxed alleles of Sik1 (32) and Sik2 (33), which we bred into the Kras-dependent mouse model to conditionally knock out expression of SIK1 and SIK2 (Fig. 2J). We first confirmed that excising Sik1 floxed alleles in mouse liver resulted in the modulation of known LKB1-dependent transcripts (Supplementary Fig. S4A and S4B), corroborating that some known CREB target genes in the liver were induced by conditional deletion of this Sik1 floxed allele. We next generated conditional floxed Kras;Sik1 (KSik1) and Kras;Sik2 (KSik2) mice and compared them against Kras (K) and Kras;Lkb1 (KL) NSCLC models. Lung tumor induction by delivery of lentivirus expressing Cre-recombinase efficiently recombined Sik1 and Sik2 in lung tumors from the corresponding mice, and we further confirmed ablation of the expressed Sik1 transcript in tumors from KSik1 mice (Supplementary Fig. S5A and S5B), which additionally displayed deregulation of known LKB1-regulated transcripts (Supplementary Fig. S5C–S5E).
KSik1 mice showed an acceleration in tumor growth over K mice (Fig. 2K and L; Supplementary Fig. S6A–S6C), resulting in an increase from 8% to 12% in tumor burden over K controls (a 1.5-fold increase), which is within the range of tumor burden fold change observed in Kras mice treated with sgSik1 guides, although this did not reach statistical significance. Intriguingly, 3 of 13 animals presented with high tumor burden, and two of them (37% burden each) matched the level of tumor burden observed in KL mice (32%–56%; Supplementary Fig. S6B), which was never observed in K mice, suggesting an estimation of about 15% to 20% of KSik1 mice displaying very high lung tumor burden. Surprisingly, genetic loss of Sik2 did not increase, and instead slightly reduced, tumor burden compared with K mice, suggesting that SIK2 alone does not play a critical role in tumor suppression in the lung. Individual tumor analysis from KSik1 mice (Fig. 2M; Supplementary Fig. S6D) revealed a highly significant difference in tumor size. Two percent of tumors in K mice were larger than 1 mm2, whereas 9% of KSik1 tumors were this size or larger (Fig. 2N).
Analysis of the Role of the SIK Family in Genetic KP Lung Tumor Models Using Floxed Sik1 Alleles
We next tested the effect of genetic ablation of SIK1 in the KP model of NSCLC. KPSik1 mice displayed an increase in tumor development compared with KP mice as measured by BLI (Fig. 2O; Supplementary Fig. S7A and S7B). Despite variability in tumor burden among individual mice (Supplementary Fig. S7C), 2 of 5 mice reached tumor burdens of 41% to 42% of the total lung area, which is similar to the tumor burden range of KPL mice (46%–62%), whereas KP mice had smaller burden (1%–27%; Supplementary Fig. S7D). As a cohort, KPSik1 mice showed an increase from 10% to 28% in tumor burden over KP mice (a 2.8-fold increase), compared with a tumor burden of 52% in the KPL mice (5.25-fold increase over KP; Fig. 2P), and this was consistent with the results observed in KP mice treated with Sik1 sgRNA guides (Supplementary Fig. S3B; 2.5-fold tumor burden increase of KP+sgSik1 over KP+sgControls). Analysis of individual tumors in KPSik1 mice, however, showed a significant increase in tumor size in these mice compared with KP (Fig. 2Q), and loss of SIK1 in this model increased the percentage of tumors larger than 1 mm2 (18% tumors in KPSik1 vs. 7% in KP), and furthermore 2% of tumors in KPSik1 mice were (>10 mm2; Fig. 2R).
To further ascertain the cell biological effect of SIK1 loss in lung cancer cells, we derived (n = 3) cell lines from KPSik1 tumors, first confirming that they expressed no detectable Sik1 transcript (Supplementary Fig. S8A), and functionally compared them to matched KP lines (n = 4) derived from the same genetic background. KPSik1 cells displayed an enhanced ability to form colonies from single cells in 2-D culture compared with control KP cells, and moreover grew into significantly larger colonies (Supplementary Fig. S8B–S8D). We next tested the ability of these cells to grow colonies in soft agar, and found that KPSik1 cells displayed an enhanced ability to grow both more and larger colonies in soft agar compared with KP cells (Fig. 2S; Supplementary Fig. S8E). Finally, we tested the dependency of SIK1 loss on the growth of these cells, and found that reexpression of the Sik1 cDNA significantly reduced their proliferative capacity (Supplementary Fig. S8F).
Targeted Coinactivation of SIK1 and SIK3 Accelerates Tumor Growth in KRAS-Driven NSCLC
Despite the enhanced tumor growth phenotype resulting from inactivation of SIK1, the loss of any individual SIK family kinase did not fully recapitulate the tumorigenic potential of LKB1 loss in our mouse models. SIK1, SIK2, and SIK3 share common substrates, and therefore it is likely that functional redundancy from the remaining SIK family members may mask the ability to further reveal LKB1-like effects on tumor growth upon loss of an individual SIK, in agreement with our observations of targeted loss of SIK1 versus combinations of SIK1+3 and SIK1+2+3 loss in A549 cells (Fig. 1D–F). To test this hypothesis in vivo, we combined our genetic floxed KSik1 model with pSECC-mediated inactivation of Sik3 (Fig. 3A). This model uses Cre recombinase to activate KRAS and inactivate SIK1, and simultaneously delivers Cas9 and an sgRNA targeting SIK3 (sgSik3-B). Thus, this system effectively inactivates SIK1 and SIK3 simultaneously. Mice with this genetic combination efficiently yielded tumors that were deleted for Sik1 and further showed disruption of reading frames at the Sik3 sgRNA target site (Supplementary Fig. S9A–S9E). Remarkably, compounding the loss of these two SIK family members accelerated tumor growth as compared with K+sgTom mice or with cohorts that inactivated SIK1 or SIK3 singly, either using the genetic model (KSik1) or CRISPR-based targeting with sgRNAs targeting Sik1 or Sik3 (Fig. 3B–D, Supplementary Fig. S9B). Floxed KSik1 mice treated with pSECC-sgTom displayed an intermediate acceleration of tumor growth between KSik1+sgSik3 and K+sgTom mice, which was comparable to inactivation of SIK1 with sgRNAs. Furthermore, BLI imaging showed that all three conditions targeting SIK1 alone resulted in enhanced tumor growth in mice as compared with targeting SIK3 by itself. At study endpoint, tumor burden quantitation showed that KSik1+sgSik3 mice showed a significantly higher burden (10%) over control mice (1%), which was further reinforced upon tumor size quantitation (Fig. 3E and F). Use of deep sequencing of tumors to analyze the frequency of frameshift indels in the tumor DNA revealed a strong selection for Lkb1 sgRNA guide A (Supplementary Fig. S9C and S9D), similar to before (Supplementary Fig. S3E). However, now the increased percentage of tumors bearing a high proportion of frameshift indels could be clearly observed for the Sik3 sgRNA B, strikingly only in KRAS SIK1 mice and not in KRAS mice alone (Supplementary Fig. S9C and S9D). We interpret these results as potential evidence for a selective advantage of tumors inactivating SIK3 only when they already lack SIK1.
Despite an apparent lower total tumor burden in the cohort of KSik1+sgSik3 mice compared with K+sgLkb1 mice (Supplementary Fig. S9F and S9G), the average size of individual tumors between K+sgLkb1 and KSik1+sgSik3 mice was instead comparatively similar (average size >1 mm2, difference in means between these two cohorts not significant) and both were significantly larger than tumors from control K+sgTom mice (Fig. 3F). Notably, mice with these two genotypes were the only ones with average tumor sizes above this threshold among all cohorts tested. We reasoned that enhanced cellular proliferation rates may underlie the dramatic increase in tumor size observed in these cohorts. BrdU positivity analysis showed that KSik1+sgSik3 tumors showed the highest average percent of BrdU incorporation among all cohorts tested (Fig. 3G), indicating that inactivation of multiple SIKs resulted in higher numbers of proliferating cells per tumor compared with control or single Sik-deleted tumors. Correlated with these observations, tumor size distribution analysis showed that KSik1+sgSik3 mice presented with the largest percent of tumors >10 mm2 among cohorts (9% of tumors of this genotype, 5% observed in K+sgLkb1, 2% in KSik1 mice, 0% in all other cohorts; Fig. 3H). Together, our data indicate that the combined inactivation of SIK1 and SIK3 yielded a tumor phenotype characterized by a high percentage of actively dividing cells, resulting in an expanded fraction of large tumors similar in size to those observed upon loss of LKB1. We additionally assessed whether SIK1+3 loss would recapitulate the reported association of LKB1 loss with an elevated recruitment of tumor-associated neutrophils, a hallmark of KL lung adenocarcinomas (11). Using IHC for the neutrophil marker Ly6G, we observed a significant increase in the percentage of neutrophils present in the KL tumors as compared with K tumors, and moreover this increase in neutrophil staining was also mirrored in the KSik1/3 tumors (Supplementary Fig. S9E). Together, these results suggest that loss of SIK1 and SIK3 in the mouse recapitulates histologic aspects of LKB1 loss in both tumor size and immune phenotype.
SIK-Mediated Control of Gene Expression
To determine which transcriptional programs driven by LKB1 inactivation are under the control of the SIKs, we performed transcriptional profiling using RNA sequencing (RNA-seq) on primary tumors plucked from K+Pgk-Cre and K+pSECCsgTom (K), KSik1+Pgk-Cre (KSik1), KSik1+pSECCsgSik3-B (KSik1±3), and K+pSECCsgLkb1-A (KL-A) genotypes. Clustering all genes across all genotypes revealed discrete clusters of genes that were clearly coregulated in KL-A and KSik1+3 tumors (Supplementary Fig. S10A). Expanding these clusters (Supplementary Fig. S10A, red and green boxed insets), it is evident that Sik1+3 deletion induced the same magnitude of transcriptional deregulation as LKB1 inactivation, suggesting that the SIKs are the main targets of LKB1 in the normal suppression of these two specific subsets of genes. Notably, these clusters did not show deregulation in KSik1 tumors (Supplementary Fig. S10A), suggesting that coincident inactivation of SIK1+3 is necessary to reveal the LKB1-dependent transcriptional programs that are under SIK-mediated regulatory control. Moreover, when we unbiasedly clustered the data from K, KSik1+3, and KL-A tumors across all genes and samples, the samples clustered cleanly by genotype, confirming that genotype-dependent transcriptional changes dominated the dataset (Fig. 4A).
We next used differential expression analysis (FDR <0.05, ± 1.5-fold) to define a KSik1+3 gene signature as compared with K tumors and, independently, a KL-A gene signature as compared with K tumors. We then identified the overlap between the KSik1+3 and KL-A gene signatures. Three hundred forty-four (38%) of the genes within the KL-A signature that were upregulated in KL-A tumors were also significantly upregulated in KSik1+3 tumors. Similarly, 627 (41%) of the genes with the KL-A signature that were downregulated in KL-A tumors were also significantly downregulated in KSik1+3 tumors (Fig. 4B). Together, we identified 971 genes commonly regulated by LKB1 or SIK1+3 inactivation (Fig. 4C). These data suggest that SIK1 and SIK3 together may be responsible for approximately 40% of the transcriptional changes induced upon LKB1 loss in primary lung tumors.
We used gene set enrichment analysis (GSEA) to unbiasedly explore the pathways regulated by SIK1+3. We considered the comparison of KSik1+3 tumors to K tumors, and queried this data against the “Hallmark” group of gene sets using GenePattern. We performed the same analysis considering KL-A tumors compared with K. We then compared the top enriched gene sets to determine which LKB1-dependent processes may be mediated by SIK1+3. Two gene sets related to well-established hallmarks of LKB1-mutant lung tumors, epithelial–mesenchymal transition (EMT) and IL6–JAK–STAT3 signaling, were among the top enriched gene sets, 1st and 8th, respectively, in the KSik1+3 data (Fig. 4D). In the KL-A data, these gene sets were the 6th and 7th most enriched. Metascape pathway analysis of the 344 genes upregulated in both KL-A and KSik1+3 tumors (Fig. 4B) confirmed a role for SIK1+3- and LKB1-dependent deregulation of the IL6 pathway, in that “IL6 family signaling” was identified as the second most significant term (Supplementary Fig. S10B). We combined the enriched genes from the GSEA plot with the genes comprised in the Metascape term to create an IL6 signaling gene list, shown in a heat map (Fig. 4D). The enriched genes from the EMT gene set are also depicted in a heat map (Fig. 4D). Thus, transcriptional analyses suggest that loss of SIK1+3 replicates effects on EMT and IL6 signaling pathways that are induced by LKB1 loss in lung tumors.
We then extended our analysis of the 344 genes upregulated in both KL-A and KSik1+3 tumors (Fig. 4B). We queried this gene list in Enrichr. IL6 signaling or related JAK–STAT signaling were the top enriched pathways within this gene list across the Kyoto Encyclopedia of Genes and Genomes (KEGG), WikiPathways, Reactome, and BioCarta analyses (Fig. 4E; Supplementary Fig. S10C). This further implicates SIK1+3 in the regulation of the IL6 pathway. Homer motif enrichment analysis identified one statistically significant motif (P < 1e-12) within the promoters, defined as ±2 kb of the transcriptional start site (TSS), of the 344 upregulated genes (Fig. 4F, top motif). This motif includes the recognition motif for the AP1 family of transcription factors, including AP1, JUND, FOS, FOSL1, and FRA2. Alignment of this motif to the JUND motif is pictured (Fig. 4F). This suggests that the AP1 family of transcription factors may be involved in driving a portion of the LKB1- and SIK1+3-dependent transcriptional changes.
A key established target of the SIKs is the CRTC family of transcriptional coactivators of the CREB transcription factor, which has also been reported to control AP1 activation (38, 39). We found that both AP1 and CREB-dependent transcription became chronically upregulated by SIK1- or LKB1-dependent loss in cultured mouse and human NSCLC cells, in correlation with bioinformatic analysis of mouse tumors (Supplementary Fig. S10D–S10G, Supplementary Fig. S11A–S11D). To query this bioinformatically, we searched the promoters of the 344 upregulated genes common to KL-A and KSik1+3 tumors for AP1 and CRE motifs. Seventy-four percent of genes have at least one motif for CREB (CRE), AP1, or both in the promoter (Supplementary Fig. S10D). It follows that the genes involved in the biological processes that are deregulated upon LKB1 and SIK1+3 loss, including EMT and IL6 signaling, are likely to be under transcriptional control by the transcription factors whose activity are modulated in these conditions, potentially CREB and AP1. Indeed, of the 344 genes commonly upregulated in murine KRAS tumors by LKB1 loss and SIK1+3 loss, 55 genes contain both predicted elements in their promoters (Fig. 4G), 31 contain CRE elements (Supplementary Fig. S10E), and 170 contain AP1 elements (Supplementary Fig. S10F). Although only some of these motifs are likely to be functionally bound by CREB or AP1 in primary lung tumors, these data suggest that both CREB and AP1 may be involved in regulating a subset of the genes upregulated upon LKB1 and SIK1+3 loss. One well-studied promoter that contains functional AP1 and CREB binding sites is that of IL6 (34). As the enrichment analysis suggested, FPKM plots of the RNA-seq data from each tumor per genotype reveals transcriptional upregulation of nearly all steps of the canonical JAK–STAT pathway, from cytokines (IL6, IL33) to receptors (IL4RA, OSMR), core components (JAK3, STAT1, STAT3), coactivators (BCL3), and well known STAT targets (SOCS3; Fig. 4H). Thus, our data support the model that LKB1 loss in lung tumors induces IL6 signaling through inactivation of SIK1+3.
To assess the overlap between the changes in gene expression by loss of Sik1+3 deletion in mouse tumors and genes regulated by loss of LKB1 in human cells, we used GSEA to query the top 500 upregulated genes in Sik1+3 tumors and compared them against a ranked list of all genes differentially regulated by the presence or absence of LKB1 in human NSCLC A549 cells based on RNA-seq expression profiling (from A549-pBABE or A549+LKB1 cells represented in Figs. 1 and 5). We found there was a significant enrichment of genes upregulated in mouse tumors by SIK1+3 loss that were also found to be upregulated in human A549 cells compared with LKB1-reconstituted cells (Supplementary Fig. S10G, graph and first line in the summary table), which notably included IL6 and IL33. We also queried the set of 344 genes that were upregulated in common by SIK1+3 loss and LKB1 loss (Fig. 4B) to ask whether the set of genes that were commonly regulated by SIK1+3 loss and LKB1 loss in mouse tumors were also regulated by LKB1 in human NSCLC cells. GSEA showed a significant enrichment of this gene set when comparing genes differentially regulated by LKB1 in A549 cells (Supplementary Fig. S10G, second line in the summary table). Furthermore, this gene set was also found to be enriched in differential transcriptional profiles generated by RNA-seq of two other human NSCLC lines that we queried (H1355, H157) in the LKB1-null state versus LKB1 cDNA reexpression (Supplementary Fig. S10G), collectively indicating that SIK1+3 loss in mouse tumors induces transcriptional changes that are mirrored in part by LKB1 loss in the mouse and in human cells. These included genes associated with IL6 pathway signaling, which further raised the possibility that LKB1 may control IL6 signaling through SIK1/3 in human NSCLC cells.
SIKs Control IL6 and STAT Signaling via CRTC2
To further explore whether the gene-expression changes seen were due to cell-autonomous effects of loss of SIK1/3 signaling on transcription within the tumor cells themselves, as compared with elevated cytokine signaling in surrounding stroma or immune cells, including the elevated presence of tumor-associated neutrophils present in the microdissected tumors, we turned to human A549 NSCLC cells in culture. Picking a limited number of well-known SIK1-regulated mRNAs (e.g., NR4A1, PDE4B), we examined a limited number of genes in the cytokine/JAK–STAT signaling pathway deregulated in the KL and KSik1+3 tumors. As seen in Supplementary Fig. S12A, mRNAs of some cytokines and cytokine receptors (IL10, IL15, OSMR) were suppressed by LKB1 reconstitution in A549 cells, and were largely reversed by SIK1/3 deletion in LKB1-expressing cells.
As IL6-dependent signaling has been shown to be a common feature of other LKB1-deficient contexts (35), we examined the cell autonomy of LKB1 effects on IL6. As a first examination, we created Lkb1 or Sik1/3 CRISPR KO murine KP cell lines using independent combinations of sgRNA guides targeting Sik1 and Sik3 in two independent KP lines, and profiled IL6 production by ELISA. In these cells, we observed significantly elevated IL6 production in both the Lkb1-deleted and Sik1/3-deleted KP cells (Fig. 5A). We next examined IL6 production in A549 cells with and without LKB1 reconstitution or Sik1/3 CRISPR deletion. Notably, A549 cells produced IL6 basally, which was strongly inhibited when LKB1 was restored (Fig. 5B), and this effect of LKB1 reintroduction was fully reversed upon deletion of SIK1/3.
Given the presence of CRE and AP1 binding motifs in the IL6 promoter (34), we next examined whether CRTC family members may mediate effects of LKB1 or SIK1/3 on IL6. First, we examined mRNA expression levels of the three members of the CRTC family in murine KP tumors as well as in A549 cells from RNA-seq data. In both mouse (Supplementary Fig. S12B) and human (Supplementary Fig. S12C) KRAS-mutant lung adenocarcinoma cells, CRTC2 was the most prominently expressed family member. Using two distinct sgRNAs, CRTC2 was deleted in A549 cells bearing LKB1 reconstitution and SIK1/3 deletion, which resulted in complete suppression of IL6 production (Fig. 5B). Given this requirement of CRTC2 for IL6 production, we examined whether gene expression controlled by SIK1/3 loss might similarly be dependent of CRTC2. We observed partial or full dependence on CRTC2 among genes well known to be CREB targets (NR4A2, NR4A3), but also among cytokine signaling (IL33, STAT1) and proliferation genes (AREG, EREG; Fig. 5C). To examine whether these elevated mRNAs would translate into altered signaling pathways, we examined total STAT1 protein levels in our A549 cell series. A549 cells expressed STAT1 protein, which was reduced by LKB1 reconstitution, but that effect of LKB1 was reversed by SIK1/3 deletion (Fig. 5D; Supplementary Fig. S12D). The increase in total STAT1 protein (both α and β splice isoforms induced by loss of SIK1/3) was in turn reversed by deletion of CRTC2. In addition, SIK1/3 loss in the LKB1-proficient cells resulted in not only elevated STAT1 proteins but also increased basal phosphorylation of STAT1 at Y701 (Fig. 5D; Supplementary Fig. S12D), perhaps due to autocrine cytokines also being produced in parallel with STAT1 upon SIK1/3 deletion. This too was reversed by CRTC2 loss (Fig. 5D).
We next examined whether CRTC2 was also a key target of SIK1/3 loss on growth in anchorage-free settings. First, we returned to growth and anoikis survival in detached conditions, and as observed earlier (Supplementary Fig. S1E), reexpression of LKB1 in A549 cells reduced their growth and survival in detachment, which was largely reversed by SIK1/3 loss, and this added growth from SIK deficiency in detached conditions was suppressed by CRTC2 loss (Fig. 5E). The enhanced growth in soft agar induced by SIK1/3 deficiency was also diminished by CRTC2 deletion (Fig. 5F), indicating that CRTC2-dependent control of transcription may be a major target of LKB1- and SIK1/3-dependent tumor suppression.
SIK-Mediated Effects Are Largely LKB1-Dependent
As we and others have previously demonstrated, despite LKB1 being a predominant upstream kinase for AMPK in most tissue types, most cell types also use CAMKK2 as an upstream kinase for the same activation-loop threonine that LKB1 phosphorylates, providing residual AMPK activity in LKB1-deficient cells, tumors, and tissues (36). Now armed with better tools and clear effects on a number of readouts, we sought to examine whether SIK family kinases might also retain catalytic activity independent of LKB1. Our first hint that this was not the case came from use of the new activation-loop threonine phospho-specific antibody, which recognizes endogenous SIK1, SIK2, and SIK3 in A549 cells (Fig. 1E). Despite an apparent absence of pSIK1, pSIK2, or pSIK3 in the A549 cells which lack LKB1, we wondered whether an alternative kinase might be upregulated under detached conditions, given that this is a critical context for the SIKs and LKB1 activity. We next performed immunoblotting analysis to profile a number of targets of LKB1 and AMPK in A549 cells with or without LKB1 add-back, and/or with SIK1/2/3 deletion (Fig. 5G). This analysis again found no evidence of residual pSIK1, pSIK2, or pSIK3 in A549 cells without LKB1 restoration, unlike pAMPK, which was still residually phosphorylated in A549 cells, particularly in detached conditions, as was the AMPK substrate pRaptor (Fig. 5G). Furthermore, we performed in vitro kinase assays on SIK1 and SIK3 isolated from LKB1-containing or LKB1-deficient A549 cells, revealing a complete loss of SIK1 and SIK3 catalytic kinase activity in the LKB1-deficient cells, consistent with the activation-loop threonine being required for catalytic activation of these kinases (Supplementary Fig. S13A–S13D). To functionally address whether SIK1 and SIK3 may have tumor-suppressor functions independent of LKB1, we used A549 cells targeted for deletion of SIK1 or SIK3 individually or in combination and performed soft-agar colony formation. First, with single disruption of SIK1, or SIK2, or SIK3, there was no statistical increase in colony number in the LKB1-deficient cells (Supplementary Fig. S13E), unlike the benefit of SIK1 loss on soft-agar colony formation in the context of A549 cells bearing LKB1 (Fig. 1F and G). However, a modest increase (∼1.8-fold) in soft-agar colony formation was observed when SIK1 and SIK3 were both removed from A549 cells (Supplementary Fig. S13F), compared with >4.5-fold increase in colony formation in the LKB1-replete A549 cells (Fig. 1F). The observed increase in soft-agar colonies when SIK1/3 are deleted in LKB1-null cells raises the possibility that these kinases may have some LKB1-independent scaffolding functions or roles not reliant on SIK1/3 kinase activity in these cells.
Here we have made the unexpected finding that the AMPK-related kinases SIK1 and SIK3 redundantly mediate some of the tumor-suppressor activity of LKB1 in KRAS-driven lung cancer. Notably, head-to-head analysis of the 14 LKB1-dependent AMPK family kinases in A549 cells revealed a unique role for the SIKs in suppressing soft-agar colony formation as well as susceptibility to cell death in detached conditions. In both assays, combined SIK1/SIK2/SIK3 deletion gave a phenotype most similar to LKB1 loss, and none of the other AMPKR subfamilies gave rise to similar effects on soft-agar colony formation or growth in detached conditions. Indeed, our previous genetic studies in mice indicate that AMPKα1 and AMPKα2 do not play a conventional tumor-suppressor role in KRAS-driven lung tumors and when deleted result in decreased tumor burden, in stark contrast to the robust tumor increases in models from loss of LKB1 (22). In both K and KP models of NSCLC, loss of SIK1 accelerated tumorigenesis, albeit in a manner not fully penetrant to phenocopy LKB1 loss. It is worth noting that p53 and LKB1 are mutually exclusively mutated in the context of lung adenocarcinomas with KRAS mutations in human patients, so the KPL mouse model is simply an additional tool for examining effects of LKB1 tumor-suppressor activity. Interestingly, LKB1 is comutated with p53 in human adenocarcinomas with wild-type KRAS, so this is a context worth exploring further in mouse models to define SIK-dependent effects.
Notably, the K and KP models with pSECC-induced targeting of the SIKs or LKB1 generated only adenocarcinomas and we did not observe an increased incidence of squamous cell carcinoma (SCC) or adeno-squamous transdifferentiating lesions at the histologic level. An increased incidence of SCC cells has been reported as a hallmark of LKB1 loss in some mouse models (5, 6, 9, 37, 38).Although the reasons for this discrepancy are not resolved, we believe that some contributing factors could be potential differences in latency of tumor development at experimental endpoint among different models, the choice of strain background, or the use of lentiviral Cre, which is thought to favor the incidence of adenocarcinoma development in models of KRAS/LKB1 over adenoviral Cre (39), which may influence the frequency at which histologically defined squamous carcinomas may be able to develop and be detected in these models.
Here we found that concomitant loss of SIK1 and SIK3 in KRAS-mutant tumors cooperatively accelerated lung tumorigenesis and promoted the ability of tumors to grow comparably with those observed by loss of LKB1 (Fig. 3). Furthermore, the loss of SIK1/3 recapitulated the reported phenotype observed in LKB1-mutant tumors denoting a significant accumulation of neutrophils within tumor areas. In our studies, we observed over a 3-fold increase in the number of infiltrating neutrophils in both LKB1- and SIK1/3-targeted tumors when compared with tumors from KRAS control mice by histologic quantification of the neutrophil marker Ly6G. An LKB1-dependent increase in tumor-associated neutrophil infiltration has been thought to contribute to the establishment of a proinflammatory, protumorigenic state in lung tumors bearing LKB1 mutations, and is characterized by an increase in the production and secretion of IL6 by tumor cells, which attracts neutrophils that exert immunosuppressive effects (11). Future studies are needed to determine whether additional codeletion of Sik2 or a distinct AMPKR might further enhance the SIK1/SIK3 tumor and immune phenotypes. It will also be interesting to examine how combined floxed alleles of Sik3 might synergize with loss of SIK1 and activation of KRAS in the lung, because pSECC targeting is not fully penetrant (ref. 30; Supplementary Fig. S9C and S9D), and thus may limit the amount of synergistic signal that loss of SIK3 could provoke in this study. Independently, in elegant studies using CRISPR gene editing in Cas9-expressing mice, Murray and colleagues have also found cooperating roles for SIK1 and SIK3 as arbiters of tumor suppression in NSCLC (40). Together, these studies underscore critical functional roles which these two kinases share in mediating LKB1′s tumor-suppressive function in the lung.
Functional analysis of the three SIK genes in mammals has largely come from studies of their roles in metabolic tissues such as liver, adipose, and muscle, where they are known to phosphorylate and inhibit the CRTC family of transcriptional coactivators as well as the class IIa family of histone deacetylases (27, 28, 32, 41–43), effects we observed here in A549 NSCLC cells (Fig. 5G). Studies of the role of the SIKs in different metabolic processes in metabolic tissues have been performed in genetic KO mice and in cell-culture systems, but the role of the SIKs in tumor cells is very underexplored. However, there are hints in the literature that SIK1 and/or SIK3 may behave as tumor suppressors in some contexts. Notably, very recently biochemical inhibition of the SIK subfamily was found to be a key driving force of PKA-dependent signals from oncogenic GNAS in pancreatic cancer (26). Moreover, an early shRNA screen for kinases involved in colony formation and tumorigenesis in mammary epithelial cells reported SIK1 as one of the top hits of the screen (23). Depletion of Sik1 using shRNAs or expression of a kinase-dead SIK1 led to increased metastatic spread in a transplant model, which was paralleled by TCGA data showing that reduced SIK1 mRNA correlated with poor clinical outcomes in breast cancer (23).
Given that the study of the SIKs to date has mainly been in metabolic tissues for their role in controlling transcriptional regulators, we performed transcriptional analysis on primary tumors from KRAS mice alone or those bearing loss of LKB1 or SIK1 and SIK3. Transcriptional profiling of primary tumors revealed that loss of LKB1 and loss of SIK1/3 in KRAS lung tumors trigger highly overlapped transcriptional responses, with 38% of the genes upregulated from LKB1 deficiency being upregulated by SIK1/SIK3 loss. Given that LKB1 activates 12 kinases in addition to SIK1 and SIK3, it is surprising that loss of just those two kinases would trigger such an overlapping extent of gene expression. Analysis of the 344 genes upregulated in common between LKB1 and SIK1/3 loss revealed many promoters bearing CREB and AP1 binding sites. Notably, both AP1 and CREB-dependent transcription have been tied to tumorigenesis before. Notably, studies on the role of the CRTCs, perhaps the best-established substrates of the SIKs, have been common in a specific type of mucoidal epithelial (MEC) tumors caused by translocations involving hyperactivation of CRTC function. Increased expression of both CREB and AP1 target genes has been studied in this tumor context extensively (44). In some of the initial studies identifying the CRTC proteins, it was noted that the phorbol ester TPA can cause nuclear translocation of CRTC1 (45) and that deletion of the AP1 binding sites in the IL8 promoter resulted in it being no longer responsive to CRTC1 (46). Subsequent studies directly showed that CRTC1 is a coactivator for AP1 and its regulation of some AP1 targets is independent of any effects on CREB (44). Importantly, overexpression of CRTC1 synergized with c-Jun overexpression to drive growth in HeLa cells (which are LKB1-deficient), and siRNA to CRTC1 blocked the ability of overexpressed c-Jun to induce BrdU staining, directly implicating CRTC1 in cell proliferation (44). Indeed, we observed elevated BrdU in the SIK1/SIK3-deficient tumors, and colony formation in SIK1-deficient KP cells was greatly enhanced as well (Fig. 3F). Besides AP1-dependent effects on proliferation, recent studies of the CRTC1–MAML2 fusion in MEC tumors have found that CRTC1-dependent induction of AREG, an EGF family ligand, occurred in a manner dependent on CREB binding the AREG promoter (47), and AREG-dependent activation of EGFR was important for the tumorigenic potential of CRTC1–MAML2 fusion proteins in MEC tumors. Indeed, treatment of such tumors with the EGFR mAb cetuximab blocked growth of xenografts (47). Given the importance of AREG–EGFR to proliferation of these MEC tumors, it is notable that we found AREG was upregulated in this study in LKB1-deficient tumors in a SIK1/3–CRTC2-dependent manner. In addition, another CRTC transcriptional target, NEDD9, was reported to be important for LKB1-mediated tumor suppression (48). Recently, working with our colleagues, Rodón and colleagues (49) reported the involvement of CRTC2 and the CREB target gene ID1 as important mediators of human NSCLC cell and tumor growth in LKB1-null settings, which further highlights the role of LKB1–CRTC–CREB signaling in the regulation of lung cancer cell growth. EMT was another well-noted LKB1-dependent effect on gene expression (50), and is also highly upregulated in the SIK1/SIK3-deficient tumors.
The single most common biological pathway upregulated at the mRNA level in common between LKB1 loss and SIK1/SIK3 loss was IL6–JAK–STAT signaling. Very recently, two studies describing conditional models of Peutz–Jeghers syndrome, the cancer predisposition syndrome exclusively driven by congenital mutations in LKB1, have drawn a consistent connection between LKB1 deficiency and upregulation of the IL6–JAK–STAT pathway in different lineages (51, 52). Loss of LKB1 in T cells or intestinal stromal cells resulted in the secretion of the cytokines IL6/IL11, which in turn drove hyperactivation of JAK–STAT signaling in the stromal compartment of the intestine required for tumor growth. One group furthermore found that reduced expression of SIK1, MARK1, or MARK4 in fibroblasts was sufficient to induce production of IL11 (51), which suggests that control of this proinflammatory axis may be a common target of SIK- and LKB1-dependent regulation in multiple tissues and cell types. We show here in both murine and human lung adenocarcinoma cell lines that SIK1/3 deletion provokes IL6 production to a similar extent as LKB1 deletion, and IL6 production is ablated if CRTC2 is deleted (Fig. 5A and B). It will be important to investigate whether inhibition of the IL6–JAK–STAT pathway could have therapeutic value in LKB1-mutated lung cancer, especially given that therapeutic options are so limited for this tumor type. Furthermore, it will be interesting to determine whether additional genes within our newly defined list have as yet unexplored targetable potential.
Mechanistically, it is also noteworthy that both SIK1 and SIK3 were previously reported to regulate TLR4 signaling to NFκB via direct binding and regulation of TRAF6 (53). Whether CREB, AP1, NFκB, or STAT-dependent gene expression underscores the primary resistance seen in KL mouse models and the striking resistance to immune checkpoint inhibitors seen in patients with LKB1 mutations remains to be determined, but several specific hypotheses can now be tested in future studies.
Although unexpected, the finding that SIK1 and SIK3 mediate much of the tumor-suppressor function of LKB1 also opens new avenues of therapeutics aimed against the pathways directly regulated by SIK1 and SIK3. Given that only a handful of substrates are known for SIK1 and SIK3, there is much to be learned mechanistically about their function in lung cancer, which may have broad implications for resistance to checkpoint inhibitors and the large fraction of NSCLC bearing LKB1 mutations, for which there are currently no robust therapeutic strategies.
Cell Culture and Cell Lines
All cell lines were maintained in DMEM supplemented with 10% FBS and incubated at 37°C with 5% CO2. All cell lines were tested for Mycoplasma once a month. A549 cells were from ATCC, and isogenic A549 lines reconstituted with pBABE-LKB1 wild-type (WT) were obtained by retroviral transduction under selection (hygromycin, 600 μg/mL). 634T KP cells were a gift from Kwok Wong (New York University). FVB-mouse lung tumor-derived KP and KPSik1 cells were isolated from freshly collected mouse tumors, minced, and digested with 1 mg/mL collagenase/dispase (Sigma) in DMEM + 10% FBS, and 1× penicillin/streptomycin for 2 hours under rotation at 37°C, and were subsequently cultured on DMEM with 10% FBS, 1× penicillin/streptomycin (Gibco), and 2.5 μg/mL Fungizone (Gibco), with media replacement every 48 hours over 6 days to establish sterile cultures. KPSik1 tumor-derived cell lines were confirmed to be SIK1 KO by qPCR of Sik1 transcript. For luciferase assays, 634T cells stably expressing sgTom, sgLKB1, and sgSIK1 guide RNAs were transfected with plasmids encoding AP1 target promoter and CREB-binding Cre promoter reporters [Cignal CRE (CCS002L), AP1 (CCS-011L) coupled to luciferase], according to the manufacturer's instructions.
In Vitro Kinase Assay
FLAG-tagged SIK1 WT cDNA or T182 activation-loop mutant (TA), or SIK3 WT or TA-mutant were expressed in A549-pBABE cells or A549-LKB1–reconstituted cells with pLenti-Neo. Equivalent amounts were immunoprecipitated (SIK1, FLAG-M2-resin, SIK3, Bethyl Laboratories A302-455A with Protein A beads) from cell lysates (3 mg/sample). SIK activity assays were performed in a final volume of 50 μL containing 15 μL bead volume of protein kinase and 35 μL of reaction mix, using 200 μmol/L of Sakamoto-tide (CRTC2 peptide ALNRTSSDSALHRRR; MRC-PPU Reagents) as a substrate, with 50 mmol/L HEPES, 1 mmol/L EGTA, 10 mol/L magnesium acetate, and 2 μmol/L [γ32P]-labeled ATP. Control reactions lacked kinase using control immunoprecipitates, or lacked Sakamoto-tide in the presence of immunoprecipitated kinase. Reactions were incubated at 30 degrees for 20 minutes, and were terminated by spotting 40 μL of reaction on P81 phosphocellulose paper followed by immediate immersion in 75 mmol/L orthophosphoric acid. Papers were washed three times for 10 minutes with 75 mmol/L orthophosphoric acid, and washed one final time for 5 minutes in acetone. Papers were air-dried, and incorporation of P-32 was quantitated by Cherenkov counting.
Soft-Agar Colony Growth Assay
KP/KPSik1 cells (0.02 million) or A549 cells (0.03 million) were mixed with agarose in 2× growth medium (final concentration 0.3%), plated on top of a solidified layer of 0.6% noble agar in 6-well plates, and fed every 4 to 5 days with growth medium. Cells were collected after 3 weeks, stained with 0.02% Giemsa stain (Sigma-Aldrich 48900) for 20 minutes at room temperature, and then washed with water and stored at 4°C overnight for stain development. Plates were imaged using a LI-COR Odyssey CLx (LI-COR Inc.) and Image Studio Lite software (LI-COR Inc.). Analysis of colony number and colony size were performed using ImageJ software (NIH).
2-D Colony Formation Assay
One thousand cells were plated on 10-cm dishes in full media (10% FBS in DMEM). After 7 days, cells derived from KP and KPSik1 GEMMs were washed with PBS, fixed for 15 minutes in 10% formalin (EMD Millipore R04586-82) at room temperature, and stained for 20 minutes with 0.25% crystal violet (Sigma C3886). Plates were washed, imaged, and analysis of colony number and colony size was performed using ImageJ software.
Cell Survival under Detachment Conditions (Anoikis Resistance).
A549 cells (5 × 105) were plated on solidified 1% noble agar–coated P60 plates in full-growth media to allow for growth in detached conditions. After 5 days, cells were collected, briefly incubated with TrypLE to produce single-cell suspensions, and transferred to 24-well plates for proliferation/survival analysis using the CyQuant Direct Cell Proliferation Assay according to the manufacturer's protocol (Invitrogen C35011). Briefly, cells were incubated in staining buffer for 1 hour at 37°C and analyzed on an Infinite M1000 Pro plate reader (Tecan). For quantification, values from LKB1 add-back and AMPKR KO cells were divided by values from control A549 cells and expressed as percentage of control.
Five thousand A549, KP, or KPSik1 cells were plated in triplicate in 96-well plates and analyzed using CyQuant assay at time-zero, 24-hour, 48-hour, and 72-hour timepoints to infer relative cell proliferation. Values for each cell line were divided by values collected at time zero and are expressed as percent growth.
To assess IL6 secretion by tumor cells, 0.5 mol/L A549 or KP CRISPR-modified cells were plated in duplicate in 6-well plates containing 1.5-mL of full media (10% FBS in DMEM). Supernatants were harvested after 48 hours and flash-frozen in liquid nitrogen before use. Human and mouse IL6 ELISA assays were performed per manufacturer's instructions (BD Biosciences). Briefly, plates were prepared overnight with IL6 antibody, and 100 μL of each sample was loaded in duplicate in 96-well plates. As needed, supernatants were diluted and reassessed to stay within the standard range of detection of the assay.
sgRNAs targeting human or mouse AMPKRs were selected using the optimized CRISPR design tool (http://crispr.mit.edu). Guides with high targeting scores and low probability of off-target effects were chosen, targeting the 1st or 2nd coding exon, or otherwise the first most common downstream exon for transcripts reported to produce multiple isoforms from searches of UniProt or Ensembl databases. At least two independent sgRNA sequences were selected for each gene. Oligonucleotides for sgRNAs were synthesized by IDT, annealed in vitro, and subcloned into either BbsI-digested pSpCas9(BB)-2A-puro (pX459) vector (Addgene no. 48139), and BsmbI-digested lentiCRISPRv.2-puro (Addgene 52961) lentiCas9-blast (Addgene 52962), or pSECC (Addgene no. 60820). Validation of guide specificity was assessed by Western blot analysis of low-passage A549 human or 634T murine cells transfected with control nontargeting guides, sgTom, or sgLacZ, or sgAMPKR guides with pX459 followed by transient selection (48 hours) with 2 μg/mL puromycin, or stable integration of pLentiCRISPRv.2 or lentiCas9-blast counterparts by lentiviral transduction with 0.45 μm–filtered viral supernatant supplemented with Polybrene followed by selection with 2 μg/mL puromycin or 10 μg/mL blasticidin. Two-dimensional proliferation assays, anoikis resistance assays, and colony growth in soft agar assays were performed with cells stably expressing sgRNAs from pLentiCRISPRv.2-puro and lentiCas9-blast. pSECC lentiviral vectors harboring identical mouse sgRNAs were utilized for large-scale viral production for in vivo delivery. Oligonucleotide sequences are listed in Supplementary Table S1.
Lentiviral Production and Titering
Lentiviruses made from pLentiCRISPRv.2, pSECC, and Lenti pGK-Cre (latter vector was a gift from Tyler Jacks, MIT) were produced by cotransfection of the lentiviral backbone constructs and packaging plasmids psPAX2 (Addgene 12260) and pMD2.G (Addgene 12259). For low-scale preps, Lipofectamine 2000 (Thermo Fisher Scientific) was used as a transfection reagent at a ratio of 3:1 lipofectamine:DNA. For large-scale preps, the transfection reagent was 1 μg/μL PEI, pH 4.5 (Polysciences #23966-2) at a ratio of 4:1 PEI:DNA). Viral supernatant was collected 48 and 72 hours post-transfection, concentrated by ultracentrifugation at 20,000 rpm for 120 minutes, and resuspended in small volumes of 1× Hank's Balanced Salt Solution prior to aliquoting, flash-freezing, and storing at −80°C. Most large-scale viral preps were made in-house and were supplemented as needed with custom orders prepared by the Salk Institute GT3 Viral Vector Core or the University of Iowa Viral Vector Core. For titering, lentiviral preps for mouse experiments (pSECC and PGK-Cre) were functionally titered by transduction of a reporter line (293-LSL-GFP), which turns on expression of GFP upon Cre-mediated recombination and allows quantitation of functional titers derived from the percent of GFP-positive cells obtained by flow cytometry. The formula used to calculate transforming units/mL (TU/mL) was TU/mL = (percent GFP+cells × number of cells seeded)/(volume of virus used × dilution fold factor). Briefly, 80,000 293-LSL-GFP cells were plated on 24-well plates on day 0 and transduced on day 1 in duplicate across four 2-fold serial dilutions of virus supernatant. After 48 hours, cells were collected, fixed, and analyzed by flow cytometry to calculate % GFP positivity with an LSR II (BD Biosciences). TU/mL titers for each virus were derived from the average TU/mL across the dilution series, assayed two independent times in duplicate wells for each virus. All viruses used together in a given experiment were titered in a batch, and were subsequently normalized to deliver an equal number of TU/mouse.
GEMM and CRISPR Models.
All procedures using animals were approved by the Salk Institute Institutional Animal Care and Use Committee. All mice were maintained on the FVB/N background. KRAS (KrasLSLG12D/+; R26LSL;luc/luc); KL (KrasLSLG12D/+; Lkb1fl/fl;R26LSL;luc/luc), KP (KrasLSLG12D/+; p53fl/fl;R26LSL;luc/luc), and KPL (KrasLSLG12D/+; Lkb1fl/fl;p53fl/fl;R26LSL;luc/luc) mice in FVB/N have been previously described (7, 22) and Sik1fl/fl and Sik2fl/fl conditionally floxed mice have also recently been described (32, 33). In this study, Sik2fl/fl was crossed into the FVB/N K background for 8 generations and Sik1fl/fl into the FVB/N K and KP background for 4 generations before generating experimental mice. All experiments used a mixture of female and male mice. Experimental endpoint was defined across experiments as the time point at which the experimental cohorts of K+sgLkb1, KL, or KPL mice reached tumor burden of 109 mean photon flux. Experiments were stopped at that time, and all mice in that experiment were collected at that point. All animals at experimental endpoint were included for analysis of lung tumor burden and tumor size analysis. No animals were excluded from longitudinal BLI measurements and graphs. Because of the different kinetics of tumor growth inherent in K and KP animals, experiments with K and KP animals were considered independent of each other, and thus specific amounts of virus were administered to each set of K and KP animals. Viruses were delivered by intratracheal intubation, according to the protocol of DuPage (54). Amounts of virus delivered per mouse are as follows: 75K TU/mL of pSECC for KRAS animals, 50K TU/mL pSECC for KP animals, 2 × 106 TU/mL PGK-Cre to KRAS animals, 4 × 105 for KP animals, and 4 × 105 for KP animals. For intravenous lung tumor studies, 8-week-old female SCID Beige mice (PrkdcscidLystbg-J/Crl:Fox Chase, catalog no. 250) were injected with 1 × 106 luciferase-expressing A549 cells into the lateral tail vein. All mice were imaged via BLI 30 minutes after intravenous injection. BLI was performed once per week for a period of 52 days.
BLI was performed biweekly for all cohorts, with the exception of the floxed KP/KPSik1/KPL cohort, which was imaged weekly. All imaging was done using an IVIS Spectrum (Caliper Life Sciences). Mice were injected intraperitoneally with 150 mg/kg d-luciferin (Caliper Life Sciences), anesthetized with isoflurane, and imaged both ventrally and dorsally 10 minutes post luciferin injection. The total lung photon flux for each animal was calculated by the combination of ventral and dorsal photon flux calculated within a region of interest encompassing the thorax.
IHC and Image Analysis
Lungs from mice were collected at each experimental endpoint as noted in the figures, and were fixed in formalin for 16 hours, transferred to 70% ethanol and paraffin-embedded (FFPE) at the Salk Institute Histology Core or Pacific Pathology (San Diego, CA). Five micron–thick sections were prepared and stained with H&E for tumor burden quantitation. For BrdU IHC, slides were deparaffinized and rehydrated, and antigen retrieval was performed for 30 minutes at high heat in citrate buffer. Endogenous peroxidase activity was quenched with hydrogen peroxide. Slides were blocked with 5% goat serum and incubated overnight with primary anti-BrdU antibody diluted in blocking buffer, and secondary antibody incubation was carried out according to the manufacturer's instructions (ImmPress HRP Reagent Kit, Vector Labs MP-7444). Staining was visualized with ImmPACT DAB peroxidase substrate (Vector Labs SK-4105), and further counterstained with hematoxylin, dehydrated through ethanol and xylenes, and mounted with Cytoseal 60 (Thermo Scientific). H&E-stained and immunostained slides were scanned using a PerkinElmer Slide Scanner (Panoramic MIDI Digital Side Scanner) for further downstream analysis using inForm v2.1 image analysis software (Cambridge Research and Instrumentation).
Lung Tumor Burden
Total lung tumor burden was quantitated from H&E sections using inForm v2.1 image analysis software (Cambridge Research and Instrumentation) in an unbiased manner. In brief, the TrainableTissue Segmentation method was trained to identify tumor, normal lung, vessel, and space. This program was then applied to all H&E images, and each of the resulting mapped images was then screened to verify that accurate tissue segmentation had occurred. The quantitation data from this analysis was then used to calculate the percentage of tumor area as normalized to total lung area (tumor area + normal lung area).
Tumor Size Quantitation
The size of each individual tumor was measured from H&E sections using morphometric analysis in Panoramic Viewer software (PerkinElmer), which calculates the size of each identified tumor by area in squared microns. The area of all tumors found in 5 lobes of each mouse was exported and compiled to plot every tumor per mouse. The data were plotted for each mouse on an x-axis and tumor size on the y-axis.
BrdU Positivity Analysis
For each tumor analyzed, image preparation and tissue and cell segmentation were performed using inForm to calculate the ratio of diaminobenzidine-positive nuclei out of the total number of hematoxylin-stained nuclei per tumor to derive a percent BrdU-positive score for each tumor. All individually cropped tumor images were analyzed simultaneously as a batch and verified for accuracy in a blinded manner to exclude extreme outlier images from final analysis.
Protein lysates from cells were isolated in cold 1% NP-40 RIPA buffer supplemented with protease inhibitors (cOmplete, Roche) and phosphatase inhibitors (PhosStop, Roche). Lysates were normalized using a BCA protein assay kit (Pierce) and resolved on 8% SDS-PAGE gels.
For Western blotting, antibodies from Cell Signaling Technology were used diluted at 1:1,000 unless otherwise noted: LKB1 (#3047), AMPKα (#2532), pAMPK(T172) (#2535), pACCS79 (#3661), pRAPTOR (S792; #2083), BRSK1 (#5935, 1:500), BRSK2 (#5460), MARK2 (#9118), MARK3 (#9311), MARK4 (#4834, 1:500), pMARK activation loop (#4836), NUAK1 (#4458), SIK2 (#6919), GAPDH (#5174, 1:10,000), NR4A1 (#3960), pHDAC4/5/7(S246; #3443), STAT3 (#9139), STAT1 (#14994S), pSTAT1-Y701 (#9167), pSIK family T-loop antibody (Abcam 199474, 1,000), SIK1: rabbit polyclonal 1-111 (kind gift from Rebecca Berdeaux, Salk Institute), SIK3 (human; Novus Biologicals, NBP2-47278), SIK3 (mouse; Bethyl Laboratories, A302-455A), NUAK2 (Novus Biologicals, NBP1-81880, 1:1,000), SNRK (ab173538, 1:500), CRTC3 (ab91654, 1:1,000), CRTC2 (Calbiochem ST1099, 1:10,000), pCRTC2 (kind gift from Marc Montminy), FLAG-M2-affinity gel (Sigma A2220). For IHC, BrdU (Abcam #6326, 1:200) and Ly-6G (BioLegend, clone 1A8, 1:200).
To validate loss of Sik1 in lung tumors, DNA was isolated from flash-frozen tumor tissue derived from control K+sgTom, KSik1, and KSik1+sgSik3 tumors using the Quick-RNA Miniprep Kit (Zymo Research), and PCR was performed with primers to amplify the recombined Sik1 allele or primers to amplify the floxed allele as described in Nixon and colleagues (32).
mRNA Preparation and qPCR
mRNA from snap-frozen tumor samples was prepared using the Quick-RNA Miniprep Kit (Zymo Research), including DNase treatment. cDNA was synthesized from 2 μg of RNA using SuperScript III (Life Technologies), and qPCR was carried out with diluted cDNA, appropriate primers, and SYBR Green PCR master mix (Thermo Fisher Scientific) using a C1000 Thermal Cycler (Bio-Rad). Relative mRNA levels were calculated using the 2−ΔΔCT method, using actin as an internal control.
Tumor RNA was isolated using the Quick-RNA MiniPrep Kit (Zymo Research), including a DNase treatment. RNA integrity numbers were determined using an Agilent TapeStation prior to library preparation. mRNA-seq libraries were prepared using the TruSeq RNA Library Prep Kit (version 2), according to the manufacturer's instructions (Illumina). Libraries were quantified, pooled, and sequenced by single-end 50 base pairs using the Illumina HiSeq 2500 platform at the Salk Next-Generation Sequencing Core. Raw sequencing data were demultiplexed and converted into FASTQ files using CASAVA (version 1.8.2). Libraries were sequenced at an average depth of 80 million reads per sample.
Deep Sequencing and Bioinformatics Analysis of Cas9 Target Loci
Tumor DNA was obtained by laser-capture microdissection at the Biophotonics Core at the Salk Institute. DNA was isolated from lung histology sections to represent all sgRNA guides used in this study and was prepared with the Arcturus PicoPure DNA Extraction Kit (ABI), with subsequent cleanup with AMPure beads (Agilent). For each targeted gene, a 300-bp amplicon centered around the PAM site was amplified using New England BioLabs Next 2X master mix. Sequencing libraries were prepared from 50–80 ng of tumor PCR product and barcoded with custom index R2 primers (kind gift from Patrick Hsu, Salk Institute). Individual libraries were quantitated after electrophoresis with analytic quantitative ladder, and relative molar amounts of barcoded amplicons were pooled, cleaned up with AMPure beads and gel purified 2×. Final pooled libraries were characterized for size by TapeStation, quantitated by qPCR using KAPA Library Quantification Kits, and sequenced on an Illumina MiSeq machine with 300-bp end-paired reads. MiSeq reads were filtered by requiring an average Phred quality (Q score) of at least 30, as well as perfect sequence matches to barcodes. Data was demultiplexed with a custom script, and pair-end stitching was done with Pear. The python program CRIS.py v2 was applied to stitched data to assess endonuclease activity. CRIS.py output was used to quantify the edit rate, frameshift rate, and average indel size of all clones.
Bioinformatic Analysis for RNA-seq
Sequenced reads were quality-tested using the online FastQC tool (http://www.bioinformatics.babraham.ac.uk/projects/fastqc) and aligned to the mouse mm10 genome using the STAR aligner version 2.4.0k. Raw gene expression was quantified across all annotated exons using HOMER (55), and differential gene expression was carried out using the edgeR package version 3.6.8 using duplicates to compute within-group dispersion and correcting for batch effects (56). Differentially expressed genes were defined as having an FDR < 0.05 and a log2 fold change >0.59. GSEA was carried out with the GenePattern interface (https://genepattern.broadinstitute.org) using preranked lists generated from FDR values, setting gene set permutations to 1,000 and using the Hallmark collection in MSigDB version5.0. Clustering was performed with Gene Cluster 3.0 and visualized by heat maps using Java TreeView (version 1.1.6r4). Area-proportional Venn diagrams were plotted using BioVenn (57). Metascape: http://metascape.org; Enrichr: http://amp.pharm.mssm.edu/Enrichr; Homer motif enrichment analysis: http://homer.ucsd.edu/homer/motif/.
Statistical analyses described in each figure were all performed using GraphPad Prism 7.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
Conception and design: P.E. Hollstein, R.U. Svensson, R.J. Shaw
Development of methodology: P.E. Hollstein, L.J. Eichner
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): P.E. Hollstein, L.J. Eichner, S.N. Brun, A. Kamireddy, R.U. Svensson, L.I. Vera, T.J. Rymoff, H.M. Galvez, R.A. Screaton, R. Berdeaux
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): P.E. Hollstein, L.J. Eichner, S.N. Brun, A. Kamireddy, A.E. Williams, M.N. Shokhirev, R.J. Shaw
Writing, review, and/or revision of the manuscript: P.E. Hollstein, S.N. Brun, R.J. Shaw
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): P.E. Hollstein, L.I. Vera, D.S. Ross, T.J. Rymoff, A. Hutchins
Study supervision: R.J. Shaw
This study was supported by grants to R.J. Shaw from the NIH (R35CA220538, P01CA120964), the Samuel Waxman Cancer Research Foundation, and The Leona M. and Harry B. Helmsley Charitable Trust grant #2012-PG- MED002. NIH grants to R. Berdeaux also supported this work (R01DK092590 and R01AR059847). R.U. Svensson was supported by a postdoctoral fellowship from the American Cancer Society (ACS#124183-PF-13-023-01-CSM). L.J. Eichner is supported by a postdoctoral fellowship from the American Cancer Society (PF-15-037-01-DMC). S.N. Brun was supported by training grant 5T32CA009370 to the Salk Institute Cancer Center and 5F32CA206400. The Salk CCSG P30 CA014195 and Helmsley Charitable Trust supported the Functional Genomics Core and the Bioinformatics Core. We thank Thales Pappagiannakopoulos (New York University) and Francisco Sánchez-Rivera (Memorial Sloan Kettering Cancer Center) for advice regarding pSECC viral preparation. We additionally are grateful to Nasun Ha and the Salk Next-Generation Sequencing Core for helpful discussions; Marc Montminy, Ezra Wiater, and Patrick Hsu for assistance and discussions regarding MiSeq; Jill Meisenhelder and Xavier Miralles Fusté for helpful advice with in vitro kinase reactions; and Jeanine Van Nostrand for help with intratracheal intubations. We thank Matthew Chun for help with transcription analysis in human NSCLC cells and Daniel Garcia for help with graphic design. We thank Chris Murray and Monte Winslow of Stanford for helpful discussions, and members of the Shaw lab for feedback and advice.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.