EIF1AX and RAS Mutations Cooperate to Drive Thyroid Tumorigenesis through ATF4 and c-MYC

Papillary carcinomas (PTC) are the most common type of thyroid cancer. They are usually indolent tumors that harbor mutually exclusive mutations in BRAF, RAS, or fusions of RET, NTRK, or BRAF (1–3). The Cancer Genome Atlas (TCGA) study of PTC identified additional driver alterations present at lower frequency in genes including EIF1AX, PPM1D, and CHEK2 (4). Poorly differentiated thyroid cancers (PDTC) and anaplastic thyroid cancers (ATC) are the most aggressive forms of the disease and are characterized by distinct genomic profiles. Although BRAF and RAS mutations are also the main drivers, as compared with PTC they are more frequently associated with mutations of the TERT promoter, TP53, genes encoding PI3K/AKT/mTOR pathway effectors, or chromatin modifiers (5–7). They are also markedly enriched for EIF1AX mutations.

Translation initiation in higher eukaryotes is orchestrated by the tight regulation of the cap binding and the 43S preinitiation complexes (PIC). Formation of the PIC involves recruitment of the ternary complex [TC; EIF2-GTP-tRNAi(Met)] onto the 40S ribosomal subunit. The PIC component EIF1A is encoded on human chromosomes X and Y by EIF1AX and EIF1AY, respectively. Their protein products are highly conserved, and expression of EIF1A is biallelic irrespective of gender (8, 9). EIF1A is required for recruitment of the TC and for assembling the 43S PIC (10), which after recruitment onto capped mRNAs scans their 5′ untranslated region (UTR) and localizes the AUG to initiate translation (11, 12). Deregulation of translation initiation is common in tumorigenesis. Increased expression of components of the EIF4F cap binding complex (EIF4E, EIF4A, and EIF4G) is seen in many cancers. The expression of these genes is under transcriptional control by c-MYC (13, 14). EIF4E is limiting for translating the mammalian genome and is frequently found in excess in cancer cells, where it may help drive a translational output supporting tumorigenesis (15).

To our knowledge, EIF1AX is the only example of a PIC subunit recurrently mutated in cancer. Mutations of EIF1AX were first reported in uveal melanomas (8). Comprehensive genomic profiling of these tumors revealed that EIF1AX-mutant tumors mark a comparatively low risk form of the disease. EIF1AX and GNA11/GNAQ mutations frequently co-occurred and were mutually exclusive with c-MYC amplification (16). EIF1AX mutations have been reported in benign thyroid adenomas (17) and follicular carcinomas (18), as well as in ∼1% of PTC in a mutually exclusive manner with other drivers (4). By contrast, they are present in 11% of PDTC and ATC and are almost invariably associated with oncogenic RAS (5, 7). The striking coevolution of EIF1AX and RAS mutations in advanced disease suggests that they may cooperate to drive tumor progression.

The core RNA binding domain of EIF1A is universally conserved from archaea to eukaryotes, whereas eukaryotes differ from bacteria by the addition of the unstructured amino-terminal [N-terminal tail (NTT)] and carboxy-terminal tails (CTT; refs. 19, 20). EIF1AX mutations identified in several cancers encode somatic substitutions in the first 2 to 15 amino acids (AA) of the NTT (8, 21, 22), whereas thyroid cancers harbor an additional hotspot splice-site mutation (EIF1AX-A113splice) in the CTT that is private to this disease (4, 5, 7). EIF1AX mutations are always heterozygous, suggesting that full occupancy of PICs by mutant EIF1AX would be detrimental to viability. Structure-function studies in yeast revealed that mutating any of the NTT residues between 7 and 16 AAs were lethal and resulted in leaky scanning of the initiation AUG codon (23, 24), and discriminated against AUGs with poor context (25). By contrast, CTT mutants enhanced noncanonical AUG initiations. Of note, the experimental CTT substitutions tested in yeast did not appropriately model the structural defects of the EIF1AX splice-site mutation (A113splice).

Here, we describe the identification of the key signaling drivers of transformation by EIF1AX mutants, particularly EIF1AX-A113splice, alone and in the context of RAS, and the therapeutic dependencies they confer.

Analysis of our institutional clinical genomics database of 148 advanced thyroid cancers coupled to data from two previously published studies (5, 7) showed that 26 of 246 (11%) tumors harbored EIF1AX mutations, 25 of which were associated with mutant RAS (25/26; P = 3.15 × 10E⁻¹³; Fig. 1A). The EIF1AX mutations clustered within the first 15 AAs of the NTT, as reported in uveal melanomas (8), or more frequently at a hotspot splice acceptor site upstream of exon 6 (A113splice) in the CTT (17/26; Fig. 1A). The A113splice mutation, not seen so far in any other cancer type, abolished the acceptor site of exon 6, resulting in two alternatively spliced transcripts (Fig. 1B): (i) c'splice (c'spl), through usage of a cryptic site within exon 6, yielding a 132AA protein through an in-frame exclusion of 12 AA. (ii) t'splice (t'spl), which retains intron 5, resulting in a 115AA-truncated protein. We confirmed the presence of these alternatively spliced mRNAs in the RNA-sequencing (RNA-seq) data of A113splice-mutant PTCs from the TCGA study (ref. 4; Supplementary Fig. S1A). Western blotting of HTH83 and C643 thyroid cancer cell lines harboring the A113splice mutation showed EIF1AX protein products corresponding to c'spl and t'spl mRNAs, with c'spl as the predominantly expressed isoform (Fig. 1C). Their predicted AA sequences are shown in Supplementary Fig. S1B.

To explore the biological consequences of the EIF1AX-A113splice mutation, either alone or in the context of mutant RAS, we generated isogenic thyroid cancer cell lines by CRISPR/Cas9 knock-in of heterozygous mutations of A113splice into KRAS^G12R-mutant CAL62 or RAS wild-type (WT) TTA1 cells, and by reversing the endogenous A113splice mutation in HRAS^G13R and EIF1AX^A113spl-mutant C643 cells (Fig. 2A; Supplementary Fig. S2A and S2B). Introduction of A113splice into CAL62 or TTA1 cells markedly increased their colony formation in soft agar, whereas transformation efficiency was decreased by editing out A113splice in C643 cells (Fig. 2A). We then tested the effects of the two EIF1AX-A113splice products, c'spl and t'spl, independently and in combination on transformation of Nthy-Ori 3-1 cells, an SV40 large-T-antigen–immortalized human thyroid cell line. The c'spl product markedly increased colony formation, whereas the t'spl did not, and dampened the effects of c'spl when both were coexpressed through a bicistronic vector, consistent with EIF1AX-c'spl being the functionally active variant (Supplementary Fig. S2C).

We next investigated the interaction of oncogenic HRAS and EIF1AX-c'spl in vivo. For this, we generated mice with thyroid-specific, doxycycline (dox)-inducible expression of EIF1AX-c'spl by targeting a TRE-EIF1AX-c'spl construct into the mouse ColA1 locus and crossing the resulting animals with Tg-rtTA mice (Fig. 2B). Dox-fed Tg-rtTA;TRE-EIF1AX-c'spl mice expressed the c'spl protein in thyroid tissue (Fig. 2C), which developed thyrocyte hyperplasia with atypical features (18/19), with one animal developing a low-grade classic PTC (Fig. 2D and E). These findings closely phenocopy the human thyroid histologies associated with an isolated EIF1AX mutation (4, 17). TPO-Cre;FR-Hras^G12V mice express Hras^G12V upon Cre-induced recombination in the thyroid (Fig. 2B), which is insufficient to drive tumorigenesis (26). By contrast, the quadri-transgenics (TPO-Cre;FR-Hras^G12V;Tg-rtTA;TRE-EIF1AX-c'spl) displayed neoplasms along the spectrum of disease progression with a penetrance of 30%, including Hurthle cell adenoma, PTC, and PDTC (Fig. 2D and E; Supplementary Fig. S2D), consistent with the histologic characteristics of human thyroid tumors harboring the combined genetic lesions (17).

EIF1AX is an essential subunit of the translation PIC (10, 12). We performed coimmunoprecipitation (co-IP) experiments to probe for possible aberrant interactions of EIF1AX mutants with components of the TC and the PIC. IP of HEK293T cells expressing HA-tagged WT EIF1AX, NTT mutants (G8R, G9R, G15V), or EIF1AX-c'spl with an antibody to HA showed pulldown of the TC component eukaryotic initiation factor 2α (EIF2α) by all EIF1AX proteins, with EIF1AX-c'spl showing greater affinity (Fig. 3A). IP of the HEK293T lysates with EIF5, a component of the PIC, did not detect EIF1AX-WT in the immunoprecipitate, likely because of the known labile interactions between these PIC subunits (10).By contrast, EIF1AX mutants, particularly G8R, G9R, and c'spl, exhibited increased affinity for EIF5 (Fig. 3B), consistent with stabilization of the PIC. This was confirmed in the isogenic EIF1AX-splice cell lines (Fig. 3C), and in thyroid cancer cells with endogenous EIF1AX mutations (Supplementary Fig. S3A). These data suggest that EIF1AX mutants result in a more stable 43S ribosomal complex. As translation initiation is a rate-limiting process, we next tested whether the EIF1AX mutants altered nascent protein synthesis in the isogenic lines. L-azidohomoalanine (AHA)-labeled proteins were markedly increased by knock-in of the A113splice mutation into RAS WT (TTA1) or mutant cells (CAL62), whereas reversion of the mutation in C643-spl-rev cells had the opposite effect (Fig. 3D and E). NthyOri cells stably expressing EIF1AX-G8R, G9R, or EIF1AX-c'spl also showed an increase in nascent protein synthesis compared with WT, with c'spl having the greatest effect (Fig. 3D and E). The increased protein synthesis by c'spl is comparable with that induced by EIF4E overexpression (Supplementary Fig. S3B) and is blocked by mTOR kinase inhibition (Supplementary Fig. S3C).

To determine whether the effects on protein synthesis were global or selective, and the candidate mechanisms involved, we performed low-pass ribosome footprinting in C643 cells to identify subsets of mRNAs that were translated with greater (TE-high) or lesser (TE-low) efficiency than its isogenic splice-reversed control (Supplementary Tables S1 and S2). Interestingly, ATF4, a known translationally regulated gene, scored as a preferentially translated candidate (log₂ fold change = 0.71; P = 0.005; P_adj: 0.09; Supplementary Table S1). Accordingly, polysome profiling by density gradient fractionation showed that ATF4 mRNA was enriched in actively translating polysome fractions of NthyOri cells expressing EIF1AX splice (Supplementary Fig. S4A). Additionally, Gene Set Enrichment Analysis (GSEA) of RNA-seq of CAL62-splice versus CAL62 cells found an ATF4 activation signature (NES: 1.7; Nom P = 0.012; Supplementary Table S3). ATF4 is a key transcription factor that integrates responses to cell stressors, such as AA deficiency or endoplasmic reticulum (ER) protein folding defects. Despite the increase in ATF4, expression of the EIF1AX-splice variants was not associated with significant activation of the ER stress pathway (Supplementary Fig. S4B). ATF4 mRNA contains two upstream open reading frames (uORF) that determine the efficiency of its translation (27). The second uORF (uORF2) overlaps with the canonical ATF4 start codon and is a strong inhibitor of ATF4 translation. Under normal conditions, translation starts at uORF1, and the ribosome dissociates at the stop codon. It reassembles at uORF2, which prevents ATF4 translation. During cellular stress, EIF2α is phosphorylated at serine 51 by stress-sensing kinases. As a result, it remains GDP-bound, dampening formation of the TC (EIF2α-GTP-Met-tRNA). When TC availability is limited, the ribosome fails to reassemble at uORF2 and instead reinitiates translation at the canonical ATF4 start codon (28). As EIF1AX is known to affect the fidelity of start codon selection (25, 29), we tested whether EIF1AX-splice preferentially translates ATF4 by altering selectivity toward the two upstream (⁻³ACCAUG/⁻³GCCAUG) and the main (⁻³AACAUG) ATF4 start codons. For this, we engineered reporter constructs in which the firefly luciferase protein was under control of the different ATF4 translation initiation contexts (Kozak + start codons). Expression of EIF1AX-WT or EIF1AX-c'spl in HEK293T cells cotransfected with appropriate reporters showed that the inhibitory uORF2 led to less efficient translation in EIF1AX mutant–expressing cells (Supplementary Fig. S4C). This would conceivably derepress translation initiation at the ATF4 main ORF (mORF). Interestingly, reporter activity under control of the ATF4 mORF was also markedly increased in EIF1AX-c'spl cells (Supplementary Fig. S4C). As a complementary strategy, we used a Translation Control Reporter System (TCRS; ref. 30) to test whether increased efficiency of ribosomal reinitiation might explain the effects of the mutant EIF1AX on ATF4 translation. Similar to ATF4, the TCRS construct has 3 ORFs: a short uORF, followed by 2 overlapping ORFs encoding a long peptide (LP) and a short peptide (SP), respectively, the latter serving as a marker for ribosome reinitiation (Supplementary Fig. S4D). Expression of EIF1AX NTT or c'spl mutants in HEK293T cells cotransfected with TCRS showed higher SP levels as compared with EIF1AX-WT, indicative of higher efficiency of ribosome reinitiation. ATF4 is believed to feed forward to induce expression of its own transcript (31). Consistent with this, ATF4 mRNA was induced by ∼3-fold in all EIF1AX-splice isogenic contexts (Supplementary Fig. S4E). The coordinate robust increase of ATF4 gene expression may have dampened the sensitivity of the ribosome profiling experiments.

Serine 51 phosphorylation of EIF2α in response to cellular stress represses global translation, but increases translational efficiency of ATF4 (27). Therefore, under physiologic conditions, ATF4 is downstream of pEIF2α. By contrast, EIF2α in EIF1AX-splice mutant cells is paradoxically underphosphorylated. Hence, EIF1AX-splice co-opts this pathway by constitutively activating expression of ATF4, placing it upstream of EIF2α, increasing availability of the TC and derepressing global translation (Fig. 4A). This is consistent with ATF4 dephosphorylating pS51-EIF2α through a negative feedback loop engaged via ATF4-dependent upregulation of GADD34 (Fig. 4B), an EIF2α-specific cofactor for protein phosphatase-1 (PP1; ref. 32). Moreover, the GADD34/PP1 phosphatase inhibitor salubrinal blocked EIF2α dephosphorylation and preferentially repressed global protein synthesis in EIF1AX-splice versus splice-reverted C643 cells (Fig. 4C).

We performed GSEA to identify the oncogenic signaling pathways activated by EIF1AX splice. In addition to ATF4, the top-ranked signatures enriched in EIF1AX-mutant cells in the RNA-seq profiles of the isogenic CAL62 and C643 models included the following terms: translation, ternary, and 43S complex formation, mTORC1 signaling, and transcriptional targets of c-MYC (Fig. 5A; Supplementary Tables S3 and S4). Accordingly, the mTOR substrates p70-S6 kinase and 4EBP1 were activated by expression of the EIF1AX-splice products in both RAS-WT and RAS-mutant cell lines (Fig. 5B). The EIF1AX-mutant induction of mTOR signaling was not associated with PI3K pathway activation in RAS-WT cells, whereas AKT and PRAS40 phosphorylation was increased in RAS-mutant cells. Despite this, the activation of mTOR by the aberrant EIF1AX gene products was neither PI3K- nor RSK-dependent, as treatment with the pan-PI3K inhibitor GDC0941, the pan-AKT kinase inhibitor MK-2206, or the pan-RSK inhibitor LJI308 did not impair the induction of p4EBP1 in the parental C643 compared with splice-reverted cells (Supplementary Fig. S5A and S5B).

On the basis of these findings, we hypothesized that c-MYC and ATF4 could be key oncogenic clients of EIF1AX. Accordingly, c-MYC and ATF4 protein levels were higher in the isogenic lines and in NthyOri cells expressing the EIF1AX-splice variants (Fig. 4B), as well as in mice with dox-inducible thyroid-specific expression of EIF1AX-c'spl (Fig. 5C). The increase in ATF4 and c-MYC was associated with greater abundance of AA transporters for glutamine (ASCT2) and leucine (LAT1), which are known to be regulated by these transcription factors (refs. 33, 34; Fig. 4B). The differential expression of these transporters in RAS-mutant thyroid cancer cell lines with endogenous EIF1AX mutations compared with those that were EIF1AX-WT was particularly striking (Supplementary Fig. S5C).

We next explored whether ATF4 and/or c-MYC accounted for the increased expression of the AA transporters in cells expressing EIF1AX splice. Silencing of ATF4 or c-MYC alone modestly decreased ASCT2 abundance in C643 cells, with minimal effects on LAT1. However, combined ATF4/c-MYC knockdown repressed both transporters and decreased p4EBP1 (Fig. 5D). This was also the case in NthyOri-splice cells (Fig. 5E). The increased expression of ASCT2 and LAT1 in EIF1AX-splice–expressing cells could induce mTOR activation through increased influx of glutamine and leucine (34, 35). Consistent with this, addition of leucine and glutamine in combination after 3 hours of AA depletion resulted in a more rapid and robust induction of p4EBP1 in C643 cells compared with their isogenic WT revertants (Fig. 5F). The mTOR pathway in EIF1AX-splice cells was also more sensitive to depletion of glutamine (Supplementary Fig. S5D).

Deregulated expression of c-MYC in cancer is commonly due to increased protein stability. Indeed, the higher c-MYC protein levels in EIF1AX-splice–expressing cells are not associated with induction of c-MYC mRNA (Fig. 6A) or increased translational efficiency (Supplementary Fig. S6A). Instead, expression of EIF1AX splice in KRAS^G12R-CAL62 shifted the half-life of c-MYC from 25 to 60 minutes as compared with the isogenic parental cells (Fig. 6B). Silencing of oncogenic KRAS also decreased c-MYC protein levels (Fig. 6C), although to a lesser extent than in EIF1AX-splice cells, consistent with the latter cooperating with oncogenic RAS to further stabilize the protein. The key contribution of RAS and MAPK signaling to c-MYC levels was further confirmed in HTH83 cells, which harbor endogenous HRAS^Q61R and EIF1AX-A113splice mutations. In these cells, HRAS silencing or MEK inhibition decreased c-MYC and ASCT2 expression, whereas the pan-PI3K inhibitor GDC-0941 was without effect (Supplementary Fig. S6B and S6C).

To explore potential therapeutic dependencies of RAS + EIF1AX-mutant thyroid cancers, we investigated the effects of the MEK inhibitor trametinib, the mTOR kinase inhibitor AZD8055, or the BRD4 inhibitor JQ1 (to target c-MYC transcription, with the caveat that JQ1 also inhibits other bromodomain proteins) alone and in various combinations in xenografts of CAL62-splice and parental cells. CAL62-splice xenografts grew to a larger size and were more sensitive to the growth-inhibitory effects of AZD8055 or JQ1 than the parental controls, whereas trametinib had equivalent efficacy in both contexts (Fig. 6D; i–iii). The combination of AZD8055 with either trametinib or JQ1 induced tumor shrinkage in CAL62-splice but not in parental cells and was superior to either drug alone (Fig. 6D; iv–vi). Consistent with their effects on growth, western blots of tumor lysates from mice treated with each condition showed that AZD8055 in combination with either trametinib or JQ1 showed the most profound inhibition of c-MYC and mTOR substrates (Fig. 6E). However, despite comparable inhibition of these signaling nodes in EIF1AX WT and mutant cells, the latter show preferential tumor shrinkage, consistent with heightened dependency on the pathways activated by these mutant proteins.

EIF1AX is the only PIC component that is recurrently mutated in cancers (4, 5, 8, 21, 22). EIF1AX mutations have been presumed to result in a change-of-function or gain-of-function because of their predilection for specific substitutions in the N- and C-terminal tails. However, functional insights so far have been primarily confined to how EIF1AX mutants alter usage of initiation codons with varying contexts in yeast (25, 29). In uveal melanomas, EIF1AX-NTT mutants are associated with relatively indolent disease. Isolated EIF1AX mutations are also found in low-risk thyroid tumors (4, 17). When coupled to RAS mutations, they mark aggressive and often lethal forms of PDTC and ATC. This is phenocopied in the mouse models we described here. However, the penetrance of these cancers in EIF1AX-c'spl/Hras^G12V mice was relatively low, suggesting that other factors may be required for transformation. In that respect, most human EIF1AX/RAS PDTCs and ATCs harbored either TERT promoter or TP53 mutations, which are major drivers of tumor progression in this disease (5, 6, 36).

Phosphorylation of Ser51 of the α subunit of EIF2 is a central common conduit for many cellular stress pathways, including nutrient/AA starvation, ER stress, and oxidative insults. This modification of EIF2α prevents its recycling into the TC, thereby inhibiting global protein synthesis, which helps cells to adapt by conserving nutrients and relieving ER stress. However, a subset of mRNAs, most prominently ATF4, are preferentially translated. ATF4 induces a transcriptional program that includes genes involved in AA transport, metabolism, and protection from oxidative stress, which allow cells to orchestrate a more sustained adaptation to these stressors (37–43). The induction of ATF4 in EIF1AX-A113splice cells is independent of EIF2α phosphorylation and takes place by modulating ATF4 translation efficiency as well as inducing its transcription. This illegitimate ATF4 activation then hijacks a negative feedback pathway that leads to dephosphorylation of Ser51 of EIF2α, by inducing expression of GADD34, a cofactor for the EIF2α phosphatase PP1 (32, 43), thus increasing global protein synthesis. Hence, this mutated translation initiation component attains a gain-of-function by deregulating the stringent control of the rate of global protein synthesis by EIF2α dephosphorylation. This in itself has significant oncogenic potential, as expression of hypophosphorylated EIF2α is sufficient to transform NIH3T3 cells (44).

Ras mutations are found along the entire spectrum of thyroid cancer, although the frequency is markedly enriched in PDTC and ATC (5). Allelic imbalance favoring oncogenic Ras gene dosage appears to be critical for transformation and can be achieved through loss of the WT copy (45) or mutant allele amplification (46–48). Intensification of RAS signaling in thyroid cancer can also be mediated through YAP-induced transcriptional activation of RAS, leading to tumor progression (26). Oncogenic RAS acts via ERK to prevent c-MYC degradation, primarily through phosphorylation of serine 62, a site within the c-MYC degron recognized by a specific E3 ubiquitin ligase (49, 50). pS62–c-MYC primes subsequent phosphorylation at threonine 58, which facilitates dephosphorylation of S62, poising c-MYC for degradation. We found that EIF1AX-splice increased c-MYC protein half-life, further augmenting the effects of oncogenic RAS signaling on c-MYC protein stability. Interestingly, transcription-independent MYC-MAX activation was seen in the uveal melanoma TCGA study in the context of EIF1AX-NTT mutations (16). Consistent with this, the EIF1AX mutations in uveal melanoma display mutual exclusivity to tumors with chromosome 8q gain, which harbors the MYC gene locus (8q^24.21). Conceivably, by analogy to the RAS-EIF1AX cooperativity on c-MYC in thyroid cancer, a potential mechanistic basis of MYC-MAX activation in uveal melanomas may involve interactions with constitutive G-protein oncogenic signaling mediated by the co-occurring GNAQ/GNA11 mutations in EIF1AX-mutant uveal melanomas (8, 16).

The intersection of ATF4 and c-MYC on regulation of AA transporter expression and mTOR activation is well established (33–35). We found that c-MYC and ATF4 coregulated AA transporters in the EIF1AX-RAS context, particularly ASCT2 and LAT1, leading to mTOR activation. Accordingly, expression of ATF4, c-MYC, and AA transporters in a panel of RAS-mutant human thyroid cancer cell lines displayed a striking concordance with EIF1AX mutation. These events cooperate to sensitize mTOR signaling to AA supply. EIF1AX-splice also slightly augments RAS-induced PI3K signaling in the isogenic cell lines, through unclear mechanisms. However, the activation of mTOR is PI3K–AKT- and RSK-independent and driven primarily by the increased influx of AAs. Whether increased PI3K contributes to tumorigenesis through alternative mechanisms in this context cannot be ruled out.

In summary, EIF1AX-A113splice, a mutation commonly encountered in thyroid cancer in association with oncogenic RAS, leads to induction of ATF4, which in turn induces a global increase in protein synthesis through GADD34-dependent dephosphorylation of EIF2α. The mutant EIF1AX, in concert with oncogenic RAS, also increases c-MYC protein stability. c-MYC and ATF4 cooperate to induce transcription of AA transporters, and the resulting AA flux activates mTOR signaling. Our data point to mTOR kinase as a primary node for pharmacologic targeting and provide a rationale for MEK or c-MYC coinhibition to maximize therapeutic responses (Fig. 7).

Besides the ∼11% of advanced thyroid cancers harboring EIF1AX and RAS mutations, these two oncogenes also co-occur in low-grade serous ovarian cancers (21) and in some widely invasive Hurthle cell carcinomas. mTOR kinase inhibitors are currently not approved for any indication. However, they are still being extensively investigated in combination with other agents. The findings reported here provide a strong rationale for combined mTOR and MEK inhibitors for tumors harboring these defects.

Cell lines were maintained at 37°C and 5% CO₂ in humidified atmosphere and were grown in RPMI-1640 for Nthy-Ori 3-1, C643, CAL62, and HTH83; DMEM for HEK293T, HTH7, and ACT1; and DMEM:RPMI for KMH2 supplemented with 10% of FBS, 2 mmoL/L glutamine, 50 U/mL penicillin (GIBCO), and 50 μg/mL streptomycin. C643, HTH7, and HTH83 cell lines were obtained from Dr. Nils-Erik Heldin, Uppsala University Hospital, Sweden. CAL62 cells were obtained from Dr. Jeanine Gioanni, Centre Antoine-Lacassagne, France. The ACT1 line was obtained from Dr. Naoyoshi Onoda, Osaka City University Graduate School of Medicine, Japan. KMH2 were obtained from the Japanese Collection of Research Bioresources Cell Bank. All cell lines tested negative for Mycoplasma and were authenticated using short tandem repeat and single-nucleotide polymorphism analyses.

The cDNAs of the EIF1AX-splice variants (c'spl and t'spl) were cloned by PCR amplification from parental EIF1AX-A113splice mutant human thyroid cell line HTH83. The full-length cDNAs of human WT EIF1AX and the EIF1AX-splice variants (c'spl and t'spl) were cloned into pLVX-puro and the dox-inducible pLVX-Tight-puro vectors (Clonetech). EIF1AX-NTT mutants were generated from pLVX-puro-EIF1AX-WT by site-directed mutagenesis (Strategene protocol). The EIF1AX c'spl cDNA was cloned into the pLVX-puro vector or the pLVX-Tight-puro vector. Bicistronic expression constructs for EIF1AX-WT and the two EIF1AX-splice variants were generated by sequential cloning into pLVX-Tight-puro, as schematically shown in Supplementary Fig. S2C. The PCR primers used to amplify EIF1AX-WT, to generate NTT mutants by site-directed mutagenesis, or to amplify the specific splice variants in cell lines harboring endogenous EIF1AX mutations are shown in Supplementary Table S5. The TCRS and the TCRS^ΔuORF vector systems were provided by Dr. Cor Calkhoven (ERIBA, Groningen, the Netherlands). The targeting plasmid ColA1-TRE (ColA-CHC system for cDNA expression) used to clone the EIF1AX-c'spl cDNA was provided by Dr. Luke Dow (Weill Cornell College of Medicine, New York, NY). All constructs were sequence verified.

CAL62 and TTA1 cells, which are WT for EIF1AX, were modified to endogenously express EIF1AX-splice variants (cryptic and truncated splice) by targeted disruption of the splice acceptor site by CRISPR/Cas9. A 20-bp CRISPR guide sequence (sgRNA) was designed to span the splice site of exon 6, exploiting the endogenous PAM sequence within the splice acceptor site (Supplementary Table S5). The sgRNA was annealed and cloned into the pLentiCRISPR vector (Addgene # 49535) that transcribes Cas9 and the CRISPR guide/tracer RNA. Parental CAL62 and TTA1 cells were transfected with pLentiCRISPR-SgRNA using FuGENE HD, followed by selection in 1 μg/mL puromycin for 3 days. Cells were then grown for 5 to 7 days and then plated as single cells into 96-well plates. The propagated clones were tested for disruption of the EIF1AX exon 6 splice site by PCR-based sequencing of genomic DNA encompassing the targeted region. The substitutions introduced in CAL62 and TTA1 cells effectively disrupted the splice acceptor site but were not identical to the naturally occurring endogenous EIF1AX-A113spl mutations. Positive clones of CAL62 splice and TTA1 splice were confirmed for expression of EIF1AX c'splice and t'splice mRNAs and their encoded proteins by immunoblotting.

The parental C643 cell line, which harbors an endogenous EIF1AX-A113splice mutation, was used to revert the mutant allele by CRISPR/Cas9 knock-in of the corresponding WT sequence through use of a homologous-directed repair (HDR) template. The sgRNA targeting intron 5 was cloned into the pX330 vector (Addgene# 42230). The homologous recombination donor vector was designed using vector builder (https://www.vectorbuilder.com/design.html). Specifically, the donor vector (Vector ID: VB160426-1022eqv) was designed to integrate into the intron 5 locus upon recombination directed by homologous arms of 629 bp (right) and 701 bp (left) flanking a puromycin-resistance cassette. The two arms encompassed the 3′ end of intron 5, the entire exon 6, and a fragment of intron 6. C643 cells were cotransfected with pX330-sgRNA and the donor vector harboring the HDR template. After 72 hours, cells were subjected to puromycin selection. Surviving cells were seeded as single cells in a 96-well plate for testing. Clones that were positive for the targeted locus by PCR-based sequencing were tested for expression of the splice variant mRNAs by RT-PCR and by immunoblotting. Clone C643-spl-rev (C643-Xcl18) was used in this article.

Nthy-Ori 3-1 cells (hereafter referred to as NthyOri; derived from WT human thyroid cells immortalized with large T-antigen; ref. 51) were used to generate stable and dox-inducible lines expressing EIF1AX-WT, NTT mutants or the splice variants. The pLVX-puro vector cloned with the respective various EIF1AX cDNAs were used for constitutive expression, whereas the pLVX-Tet-On Advanced vector system (Clontech) was used to generate dox-inducible EIF1AX or bicistronically expressed EIF1AX-splice variants, as described in the Plasmids and Constructs section.

To generate stable EIF1AX-expressing NthyOri clones, the constructs described above were used for lentiviral production in HEK293FT cells using the Mission Lentiviral Packing Mix (Sigma). The constitutively expressing NthyOri-EIF1AX stable lines were generated by infecting with the corresponding viral particles, and the dox-inducible NthyOri-EIF1AX cells by coinfecting with lentiviral-transduced pLVX-Tight-puro-EIF1AX and pLVX-Tet-On particles in the presence of 8 μg/mL polybrene (Sigma) overnight. After 24 hours in complete medium, cells were selected in 1 μg/mL puromycin with or without 300 μg/mL G418 as required. The mass culture was then cloned into single cells, which were expanded and tested for the expression or induction of EIF1AX by immunoblotting.

The dox-inducible EIF1AX-c'spl mouse model was developed by engineering mouse embryonic stem cells (ESC) as previously reported (52). ESCs derived from TPO-Cre;FR-Hras^het;RIK^het;CHC^het mice were used to target the human EIF1AX-c'spl cDNA into the homing cassette that directs site-specific integration of the transgene downstream of the Col1a1 locus by recombination-mediated cassette exchange. The TPO-Cre drives Cre recombinase under the control of the thyroid peroxidase gene promoter, which is active in thyroid cells at E14.5 (53). ESC clones targeted with TRE-EIF1AX-c'spl into the Cola1-homing cassette (CHC) were microinjected into blastocysts produced from NCI C57BL/6-cBrd/cBrd/Cr (C57BL/6 albino) mice and implanted into CD-1 pseudopregnant mothers enabling production of chimeric pups. To achieve higher expression of rtTA, we bred out the RIK^het cassette and bred in thyroglobulin-driven rtTA (Tg-rtTA(s)M2; ref. 54). The resulting TPO-Cre;FR-Hras^het;Tg-rtTA;TRE-EIF1AX^het mice were intercrossed to generate the following lines: TPO-Cre;FR-Hras^hom,Tg-rtTA;TRE-EIF1AX^hom, and TPO-Cre;FR-Hras^hom;Tg-rtTA;TRE-EIF1AX^hom. All animals were fed doxycycline-impregnated chow (TD01306, Envigo), and the appropriate lines were verified by immunoblotting to achieve dox-inducible expression of EIF1AX-c'spl in thyroid follicular cells (Fig. 2C). Animal care and all procedures were approved by the Memorial Sloan Kettering Cancer Center (MSKCC) Institutional Animal Care and Use Committee.

Mouse thyroids dissected from surrounding tissues were fixed in 4% paraformaldehyde, embedded in paraffin, sectioned, and stained with hematoxylin and eosin (H&E). Histologic diagnosis was performed by a thyroid pathologist (R.A. Ghossein) blinded to mouse genotype. Sections were also immunostained for Ki67. Slides were scanned with Pannoramic Flash 250 (3DHistech), and whole thyroid lobes or regions of interest were viewed using Pannoramic Viewer and exported as tiff images. H&E and IHC were performed by the MSK Molecular Cytology Core Facility.

Cells were lysed in 1× RIPA buffer (Millipore) supplemented with protease (Roche) and phosphatase inhibitor cocktails I and II (Sigma). Tumors or xenografts were homogenized in 1× Lysis Buffer (containing 10 mmol Tris-HCl, 5 mmol EDTA, 4 mmol EGTA, 1% Triton-X100) with protease/phosphatase inhibitors. Lysates were briefly sonicated to disrupt the tissue and cleared by centrifugation. Protein concentrations were estimated by BCA kit (Thermo Scientific) on a microplate reader (SpectraMax M5); comparable amounts of proteins were subjected to SDS-PAGE using NuPAGE 4%–12% Bis–Tris gradient gels (Invitrogen) and were then transferred to PVDF membranes. Following overnight primary antibody incubation, membranes were incubated with secondary antibodies coupled to horseradish peroxidase (HRP) or IRDye fluorophores for 1 hour at room temperature. HRP probed blots were developed using enhanced chemiluminescence reagent (Amersham Biosciences), and signal was captured using X-ray films or with the KwikQuant Imager (http://kindlebio.com/index.php). IRDye-probed blots were imaged using the LI-COR Odyssey imaging system (LI-COR Biosciences). Immunoblot quantification was done using ImageJ.

For coimmunoprecipitation experiments, cells were lysed in buffer containing 75 mmol/L NaCl, 50 mmol/L Hepes, 10 mmol/L MgCl2, 8 mmol/L EGTA, 10 mmol/L β-glycerophosphate, 1 mmol/L DTT, and 0.5% Triton X-100, along with protease and phosphatase inhibitors. Equal amounts of lysate (500 μg) were diluted with lysis buffer in 300 μL final volume; 1/10th of the volume was denatured and used as input control. Antibodies were incubated with lysates overnight by end-to-end rotation at 4°C. Antigen–antibody complexes were immobilized by incubating with Dyna beads for 1 hour, the antibody–antigen beads were collected by DynaMag-spin (Invitrogen), washed with lysis buffer, denatured, and subjected to western blotting.

The following primary antibodies were used for immunoblots at a dilution of 1:1,000, except where indicated. EIF1AY (sc-84243) was used to immunoblot EIF1AX, EIF5 (sc-28309), EIF2α (sc-11386), HRAS (sc-520; 1:500), KRAS (sc-30; 1:500), YB1 (sc-101198), and HA-tag (sc-805) were obtained from Santa Cruz biotechnology. c-MYC (5605), ATF4 (11815), LAT1 (5347), ASCT2 (5100), pEIF2α-S51 (9721), pAKT-S473 (4051), pAKT-T308 (4056), AKT (2920), pP70S6K-T389 (9234), P70S6K (2708), p4EBP1-S65 (9451), p4EBP1-T37/46 (2855), 4EBP1 (9452), pYB1-S102 (2900), pRPS6-S240/244 (2215), pRPS6-S235/236 (2211), RPS6 (2317), pERK (9101), ERK (4696), HA-Tag (3724), Biotin (5597), ER stress antibody sampler kit (9956) were obtained from Cell Signaling Technology; pPRAS40-T246 (441100G) and GADD34 (PA1-139) from Invitrogen; ASCT2 (HPA035240) and β-actin (A2228; 1: 10,000) from Sigma-Aldrich.

The secondary antibodies were used at a dilution of 1:5,000. HRP-conjugated antibodies included goat anti-rabbit (Santa Cruz; sc-2004) and goat anti-mouse (Santa Cruz; sc-2031). IRDye fluorophore-conjugated antibodies were IRDye 800CW Goat anti-Rabbit IgG (LI-COR; 926-32211), IRDye 800CW Goat anti-Mouse IgG (LI-COR; 926-32210), IRDye 680RD Goat anti-Rabbit IgG (LI-COR; 926-68071), and IRDye 680RD Goat anti-Mouse IgG (LI-COR; 926-68070). We also used the following additional reagents: Doxycycline (2 μg/mL) and cycloheximide were from Sigma. Salubrinal was from Calbiochem. LJI308 was from Selleckchem.

Isogenic cell lines used to assess nascent protein synthetic rate were grown in 60-mm dishes until ∼70% confluent. Prior to labeling, cells were incubated with methionine-free media containing 2% FBS for 1 hour, followed by the addition of 50 μm AHA (Life Technologies, C10102) for 20 minutes. Cells were then washed in cold PBS and immediately lysed in buffer containing 50 mmol/L Tris-HCl, pH 8.0, and 1% SDS. Complete lysis was achieved by sonication. Comparable amounts of lysates were then subjected to Click-iT reaction for switching azido-modified nascent proteins to alkyne-biotin (Life Technologies, B10185) using the Click-iT Protein Reaction Buffer Kit (Life Technologies, C10276) following the manufacturer's protocol. Biotinylated nascent proteins were subjected to immunoblotting using either anti-biotin, EIF1AX, or β-actin primary antibodies and the corresponding IRDye-fluoropore–conjugated secondary antibodies. Images were captured by the LI-COR Odyssey imaging system. Biotinylated proteins on the entire lane were quantified using Odyssey application software version 3.0 (LI-COR Biosciences).

We used SMARTpools (Dharmacon) for ATF4 and c-MYC silencing (Dharmacon). Cells grown without antibiotics at 70% confluence were transfected using 50 nmol/L of SMARTpools and 6–8 μL of DharmaFECT (Dharmacon). siRNAs for HRAS and KRAS were from ORIGENE and were transfected using SiTran1.0 as per the manufacturer's instructions. Cells were harvested 72 hours after transfection and analyzed by western blotting.

Total RNA from isogenic cell lines was extracted using the PrepEase Kit (USB Corporation). Comparable amounts of RNA (1 μg) were subjected to DNase I (Invitrogen) treatment and reverse transcribed using SuperScript III Reverse Transcriptase (Invitrogen) following the manufacturer's protocol. qPCR was then performed with the Power SYBR Green PCR Master Mix (Applied Biosystems). Primers used are shown in Supplementary Table S5. The Ct values of the target genes were normalized to β-actin, and the normalized expression analyzed by the ΔΔCt method.

The relative half-life of c-MYC protein in CAL62 versus CAL62-splice cells was analyzed by measuring c-MYC protein abundance after treatment with cycloheximide. Cells plated in 60-mm dishes and grown 24 hours in low serum condition (0.5% FBS) were treated with 100 μg/mL cycloheximide (Sigma), harvested, and lysed at the indicated time points and analyzed by western blotting.

Dishes were first coated with a bottom layer of 0.4% agar in RPMI. Cells were resuspended in a top layer of 0.2% agar in RPMI with 10% FBS and then fed every other day by adding drops of media onto the top layer. After 3 weeks, the colonies were stained with crystal violet and counted in a GelCount colony counter (Oxford OPTRONIX). Minimum diameter of the colonies was set to 100 μm.

Approximately 6- to 8-week-old female SCID mice (NOD.CB17-Prkdc; Envigo RMS Inc.) were injected subcutaneously with 5 × 10⁶ CAL62 or CAL62-splice cells grown to 70% confluence and resuspended in 50% Matrigel (Corning) into the right and left flanks, respectively. Treatments were administered by oral gavage when tumor volume approached 200 mm³ as estimated by measuring the length and width with calipers (width² × length × 0.52). Tumor-bearing mice were randomly assigned into 5 treatment arms: controls (vehicle–4% DMSO in 30% PEG 300); AZD8055 (10 mg/kg); trametinib (0.75 mg/kg); JQ1 (40 mg/kg); AZD8055 + trametinib and AZD8055 + JQ1 (all drugs were from Selleckchem). Mice were weighed at the start of treatment and every second day during the treatment period. AZD8055 was dissolved in a mixture of 4% DMSO and 30% PEG 300 (Sigma), trametinib in 4% DMSO in corn oil and JQ1 in 2% DMSO, 30% PEG 300, and 5% Tween 80. Treatments were administered by oral gavage in a volume of approximately 200 μL. Tumor volume was measured every 2 to 3 days with calipers. After 21 days, mice were humanely killed, and dissected tumors were flash-frozen for subsequent protein isolation. All animal experiments were performed in accordance with a protocol approved by the Institutional Animal Care and Use of Committee of MSKCC.

Statistical analysis for animal studies and cell lines was performed using GraphPad Prism 7. P value was determined by two-tailed t tests. F test was used to compare variances between the groups; if different, Welch correction was applied. Data are shown as mean with SD or mean with 95% CI (n = 3 or more biological replicates).

Ribosome profiling and RNA-seq were performed in C643 and C643-spl-rev cells. Triplicates of cells grown in 150-mm dishes were treated with cycloheximide for 10 minutes and ribosome-protected RNA fragments isolated following the published protocol for ribosome profiling (55) with a modification of including unique molecular identities (UMI) in the library reverse transcription primer (Supplementary Table S5) in order to remove PCR duplicates during analysis. Parallel total RNA extraction was performed for RNA-seq. Ribosome profiling reads and RNA-seq reads were aligned using STAR v2.5. (56) using the UCSC human genome reference, hg19 (http://hgdownload.cse.ucsc.edu/goldenPath/hg19/chromosomes) with ERCC spike-ins included as an extra chromosome. To process the ribosome profiling reads before alignment, linker sequences (5′-CTGTAGGCACCATCAAT-3′) were removed using Trimmomatic v0.32 (57) with the following parameters: number of mismatches between read and adapter: 2; length of alignment between read and adapter: 6. Clipped reads were then filtered to be a minimum length of 25. Reads with UMIs (5′-NNNNTGANNNNCC-3′) were removed from the sequence and inserted into the read name using UMI tools v2.1.1 (58). The set parameters for STAR during alignment were as follows: each read must uniquely map; number of mismatches: 2; maximum intron alignment length: 500,000; 3′ adapter sequence: CTGTAGGCAC; maximum proportion of mismatches within adapter: 0.1; default values were used for all remaining parameters:

Read alignments are available on NCBI SRA under accession number SRP142722. To remove possible rRNA contamination, both the ribosome profiling and RNA-seq reads were aligned to ribosomal sequences gathered from BioMart Ensembl (59) and SILVA (60) databases and merged into a single FASTA reference file (GEO accession GSE113695). Reads were again aligned using STAR with almost all the same parameters, but reads were allowed to align to a maximum of 3 other regions in our rRNA FASTA reference. All reads that aligned to the rRNA reference according to the criteria above were filtered from the original genome reference aligned reads:

For final filtering, all ribosomal reads larger than 35 base pairs were removed and all reads aligning to the same position with the same UMI were reduced to a single read using UMI tools. The final alignment files used for quantifications are available on GEO accession GSE113695.

To estimate abundance on the aligned BAM files, a custom script was used with gencode annotation version 19 and additional ERCC spike-in sequences (ERCC92). For RNA-seq samples with spike-ins, the library size normalization factor was estimated using DESeq2 v3.6 (61). The library size normalization was used in the differential translation efficiency analysis. RiboDiff v0.2.1 (62) with default parameters was used to estimate the change in translational efficiency between sample conditions. Only protein coding genes (Gencode v19) were considered in RiboDiff. The result of RiboDiff is available in Supplementary Tables S1 and S2. We used the following command line:

NthyOri-splice dox-inducible cells were treated with or without dox for 72 hours, followed by cycloheximide (100 μg/mL) for 15 minutes. Cells were then trypsinized and lysed with buffer containing 15 mmol/L Tris-HCl, 300 mmol/L NaCl, 15 mmol/L MgCl₂, 1% Triton X-100, 0.1 mg/mL cycloheximide and ribonuclease inhibitor (RNasin, Promega). Comparable amounts of lysates (1.5 mg protein) were then layered onto a 10% to 60% sucrose density gradient prepared in 15 mmol/L Tris-HCl, 300 mmol/L NaCl, 15 mmol/L MgCl₂, 0.1 mg/mL cycloheximide and RNasin, and fractionated using an SW60Ti rotor in a Beckman ultracentrifuge for 2 hours at 37,000 rpm at 4°C. After centrifugation, the gradients were collected manually from the top into 12 fractions. Fractions were subjected to RNA extraction for ATF4, c-MYC, and β-actin mRNA measurements by real-time PCR as described (63).

GSEA (64) was performed with GSEA software (http://www.broadinstitute.org/gsea/) using the predefined Canonical Pathways and Hallmarks Molecular Signatures Database (MSigDB) gene sets (http://software.broadinstitute.org/gsea/msigdb/index.jsp). The normalized counts of each replicate (GEO accession GSE113695) derived from the RNA-seq of C643 versus C643-spl-rev and CAL62-splice versus CAL62 cells were used as a data set to run GSEA (Identification-Gene symbol; permutations-1000 and permutation type-gene sets). Briefly, GSEA software-derived enrichment scores (ES) identified the functional group of genes (predefined data sets in the MsigDB) overrepresented in the given data set. The normalized enrichment score (NES) was used to determine statistical significance from the nominal P value after controlling for false positives by calculating false discovery rate (FDR). Key top-ranked signatures based on NES, Nom-P value, and FDR-q value prompted validation in our experimental models.

The pGL3-firefly vector was engineered with the ATF4 uORF initiation context by site-directed mutagenesis (primers shown in Supplementary Table S5). HEK293T cells were cotransfected with EIF1AX-WT, G9R, or c'splice expression vectors along with the engineered firefly vectors using FuGENE HD. Renilla luciferase (pRL-null) was cotransfected as a transfection efficiency control. Cells were incubated in 0.5% FBS for 48 hours, and luciferase activity was measured using the Dual-Glo Luciferase Assay system on the GloMax-Multi Microplate Reader (Promega).

The efficiency of a ribosomal reinitiation mode of translation in EIF1AX-WT and the mutants was assessed with the TCRS as described previously (30). The TCRS construct has a short uORF and 2 overlapping ORFs encoding an LP and an SP, respectively, the latter being a marker of ribosomal reinitiation. HEK293T cells were cotransfected with TCRS and EIF1AX-WT, EIF1AX-NTT mutants, or EIF1AX-c'spl expression vectors. The expression of the SP was analyzed by immunoblotting.

Ribosome profiling and RNA-seq data from this study have been submitted to the NCBI Gene Expression Omnibus under accession number GSE113695.

S.D. Leach is a consultant/advisory board member for Nybo Pharmaceuticals. S.W. Lowe is on the scientific advisory board at, has ownership interest (including stock, patents, etc.) in, and is a consultant/advisory board member for Mirimus, Inc. G. Rätsch is on the adjunct faculty at MSKCC, Weill Cornell Medical School, and University Hospital Zurich; reports receiving a commercial research grant from Roche Switzerland; has ownership interest (including stock, patents, etc.) in Computomics GmbH; is a consultant/advisory board member for Computomics GmbH; and has received other remuneration from Novartis Freenovation. No potential conflicts of interest were disclosed by the other authors.

Conception and design: G.P. Krishnamoorthy, S.D. Leach, Z. Zhao, J.A. Knauf, G. Rätsch, J.A. Fagin

Development of methodology: G.P. Krishnamoorthy, S.D. Leach, Z. Zhao, K. Singh, H.-G. Wendel, G. Rätsch

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): G.P. Krishnamoorthy, S.W. Lowe, J. Nagarajah, M. Saqcena, P.P. Tamarapu, J.A. Knauf, J.A. Fagin

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): G.P. Krishnamoorthy, N.R. Davidson, G. Lee, I. Landa, M. Saqcena, R.A. Ghossein, G. Rätsch, J.A. Fagin

Writing, review, and/or revision of the manuscript: G.P. Krishnamoorthy, N.R. Davidson, Z. Zhao, G. Lee, M. Saqcena, S. Dogan, J. Blenis, G. Rätsch, J.A. Fagin

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): G.P. Krishnamoorthy, Z. Zhao, G. Lee, S. Dogan

Study supervision: G.P. Krishnamoorthy, J.A. Fagin

This study was supported by NIH P50-CA72012, R01-CA72597, and R01-CA50706 (J.A. Fagin), R01-CA204228 (S.D. Leach), U54 OD020355-01 (S.W. Lowe), and P30-CA008748. We thank the MSKCC Research Animal Resource Center and the following core labs for their support: Molecular Cytology, Pathology, and the Integrative Genomics Operation funded by Cycle for Survival and the Marie-Josée and Henry R. Kravis Center for Molecular Oncology. We are also grateful to Dr. Cor Calkhoven (ERIBA, the Netherlands) for helpful advice and reagents.

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.