RAS mutation patterns arise from tissue-specific responses to distinct oncogenic signaling

Despite multiple possible oncogenic mutations in the proto-oncogene KRAS, unique subsets of these mutations are detected in different cancer types. As KRAS mutations occur early, if not being initiating, these mutational biases are ostensibly a product of how normal cells respond to the encoded oncoprotein. Oncogenic mutations can impact not only the level of active oncoprotein, but also engagement with effectors and other proteins. To separate these two effects, we generated four novel inducible Kras alleles encoded by the biochemically distinct mutations G12D versus Q61R encoded by native (nat) rare versus common (com) codons to produce either low or high protein levels. Each allele induced a distinct transcriptional response in normal cells. At one end of the spectrum, the KrasnatG12D allele induced transcriptional hallmarks suggestive of an expansion of multipotent cells, while at the other end, the KrascomQ61R allele exhibited all the hallmarks of oncogenic stress and inflammation. Further, this dramatic difference in the transcriptomes of normal cells appears to be a product of signaling differences due to increased protein expression as well as the specific mutation. To determine the impact of these distinct responses on RAS mutational patterning in vivo, all four alleles were globally activated, revealing that hematolymphopoietic lesions were sensitive to the level of active oncoprotein, squamous tumors were sensitive to the G12D mutant, while carcinomas were sensitive to both these features. Thus, we identify how specific KRAS mutations uniquely signal to promote the conversion of normal hematopoietic, epithelial, or squamous cells towards a tumorigenic state.


Introduction
Why specific driver mutations track with different cancers is unknown, yet holds the key to understanding the origins of cancer, with implications for early detection and prevention. This is particularly well illustrated with the small GTPase KRAS. Single point mutations at one of three hotpot positions (G12, G13, and Q61) result in six possible substitutions that inhibit the intrinsic or extrinsic GTPase activity of the protein, rendering KRAS constitutively GTP-bound and active, which is well known to be oncogenic (Simanshu et al., 2017). Despite 18 possible oncogenic mutations, specific subsets of these mutations tend to be found in specific cancer types. For example, G12C is the most common KRAS mutation in non-small cell lung cancer while it is Q61H  Zafra et al., 2020). While this tissue 'tropism' of cancers towards specific KRAS mutations has been appreciated for decades (Bos, 1989), the underlying mechanism is unclear.
Variation in the ability of specific KRAS mutants to be tumorigenic (or not) in different tissues ostensibly results from differences in oncogenic signaling between mutants. Generally speaking, oncogenic signaling is a product of the amplitude of the signal (quantitative signaling) and/or the effector pathway engaged (qualitative signaling). In terms of quantitative signaling, different mutations can exhibit different degrees of activation (GTP-loading) and/or different sensitivities Activating an inducible Kras G12R allele in the pancreas led to early premalignant lesions compared to the much more tumorigenic Kras G12D allele (Zafra et al., 2020). Despite an appreciation that different KRAS mutations can manifest in quantitative or qualitative signaling differences, how these two signaling outputs impact the mutational patterns of this oncogene was unclear.
As KRAS mutations are often initiating, being sufficient to induce tumorigenesis in mice and truncal in many human cancers, the bias of specific mutations towards distinct cancers may arise from tissue-specific responses of normal cells to quantitative and qualitative features of KRAS signaling (Li et al., 2018). Determining the immediate response of normal cells to different Kras mutations in vivo thus holds the key to understanding the mutational patterning of this oncogene. Identifying a point mutation arising in KRAS from the cell-of-origin prior to becoming a tumor is challenging in humans. Mice, on the other hand, provide an ideal model system to experimentally explore this phenomenon, as the point of tumor initiation can be precisely defined using inducible oncogenic Kras alleles. To thus determine why specific KRAS mutations have such a strong bias for different cancer types we created four novel inducible murine Kras alleles designed with very different oncogenic mutations that were expressed at either low or high levels. In this way, tissues sensitive to quantitative signaling would develop tumors dependent on the activation status of the Kras oncoprotein, whereas those sensitive to qualitative signaling would develop tumors dependent upon the mutation type.
We chose two completely different oncogenic mutations for these experiments, namely G12D and Q61R. G12D places a negatively charged head group into the catalytic cleft of RAS and blocks extrinsic (RASGAP-mediated) GTPase activity (Parker et al., 2018). On the other hand, Q61R replaces the catalytic amino acid with one that has a positively charged headgroup, disrupting the position of the active site water molecules necessary for intrinsic GTP hydrolysis (Buhrman et al., 2010). Q61 is also essential for extrinsic GTP hydrolysis, as it stabilizes the transition state via hydrogen bonds to the g-phosphate and nucleophilic water while providing another hydrogen bond to the GAP arginine finger (Grigorenko et al., 2007;Kotting et al., 2008;Rabara et al., 2019;Scheffzek et al., 1997 Smith et al., 2013). In those few cases in which the tumorigenic potential of the G12D and Q61R mutants have been directly compared in mice, tissue specific expression of Nras G12D or Kras G12D was less potent than their Q61R counterparts at inducing melanoma (Burd et al., 2014) or myeloproliferative neoplasm (Kong et al., 2016), respectively.
To parse out the contribution of quantitative oncogenic signaling, the Kras alleles encoding these two different mutations were expressed at different levels by manipulating their codon usage. Namely, the first three coding exons were fused and encoded by either their native rare codons, which is known to retard protein translation, or common codons to increase translation (Lampson et al., 2013;Pershing et al., 2015). We chose the novel approach of altering mammalian codon usage to modulate protein expression (Pershing et al., 2015), as no additional elements are required to change protein levels, providing a simple, reproducible, and uniform way of precisely controlling Kras levels in mice and derived cell lines.
These four alleles were activated and immediately thereafter the transcriptome of normal cells was determined, which revealed that each allele induced a specific transcriptional response in normal cells. Increased expression shifted the transcriptional hallmarks consistent with an expansion of multipotent cells to that of oncogenic stress and inflammation. Changing the mutation shifted the hallmark of estrogen response in the G12D mutant to that of the p53 pathway and DNA repair in the Q61R mutant. To determine how these two types of signaling contribute to RAS mutational patterning in vivo, all four alleles were globally activated, revealing that hematolymphopoietic lesions were sensitive to the level of active oncoprotein, squamous tumors were preferentially sensitive to the G12D mutant, while carcinomas tended to be sensitive to both these changes. Thus, we identify how specific KRAS mutations uniquely signal to promote the conversion of normal hematopoietic, epithelial, or squamous cells towards a tumorigenic state.

A panel of Kras alleles designed to separate the effects of a mutation from the activity of the oncoprotein
To elucidate how specific cancers are driven by specific KRAS mutations, we reasoned that the contribution of quantitative signaling of the oncoprotein could be parsed out by simply changing the amount of protein made, while the contribution of qualitative signaling could be parsed out by using two different oncogenic mutants. To this end, we created the four novel LSL-Kras natG12D , LSL-Kras natQ61R , LSL-Kras comG12D , and LSL-Kras comQ61R alleles ( Figure 1A) in which an LSL transcriptional/translational repressor sequence (STOP) flanked by loxP sites (Jackson et al.,

2001)
was engineered into the first intron of Kras after the non-coding exon 0 to provide temporal and special control of gene expression. This was followed by a fusion of coding exons 1 to 3 encoded by either their native (nat) rare codons, which are known to retard protein translation, or 93 of these rare codons converted to their common (com) counterparts to increase protein expression (Figure 1-figure supplement 1,2), and either a G12D or Q61R mutation. As noted above, each of these mutants alters RAS activity in a biochemically different manner  Pershing et al., 2015). This was followed by the next intron containing an FRT-NEO-FRT cassette for ES selection, which was excised via Flp-mediated recombination after which the Flp transgene was outbred. Finally, the remainder of the gene was left intact so as to generate the two Kras4a and Kras4b isoforms, as both contribute to tumorigenesis , potentially through unique protein

interactions (Amendola et al., 2019).
All four alleles were successfully recombined by Cre recombinase, as confirmed with two separate mouse embryonic fibroblast (MEF) cultures derived from each genotype ( Figure 1B, and leading to higher levels of GTP-bound RAS, all four alleles displayed the expected gradual increase the level of active Kras. Namely, we confirm a stepwise increase in ectopic GTP-bound Kras in the ascending order of Kras natG12D <Kras natQ61R <Kras comG12D <Kras comQ61R by both immunocapture 7/42 and ELISA-based assays ( Figure 1C and Figure 1-figure supplement 3). Finally, we confirm that these four alleles exhibit the expected increase in oncogenic signaling in vivo. Namely, we crossed the LSL-Kras natG12D /+, LSL-Kras natQ61R /+, LSL-Kras comG12D /+, and LSL-Kras comQ61R /+ genotypes into a Rosa26-CreERT2 background, which expresses a tamoxifen-inducible Cre from the endogenous Rosa26 promoter that is active in a broad spectrum of tissues (Ventura et al., 2007). Two adult mice from each of the four derived cohorts, as well as the control strain (Rosa26-CreERT2/+), were injected with tamoxifen and seven days later humanely euthanized, their lungs removed, RNA isolated, and the level of mRNA encoded by RAS target genes determined by qRT-PCR (Figure 1-figure supplement 4A). This revealed the expected increase in three of the target genes by the LSL-Kras alleles when encoded with common codons, which was further increased in the Q61R-mutant background (Figure 1-figure supplement 4B,C).
Thus, the four novel LSL-Kras alleles function as designed.

The effect of codon usage and mutation type on Kras biological activity
As the four alleles exhibited the expected biochemical and signaling properties of the encoded proteins, we next confirmed that these alleles were proportionally tumorigenic in a side-by-side comparison in the same organ. Each of these four alleles was specifically activated by tamoxifen injection to induce expression of Cre recombinase in the lung, after which every month thereafter for six months, five mice from each of the four cohorts were humanely euthanized (Figure 2figure supplement 1A). The lungs from all mice were visually analyzed for the presence of surface pulmonary tumors (Figure 2-figure supplement 1B), and in addition, two H&E-stained sections from pairs of mice were assayed for the presence and type of pulmonary tumors by a veterinarian pathologist blinded to the genotype (Figure 2A). This revealed a stepwise increase in early onset and tumor burden of pulmonary lesions in lock step with the increased biochemical activity and signaling amplitude of the oncoproteins in the ascending order of Kras natG12D <Kras natQ61R <Kras comG12D <Kras comQ61R (Figure 2A,B). These tumorigenic phenotypes were the result of activating the LSL-Kras alleles in the lungs, as five control Kras +/+ mice injected with tamoxifen failed to develop tumors after 13 months, twice the length of the study (Figure 2-figure   supplement 1C,D). Finally, we find preliminary evidence that the G12D-mutant alleles 8/42 preferentially led to atypical alveolar hyperplasia (AAH), whereas bronchiolar hyperplasia/dysplasia (BH) lesions were more prevalent with the Q61R-mutant alleles (Figure 2figure supplement 1E,F). We conclude that this allelic set exhibits the expected increase in tumorigenic potential consistent with the activity and signaling of the encoded oncoproteins, but also exhibited evidence of mutation-specific effects on tumorigenesis.

The response of normal cells to different Kras codon usage or mutations
As different tumor types and grades induced by each of these four oncogenic LSL-Kras alleles are presumably a product of tumor initiation, the key to understanding these effects must lie in how normal cells respond to these different oncoproteins. As the lung was sensitive to both the type of oncogenic mutation and the codon usage of the activated Kras allele, we compared the immediate transcriptional response to activation of each allele in this organ in the aforementioned Rosa26-CreERT2/+ background. Cohorts of three adult male and female mice from each of the four cohorts were injected with tamoxifen, seven days later the animals were humanely euthanized, their lungs removed, and RNA isolated for bulk transcriptome sequencing This revealed that increasing Kras activity (through the combination of mutation type and codon usage) progressively moved the transcriptional signature away from one more similar to control tissue (Kras natG12D ) to one much more complex and with features consistent with high oncogenic activity (Kras comQ61R ) ( Figure 3A and We interpret this as expression level largely dictating the degree of oncogenic signaling in this allelic set. With regards to qualitative signaling, there were transcriptional signatures preferentially induced by specific mutations that were independent of codon usage. Namely, activation of both G12D-mutant alleles induced the Estrogen Response Late hallmark. On the other hand, we found that activation of the Q61R-mutant alleles induced DNA Repair and P53 Pathway and decreased Interferon Alpha Response hallmarks, suggesting that there was a strong tumor suppressive response uniquely activated by these mutant alleles ( Figure 3D and

10/42
Finally, we validated the specific signaling responses detected by bulk RNAseq analysis by quantifying four to six marker genes in a subset of selected hallmarks. Namely, two adult mice from each of the four cohorts and control mice were injected with tamoxifen as above, and seven days later the animals were humanely euthanized, their lungs removed, RNA isolated, and the level of select transcripts determined by qRT-PCR. We identified similar expression patterns to the transcriptome signature in the hallmarks of TNFa Signaling via NFκB and Interferon-g, which increase with Kras activity (Figure 3-figure supplement 7A,B), and EMT and Myogenesis, as these hallmarks were enriched in G12D mutants but depleted in Q61R mutants ( Thus, in broad terms, we find that increasing Kras activation leads to progressively higher oncogenic signaling, eventually reaching the point of oncogenic stress, with the type of oncogenic mutation further modulating this response. Thus, quantitative signaling differences predominate between these mutants, with evidence of qualitative differences, potentially amplified at higher activation levels.

Tissue sensitivities to different RAS signaling
With the biochemical, biological, and signaling output of each allele now determined, we next addressed the tropism of tissues towards specific oncogenic signaling by determining the tumor landscape upon globally activating each allele. As each of these alleles is expected to have different oncogenic potential, we opted for a moribundity endpoint, as opposed to a fixed endpoint, to identify the tissues most sensitive to tumorigenic conversion in a competition-based approach. To this end, each allele was again activated by tamoxifen using the aforementioned ubiquitous Cre driver Rosa26-CreERT2. Mice were then regularly monitored for moribundity endpoints, indicative of ensuing mortality due to cancer, at which point the animals were humanely euthanized (Figure 4-figure supplement 1). As a control to ruled out a tumor phenotype being a product of variations in gene activation, we validated Cre-mediated recombination between all four alleles across 14 diverse tissues seven days after tamoxifen injection, with only the ovary displaying reduced recombination, and hence was not included in the study (Figure 4-figure supplement 2, Figure 4-source data 1, and Supplementary File 1).

11/42
Plotting the percent survival for each of the four genotypes by the Kaplan-Meier approach revealed that the median lifespan of mice progressively increased from 14 days in the case of the activated Kras comQ61R allele to 150 days with the Kras natG12D allele ( Figure 4A). Not surprisingly, the number of tumors per animal mirrored these survival differences (Figure 4- To identify the tissue sensitivity to these different alleles, eight different organs were removed from the mice at necropsy and analyzed for pathologic changes as above. While there was much overlap, the prevalence and severity of specific cancer types varied between alleles, arguing that the different alleles can lead to differences in the tumor landscape ( Figure 4B and Collectively, these observations argue that all tested LSL-Kras alleles are oncogenic, with some tissues being sensitive and others resistant, and the response to these four alleles being variable, implying both qualitative and qualitative signaling differences contribute to the bias of tissue-specific oncogenicity of RAS mutations.

A tissue sensitive to quantitative signaling
The incidence of hematolymphopoietic neoplasias increased with the biochemical activity, tumorigenic potential, and signaling amplitude of the oncoproteins. In more detail, the spleen and thymus from eight to ten mice from all four cohorts were removed at necropsy, after which 12/42 H&E-stained sections were assayed for the presence and grade of hematolymphopoietic neoplasias as above. Beginning with the least active oncoprotein, most of the Kras natG12D mice had no evidence of hematolymphopoietic neoplasms, although some mice had pathologic features consistent with malignant lymphoma. The incidence of malignant lymphoma increased with Kras natQ61R allele, and then again with the Kras comG12D allele. Pathological analysis also revealed medullary hyperplasia in the thymus, as well as leukemic infiltrates in the kidneys and pancreas of these latter mice. Lastly, Kras comQ61R induced severe myeloproliferative disease (MPD) at 100% penetrance, with extensive myeloproliferative infiltrates throughout many tissues ( Figure 4B,C, Figure 4-figure supplement 5, Supplementary files 3,4, and not shown). While we cannot discount high oncogenic activity leading to a different hematopoietic disease in these later mice, it seems more likely that this incredibly short latency for the onset of severe systemic myeloid neoplasia results in high mortality that prevents development of longer latency tumors such as the development of lymphopoietic neoplasms.
Thus, with this proviso, hematolymphopoietic neoplasia are hyper-sensitive to quantitative oncogenic signaling, being induced at the lowest level of active Kras and progressively worsening with increased activity.

Tissues sensitive to qualitative signaling
Proliferative lesions of forestomach and oral squamous epithelium exhibited evidence of qualitative signaling impacting tumorigenesis, namely tumors in these tissues were preferentially induced by the oncogenic G12D mutant of Kras encoded by either native or common codons. In the case of the forestomach tumors, pathological analysis performed as above revealed that the and Supplementary file 4). As such, in these two organs, the G12D mutant was associated with more severe phenotypes than the Q61R mutant, with changing codon usage to increase protein expression shifting this difference to a more advanced stage. These data collectively support a dominant role for qualitative differences in the tumorigenic sensitivity of the upper gastrointestinal epithelium, especially at lower activity levels.

Tissues sensitive to both oncogenic activity and mutation type
The lung appeared to be a tissue whereby the effect of qualitative and quantitative signaling differences were both evident. In terms of qualitative signaling, pathological analysis performed as above revealed that the G12D mutation consistently induced more AAH and adenomas than the Q61R mutation in both the Kras com and Kras nat contexts (Figure 4B,F, Figure 4 Further, the number of animals with BH lesions was higher upon activating the Kras comQ61R compared to the Kras comG12D allele (Figure 4-figure supplement 5,6C), akin to hematolymphopoietic neoplasia. Assuming that AAH and BH represent different types of tumors (and not different stages of the same tumor type), these data collectively support the lung being sensitive to both qualitative and quantitative oncogenic RAS signaling. 14/42

Tissues resistant to oncogenic RAS driven tumorigenesis
Despite widespread tumorigenesis, we note that no overt proliferative lesions were detected at necropsy or by histopathologic analysis in the pancreas, kidney, or liver (Figure 4-figure   supplement 5 and Supplementary file 3). A gross survey of other organs such as the colon, intestine, heart, skin, and mammary glands also did not reveal any macroscopically detectable tumors (not shown). In agreement, many of these same tissues were reported to be refractory to tumorigenesis upon activation of an LSL-Kras G12D allele in the adult by CreER driven by a Rosa26 (Parikh et al., 2012), CK19 (Ray et al., 2011) or Ubc9 (Matkar et al., 2011) promoter. Thus, many organs appear to be intrinsically resistant to the tumorigenic effects of oncogenic Kras, regardless of the mutation type or expression levels tested, at least within the timeframe of this study and in the tested Rosa26-CreERT2 background.

Discussion
Here we describe how separating oncogenic mutations from expression level in the endogenous Kras gene in vivo revealed quantitative and qualitative signaling contributions to RAS mutation patterns. We acknowledge three caveats to this approach. First, the four oncogenic LSL-Kras alleles were generated by fusing coding exons 1 to 3, an artificial gene architecture, and hence can only be compared to themselves and not to other types of Kras alleles. Second, these alleles were induced by an injection of tamoxifen to activate CreER expressed from the Rosa26 locus, which is admittedly an unnatural situation whereby oncogenic Kras is expressed all at once in a multitude of cells and tissues, potentially perturbing homeostasis in the whole animal.
Nevertheless, as the identical design was applied to all four alleles, comparisons can be made within this allelic set, and we find that targeted activation in the lung phenocopied activation by CreER expressed from the Rosa26 locus. Third, Kras activity of this allelic set was defined in the lung. We therefore acknowledge that cell-type differences in feedback regulatory pathways hyper-sensitive to oncogenic Ras signaling. Second, proliferative lesions of forestomach and oral squamous epithelium were preferentially induced by G12D-mutant Kras alleles, pointing towards qualitative differences driving these tumors with higher expression shifting the effects to more aggressive grades. We acknowledge, however, that other interpretations beyond qualitative signaling differences between G12D and Q61R mutants are possible. It is worth mentioning that 16/42 the survival is similar in mice in which Kras natG12D and Kras natQ61R alleles were globally activated, ruling out early moribundity masking the development of squamous lesion. Nevertheless, increasing oncoprotein expression does render the Q61R mutant tumorigenic in this tissue, so it is possible that qualitative signaling differences can be overcome by robust oncogenic activity.
Third, the lung was sensitive to both forms of signaling, with qualitative (G12D-specific) signaling favoring AAH and adenomas while quantitative (Ras activity) favoring BH lesions, perhaps reflecting a different cell-of-origin for these different lesions (Sutherland et al., 2014;Xu et al., 2014). Fourth, with the two caveats that one, the short lifespan of some of the tested mice may prevent longer latency tumors from developing, and two, even though Rosa26-restricted Cre expression activated the four Kras alleles in tissues that did not form tumors, Cre may not be expressed in the tumor cell-of-origin within these tissues, we show that many organs appear to be intrinsically resistant to the tumorigenic effects of oncogenic Kras, at least within the timeframe of this study and in the Rosa26-CreERT2 background. This suggests that the hematopoietic system, lungs, forestomach, and oral mucosa are unique in being sensitive to oncogenic Kras tumorigenesis, again however, within the confines of the experimental design.
Interestingly, these tissue sensitivities share some similarity to that of humans, namely both species have RAS-associated cancers in the lung, mouth, and hematopoietic system but not in the mammary gland, skin, central nervous system and so forth, but there is also some discordance (Prior et al., 2020).
In summary, we interpret these results to imply that there are tissue sensitivities to quantitative and/or qualitative RAS signaling, with quantitative signaling favoring hematolymphopoietic neoplasias, qualitative signaling favoring squamous tumors, while 17/42 carcinomas tended to be a combination of both types of signaling. The unique signaling dependencies of these tissues may, in turn, be capitalized upon to identify new therapeutic opportunities to target early tumorigenesis, when the tumors are particularly vulnerable, perhaps either as an early intervention or as a preventative measure in high-risk populations.

Generation of LSL-Kras natG12D , LSL-Kras natQ61R , LSL-Kras comG12D , and LSL-Kras comQ61R alleles
A bacteria artificial chromosome was engineered with 7.5 kbp of 5' flanking sequence Kras intron 1 DNA, a Lox-STOP-Lox cassette (LSL) (Feil et al., 1996), exons 1 to 3 fused together and encoded by either native (nat) codons or with 95 rare codons converted to the most commonly used codons in the mouse genome (com) and either a G12D or Q61 oncogenic mutation, followed by the N-terminal 564-bp of intron 4, an FRT-Neomycin-FRT cassette, and a further 1.5 kbp of 3' flanking sequence (Figure 1A and Figure 1-figure supplement 1)

Ectopic expression, immunoblots, and Ras activity assay
To validate the expression levels of FLAG-tagged Kras constructs 2 x 10 6 HEK-HT cells (Counter et al., 1992) were seeded in 10 cm tissue culture plates in DMEM with high glucose (Sigma-Aldrich, immunoblots and replicates are provided (Figure 1-source data 2).

Tissue analysis
Animals were humanely euthanized with inhaled carbon dioxide and subjected to a complete necropsy. Selected organs were sampled for microscopic examination including lung, liver, kidney, spleen, thymus, stomach, pancreas, and macroscopic lesions. All tissues were fixed for 48 h in 10% neutral buffered formalin (VWR, 89370-094) and then post-fixed in 70% ethanol (VWR,, processed routinely, embedded in paraffin with the flat sides down, sectioned at a depth of 5 µm, and stained by the H&E method. Routine processing of the lungs from CC10-CreER/+;R26-CAG-fGFP/+ mice was performed by the Duke Research Immunohistology Lab, while all tissues from Rosa26-CreERT2/+ mice were processed by IDEXX Laboratories. Tissues and H&E slides were evaluated by a board certified veterinary anatomic pathologist with experience in murine pathology.

Tissue recombination analysis
At 6 to 8 weeks of age, two female mice from each of the genotypes LSL-Kras natG12D /+, LSL-Kras natQ61R /+, LSL-Kras comG12D /+, and LSL-Kras comQ61R /+ in a Rosa26-CreERT2/+ background were injected with tamoxifen as above, and seven days later humanely euthanized. One age-matched female Rosa26-CreERT2/+; LSL-Kras comG12D mouse was also humanely euthanized as a notamoxifen control. At necropsy the colon, duodenum, ileum, cecum, jejunum, pancreas, spleen, glandular stomach, forestomach, kidney, liver, lung, heart, and ovaries were removed, genomic DNA extracted, and the status of recombination of each of the four oncogenic mutant LSL-Kras alleles determined by recombination PCR in duplicate, as described above. The intensities of bands corresponding to the unaltered wild-type Kras allele (WT) as well as the unrecombined and recombined oncogenic mutant LSL-Kras alleles was quantified with ImageJ software and the recombination rates for each tissue type from each mouse calculated by dividing the densitometry of recombined allele to that of the WT Kras allele. Full-length gels and replicates are provided (Figure 4-source data 1).

Codon usage plots
The codon usage index (Sharp & Li, 1987) was calculated using the relative codon frequency derived from codon usage in the mouse exome (Nakamura et al., 2000) with a sliding windows 24/42 of 25 codons across the open reading frame (ORF) of each transcript. A theoretical Kras ORF encoded by the rarest codons at each position (grey dotted line) was plotted for reference ( Figure   1-figure supplement 1).

Statistical analysis
Statistical analyses were performed using GraphPad Prism software, version 8 (GraphPad Software). One-way ANOVA with Bonferroni's multiple comparisons test with a single pooled variance and a 95% CI were used for experiments with more than 2 groups. Reported P values are adjusted to account for multiple comparisons. A P value of less than 0.05 was considered statistically significant.

RNAseq
Three mice with random distribution of males and females were selected from cohorts of LSL-Kras natG12D /+, LSL-Kras natQ61R /+, LSL-Kras comG12D /+, and LSL-Kras comQ61R /+ mice in a Rosa26-CreERT2/+ background, injected with tamoxifen at 6 to 8 weeks of age as described above, and humanely euthanized seven days later to harvest lungs. Three mice with random distribution of males and females in Rosa26-CreERT2/+ background without the LSL-Kras alleles were used as negative control. For the second RNAseq experiment, three mice with random distribution of males and females were selected from LSL-Kras natG12D /+, and LSL-Kras comQ61R /+ mice in a Rosa26-CreERT2/+ background, injected with tamoxifen at 6 to 8 weeks of age as described above, and humanely euthanized seven days later to harvest lungs. All tissue lysis and RNA extraction steps were performed in a chemical hood and all instruments and tools were sprayed with RNaseZAP™ Samples with RIN less than 7 were not sequenced. RNA-seq libraries were prepared using the commercially available KAPA Stranded mRNA-Seq Kit (Roche). In brief, mRNA transcripts were captured from 500 ng of total RNA using magnetic oligo-dT bead. The mRNA was then fragmented using heat and magnesium, and reverse transcribed using random priming. During second strand synthesis, the cDNA:RNA hybrid was converted into to double-stranded cDNA

Transcriptome analysis
RNA-seq data was processed using the TrimGalore toolkit (http://www.bioinformatics. babraham.ac.uk/projects/trim_galore) which employs Cutadapt (Martin, 2011) to trim lowquality bases and Illumina sequencing adapters from the 3' end of the reads. Only reads that were 20 nucleotides or longer after trimming were kept for further analysis. Reads were mapped 26/42 to the GRCm38v73 version of the mouse genome and transcriptome (Kersey et al., 2012) using the STAR RNA-seq alignment tool (Dobin et al., 2013). Reads were kept for subsequent analysis if they mapped to a single genomic location. Gene counts were compiled using the HTSeq tool (http://www-huber.embl.de/users/ anders/HTSeq/). Only genes that had at least 10 reads in any given library were used in subsequent analysis. Normalization and differential expression was carried out using the DESeq2 (Love et al., 2014) Bioconductor (Huber et al., 2015) package with the R statistical programming environment (http://www.R-project.org). The false discovery rate was calculated to control for multiple hypothesis testing (Figure 3-figure supplement 2-5).
Gene set enrichment analysis (Mootha et al., 2003) was performed to identify hallmarks and pathways associated with altered gene expression for each of the comparisons performed Kras-specific druggable kinases, the transcriptome enriched in Kras natG12D /+ and Kras comQ61R /+specific GSEA hallmarks were cross referenced with kinases whose clinical inhibitors were previously surveyed (Klaeger et al., 2017). The kinases enriched in Kras natG12D /+ and in Kras comQ61R /+ are shown in blue and red, respectively with adjusted p-value less than 5% were highlighted ( Figure 5).

Quantitative Real-Time PCR
Two mice with random distribution of males and females were selected from cohorts of LSL-Kras natG12D /+, LSL-Kras natQ61R /+, LSL-Kras comG12D /+, and LSL-Kras comQ61R /+ mice in a Rosa26-CreERT2/+ background, injected with tamoxifen at 6 to 8 weeks of age as described above, and humanely euthanized seven days later to harvest lungs. Three mice with random distribution of males and females in Rosa26-CreERT2/+ background without the LSL-Kras alleles were used as negative control. RNA extraction was performed as mentioned above, followed by first strand cDNA synthesis from 2 µg RNA, and real-time PCR using GoTaq 2-Step RT-qPCR kit (Promega, A6110). All measurements were normalized against Actin as the internal control using the 2 -DDCt method (Figure 1-source data 3 and

Generation of LSL-Kras rareG12D /+ mice and tumorigenesis study
An additional inducible Kras allele with the most rare codons was generated in the same manner as mentioned above for the four LSL-Kras alleles from BAC design to chimera production ( Figure   1-figure supplement 1). LSL-Kras rareG12D (+neo) chimeras were crossed back to 129S6 mice followed by ACTB:FLPe/+ (Jackson Laboratory, strain 003800) mice to remove the neo selection marker via FLP-mediated excision (Dymecki, 1996). Both germline transmission and the removal of the neo cassette were confirmed with genotyping PCR as mentioned above. Resultant strain was backcrossed with 129S6 mice for five generations, generating the LSL-Kras rareG12D /+ strain. Online Supplemental Material Figure 1-figure supplement 1 shows the codon usage index versus codon number of Kras with native, common, and rare codons. Figure 1-figure supplement 2 shows the sequence alignment of the coding exons of the four novel LSL-Kras alleles used in this study. Figure 1-figure   supplement 3 shows protein expression and activity levels of the encoded Kras proteins. Figure   1-figure supplement 4 shows the qRT-PCR analysis of the RAS target genes.    (A) Schematic of generating and activating LSL-Kras alleles with coding exons 1 to 3 encoded by native (nat) versus common (com) codons with either a G12D or Q61R mutation.

LSL-Kras
(B) PCR genotyping of two independently derived MEF cultures (two biological replicates) with the indicated LSL-Kras alleles in the absence and presence of Cre recombinase (CRE) to detect the unaltered wild-type Kras allele product (WT, 488 bp) as well as the unrecombined (LSL-Kras*, 389 bp) and recombined (LoxP recombined, 616 bp) LSL-Kras allelic products. Gel images were cropped and color inverted for better visualization. Full-length gel images are provided in Figure  1-source data 1.
(C) Levels of active KRAS determined by RBD pull-down (RBD-PD) followed by ELISA analysis using lysates derived from HEK-HT cells transiently expressing the indicated FLAG-tagged Kras proteins. Tubulin and empty vector serve as loading and negative controls, respectively. One of two biological replicates, see     Druggable kinases positively enriched in GSEA hallmarks identified in Figure 4B (blue Kras natG12D , red Kras comQ61R ). * adjusted p-value less than 5%.