Distinct clinical outcomes and biological features of specific KRAS mutants in human pancreatic cancer

SUMMARY

KRAS mutations in pancreatic ductal adenocarcinoma (PDAC) are suggested to vary in oncogenicity but the implications for human patients have not been explored in depth. We examined 1360 consecutive PDAC patients undergoing surgical resection and find that KRAS^G12R mutations are enriched in early-stage (stage I) disease, owing not to smaller tumor size but increased node-negativity. KRAS^G12R tumors are associated with decreased distant recurrence and improved survival as compared to KRAS^G12D. To understand the biological underpinnings, we performed spatial profiling of 20 patients and bulk RNA-sequencing of 100 tumors, finding enhanced oncogenic signaling and EMT in KRAS^G12D and increased NF-κB signaling in KRAS^G12R tumors. Orthogonal studies of mouse Kras^G12R PDAC organoids show decreased migration and improved survival in orthotopic models. KRAS alterations in PDAC are thus associated with distinct presentation, clinical outcomes, and biological behavior, highlighting the prognostic value of mutational analysis and the importance of articulating mutation-specific PDAC biology.

eTOC

McIntyre et al. demonstrate in an analysis of 1360 patients the key features of early-stage pancreatic ductal adenocarcinoma (PDAC), including an enrichment of KRAS^G12R disease that extensively associates with improved outcomes. Spatial profiling, transcriptomic analyses, and orthogonal mouse organoids recapitulate several allele-specific differences, underscoring the importance of mutation-specific biology in human PDAC.

Graphical Abstract:

INTRODUCTION

Pancreatic ductal adenocarcinoma (PDAC) is predicted to be the second leading cause of cancer death by the year 2030.¹ The majority of patients present with locally advanced or metastatic disease, while only approximately 20% of patients are resectable at the time of diagnosis. The inability to treat most patients with curative intent surgical resection and (neo)adjuvant chemotherapy has limited overall survival at 5 years to <10%.²

Stage at diagnosis is associated with overall survival following surgical resection,^3–4 yet many of the biologic underpinnings that drive early-stage PDAC are not well understood.^5–7 Specifically, whether early-stage disease represents a distinct biological subset of PDAC or is simply a cohort of patients in whom disease is discovered earlier remains largely unexplored. Additionally, there is a need for enhanced biomarkers that improve patient selection for treatment and accurately predict clinical outcome.

In parallel to improving characterization of early-stage disease, there is an increased understanding of the genomics of PDAC. There are four main driver genes – KRAS, TP53, SMAD4 and CDKN2A – that drive pancreatic tumorigenesis.^8–11 There are a number of additional genes that play a role in pro-tumorigenic pathways, including genes involved in DNA repair, epigenetic modifications and cell cycle regulation.^8–11 Prior data has demonstrated that alterations in driver genes are associated with outcomes in patients with PDAC.^12–13 Specifically, a KRAS^G12D mutation (35% of patients) has been associated with worse survival as compared to non-G12D KRAS alterations,^12–13 both alone and in conjunction with TP53 mutations¹³. CDKN2A loss has also been suggested to associate with poorer outcomes, with more frequent driver mutations portending worse survival.¹² However, a recent study demonstrated improved overall survival when there was co-mutation of KRAS^G12D and TP53 as compared to other KRAS alleles with TP53 mutation.¹⁴ These contradictory findings raise the question of whether alterations in driver genes alterations impact survival. Further, these data do not account for stage in assessing the impact of mutational burden, nor has the significance of other frequent KRAS alleles – such as KRAS^G12V (30% of patients) or KRAS^G12R (15% of patients) – been well-studied.

Emergent preclinical data suggest that Kras mutants are heterogeneous in their biological behavior. The most frequent and well-studied Kras mutation in PDAC, Kras^G12D, recapitulates the progression from normal epithelium to acinar-ductal metaplasia (ADM) to pre-malignant pancreatic intraepithelial neoplasia (PanIN) and PDAC found in human disease.^15–18 Kras^G12V initiates tumors in mice in a manner similar to Kras^{G12D 19–20}, but Kras^G12R appears deficient in its ability to initiate PanIN²¹, likely owing to differential engagement of active KRAS with the PI3K subunit p110α²². Whether the divergent biology of Kras^G12R PDAC in mouse models extends to human disease remains largely unknown.

In the current study, we evaluated in 1360 patients undergoing PDAC resection the differences in clinical and pathologic factors, including commonly altered genes, associated with early-stage as compared to late-stage disease at diagnosis. In so doing, we uncover an enrichment for KRAS^G12R mutations in stage I PDAC. We find that KRAS^G12R mutant PDAC is more likely to be treated with neoadjuvant chemotherapy and to be absent of nodal involvement. Over time, KRAS^G12R disease is associated with improved overall survival, and is less likely to recur at distant sites but features an increased rate of local recurrence compared to KRAS^G12D. Interestingly, KRAS^G12V disease also features improved OS but without the same enrichment in early-stage disease nor a decreased tendency to metastasize. To understand why KRAS^G12R is associated distinct clinical outcomes, we performed spatial profiling on 20 patients (14 tumors, 6 normal) and bulk RNA-sequencing (RNA-seq) on 100 patients, finding recurrent enrichment of KRAS signaling and epithelial-mesenchymal transition (EMT) among KRAS^G12D, and TNF/NF-κB signaling in KRAS^G12R. Isogenic Kras^MUT; p53^KO organoids recapitulate these transcriptional differences, and show decreased migratory potential and improved survival in orthotopic transplantation studies. These findings highlight distinct patient outcomes that reflect unique biological behavior of different mutant KRAS alleles.

RESULTS

Unique clinical features of early-stage PDAC

There were 1,360 patients who underwent resection between 2004 and 2019, of whom 397 (29%) had early-stage (stage I) tumors and 963 (71%) had late-stage (stage II-III) tumors using the AJCC staging 8^th edition. Baseline clinical and pathologic variables of the cohort, as well as early- and late-stage groups are displayed in Table S1. Early- and late-stage patients were not different with respect to age and risk factors such as smoking, obesity, diabetes, and a prior history of cancer.

Median follow-up of survivors was 84.9 months (95% CI 76.5–94.7 months) and the median OS of the cohort was 30.7 months (95% CI 28.0–32.8 months). Patients with early-stage disease had significantly improved survival as compared to those with late-stage disease (median [95% CI], 42.5 [37–48] months vs 26.2 [24–28] months, p<0.001; Figure 1A). Median RFS in all patients was 12.9 months (95% CI 12.2–13.8 months), and again, patients with early-stage disease had improved RFS as compared to those with late-stage tumors (median [95% CI], 21.4 [16.9–27.5] months vs 11.4 [10.4–12.4] months, p<0.001; Figure 1B). Because we observed no difference in OS between stage IA (T1N0) and stage IB (T2N0) patients (Figure S1A) nor between T0 (pathologic complete response, pCR), T1a/b, and T1c patients (Figure S1B), we focused primarily on the distinction between stage I and stage II-III in further analyses.

Figure 1. Distinct clinical features of resected early-stage pancreatic adenocarcinoma.

A-B, Kaplan-Meier curves of (A) overall survival and (B) recurrence-free survival (RFS) between early- (n = 397) and late- (n = 963) stage patients. C-F, Cumulative incidence of (C) RFS, (D) local recurrence (LR), I distant recurrence (DR), and (F) simultaneous local and distant recurrence between early- (n = 397) and late- (n = 963) stage patients. G, Bar charts for the indicated clinical variables between early- and late-stage disease. H, Tumor size (largest dimension) dot plot between early- and late-stage disease (box-and-whiskers indicate mean +/− SEM). I, Rates of indicated margin positivity among pancreaticoduodenectomy patients with early- and late-stage disease. See also Figure S1 and Tables S1–S2.

We assessed patterns of recurrence in this patient cohort. Patients with early-stage disease had fewer recurrences than patients with late-stage disease (5-year CIF estimate [95% CI], 67% [61–71%] vs 86% [83–88%], p < .001) (Figure 1C). There was no difference in the incidence of local recurrence between groups (5-year CIF estimate [95% CI]; 21% [17–25%] vs 21% [18–24%], p=0.920) (Figure 1D); however, patients with late-stage disease were significantly more likely to have distant metastases or simultaneous local and distant metastases at 5-years (5-year CIF estimate [95% CI]; 44% [41–48%] vs 34% [29–38%], p<0.001; and 21% [18–23%] vs 12% [9–16%], p <0.001) (Figure 1E–F).

We then asked if specific clinical features associate with early-stage disease. We found that early-stage disease is more likely to occur in women (54% vs 46%, p=0.012) and to be identified in distal pancreatectomy specimens (25% vs 21%, p=0.014). Early-stage patients were also more likely to have received neoadjuvant chemotherapy (NAC) and radiation (NAR) (NAC 36% vs 14%, p<0.001; NAR 19% vs 2%, p<0.001), and to not have received adjuvant therapy (68% vs 83%, p<0.001) (Figure 1G; Table S2), reflecting the effect of NAC to downstage patients. Pathologically, as expected by the classification, late-stage tumors were significantly larger in diameter as compared to early-stage tumors (median; 3.0 cm vs 2.4 cm, p<0.001) (Figure 1H). Furthermore, these tumors were more likely to have extrapancreatic invasion (98% vs 86%, p<0.001), lymphovascular invasion (81% vs 38%, p<0.001), perineural invasion (96% vs 78%, p<0.001), poor differentiation (35% vs 23%, p<0.001) and a positive margin (42% vs 24%, p<0.001) (Figure 1G). Notably, late-stage tumors removed via pancreaticoduodenectomy were more likely to have positive posterior (retroperitoneal) and anterior (peritoneal) margins (30% vs 13%, p<0.001; 6% vs 2%, p=0.03), but were not statistically different in terms of the modifiable pancreatic parenchymal, bile duct, or duodenal margins (Figure 1I).

Because we observed an enrichment of patients treated with neoadjuvant therapy in early-stage disease, we asked if the differences in outcome between early- and late-stage patients were due to the effect of preoperative treatments. We observed that patients who had early-stage disease and did not receive neoadjuvant therapy had the longest median overall survival at 46 months, as compared to those with early-stage tumors who were treated with neoadjuvant therapy (35 months) and late-stage disease whether or not they received neoadjuvant treatment (26 months and 26 months, respectively). Using a proportional hazards model, we found no significance to the interaction term between early / late status and neoadjuvant chemotherapy in terms of OS (p=0.439) (Figure S1C). but did find significance for RFS p=0.012 from Cox model) (Figure S1D). Late-stage disease is associated with worse RFS, but the magnitude of association is dependent on whether or not the patient received neoadjuvant chemotherapy (no neoadjuvant: HR 2.11, 95% CI 1.78–2.51; neoadjuvant: HR 1.42, 95% CI 1.09–1.87). Within the cohort of patients with early-stage disease, patients who received neoadjuvant therapy were 1.49 times more likely to recur (HR 1.49, 95% CI 1.16–1.9), yet there was no significant difference between late-stage patients who received or did not receive neoadjuvant treatment (HR 1.01, 95% CI 0.82–1.28) (Figure S1E).

Specific genomic features are associated with early-stage PDAC

We then sought to determine if early-stage disease is genomically distinct or not different from late-stage disease. There were 397 patients who underwent genomic sequencing with MSK-IMPACT; no selection criteria were employed for sequencing (Table S3). Of these, 103 (26%) patients had early-stage disease, while 294 (74%) had late-stage disease. The median age at operation was 68 years, 89 (22%) patients received neoadjuvant treatment and 309 (82%) received adjuvant therapy (Table S4). Neoadjuvant treatment was associated with early stage disease and smaller tumor size and N0 disease (Table S5), and the use of 5-FU based regimens (most frequently FOLFIRINOX) in either the neoadjuvant or adjuvant setting was associated with younger age and no prior history of cancer (Table S6).

KRAS was altered in 361 (90%) of patients, TP53 in 284 (71%), CDKN2A in 95 (24%) and SMAD4 was mutated in 68 (17%) (Figure 2A). There were no differences in the rate of these four driver gene mutations between early versus late-stage disease (Figure 2B). Of note, the frequency of CDKN2A/B deletions determined by IMPACT was quite low (Figure S2A) and therefore not incorporated into these analyses. The 100 most commonly altered genes were evaluated in both early- and late-stage tumors; patients with early-stage tumors were more likely to have putative oncogenic mutations in BRAF (3.9% vs 0.7%, p=0.042; q>0.95) and TGFBR2 (8.7% vs 2.4%, p=0.008; q>0.95) as compared to those with late-stage tumors (Figure 2C; Table S7). Importantly, after adjusting for multiple comparisons no differences were seen. Furthermore, there was no significant difference in the overall number of mutations between early- and late-stage tumors (Figure 2D).

Figure 2. Specific genomic features are associated with early-stage PDAC.

A, OncoPrint of selected gene mutations between early- (n = 103) and late- (n = 294) stage disease patients. B-C, Bar charts showing frequencies of (B) top 4 driver mutations or (C) other selected mutations, between early- and late- stage disease patients. D, Violin plot of overall number of mutations between early- and late-stage tumors. E, Frequency of KRAS mutations (KRAS^G12D, KRAS^G12R and KRAS^G12V; Other = other KRAS mutation; KRAS^WT = Wild-type) in early- and late- stage tumors. F, Frequency of early- and late-stage tumors for each of the indicated KRAS mutants or WT. See also Figure S2 and Tables S3–S7.

Next, we evaluated the cohort for distinct KRAS mutational variants. There were 145 (36.5%) patients with KRAS^G12D mutations, 129 (32.5%) with KRAS^G12V, 55 (13.9%) with KRAS^G12R, 32 (8.1%) with other KRAS alterations, and 36 (9.1%) that were KRAS wildtype (WT) (Table S7). KRAS^G12R was enriched in early-stage PDAC relative to late-stage disease (23% vs 11%; p=0.022). KRAS^G12V, other KRAS mutations, and KRAS^WT were not enriched in either early- or late-stage disease (Figure 2E). Conversely, whereas KRAS^G12D mutant disease was equally enriched among early- and late-stage tumors, patients with KRAS^G12R mutation were more likely to be early-stage as compared to KRAS^G12D or other KRAS alterations (43.6% vs 23.6% for KRAS^G12D; p=0.022). (Figure 2F).

KRAS^G12R mutant PDAC has distinct clinical features

Given the frequency of KRAS^G12R mutant disease in early-stage PDAC, we asked if KRAS alleles associate with specific clinical features. We observed, as expected, that AJCC stage was significantly different for KRAS^G12R (p=0.016) but that this was not the case for other KRAS mutations (Table S8). The migration to earlier overall stage was not attributable to T-stage, as this was not different between KRAS^G12D and the other KRAS alterations (Figure 3A). Indeed, primary tumor size differences were limited to KRAS^WT tumors, which tended to be smaller (median: KRAS^WT 2.2 versus KRAS^G12D 2.9; p=0.042). On the other hand, we observed strong differences in the rate of lymph node positivity, with KRAS^G12R tumors more often node-negative (KRAS^G12R 47% vs KRAS^G12D 26%; p=0.019) (Figure 3B–C; Table S8). In addition, KRAS^G12R PDAC patients were more likely to have a family history of pancreatic cancer (16.4% vs. 4.2%, p=0.003) and tended to be non-smokers (58.2% vs 43.1%, p=0.184), although these findings were not significant (Figure 3D–E).

Figure 3. KRAS^G12R mutant PDAC has distinct clinical features and increased tumor suppressor inactivation.

A, Bar chart of frequency of T-stage by KRAS status. B, Tumor size (largest dimension in cm) scatter plot by indicated KRAS status (mean +/− SEM shown). C, Bar chart of frequency of N-stages by KRAS status. D, Bar chart of frequency of a family history of pancreas, breast or pancreas and breast (both) or no cancer (none) by KRAS. E, Bar chart of frequency of smoking history among patients, by KRAS status. F, Bar chart of frequency of neoadjuvant chemotherapy (NAC) or upfront resection by KRAS status. G, Pie chart of KRAS status for cohorts receiving upfront resection (n = 308) or NAC (n = 89). H-I, Frequency of AJCC stage for upfront resected (H) and neoadjuvant chemotherapy (I) patients by KRAS status. J, Bar charts of frequency of T (top) and N (bottom) stage for either upfront resection (left) or neoadjuvant chemotherapy (right) groups. K, Chromatin modifier mutation frequency per KRAS mutations. L, Bar chart of tumor suppressor mutation frequency for each KRAS mutant. M, Donut charts representing each combination of TP53, CDKN2A and SMAD4 mutation for each of KRAS^G12D, KRAS^G12R and KRAS^G12V. Dunnett’s test with KRAS^G12D as reference was performed for the 0–1 versus 2–3 tumor suppressors mutated comparison. See also Figure S2 and Tables S8–S10.

We asked if KRAS^G12R disease was more likely to be early-stage and node-negative because of different preoperative treatment. Indeed, we observed that resected KRAS^G12R-mutant PDAC patients tended to more frequently have received NAC (32.7% vs 17.9% for KRAS^G12D, p=0.094) (Figure 3F), and conversely, resected patients who received NAC tended to be more likely to have KRAS^G12R alterations than those who had upfront resection (20.2% vs 12%; p=0.094) (Figure 3G; Table S8). We therefore examined if KRAS^G12R patients were enriched among all patients who received NAC and not only those who went on to resection. We analyzed our last 309 consecutive patients who received NAC²³, and found no enrichment for KRAS^G12R, with 17% of all patients who received NAC bearing this allele (Figure S2B). These data suggest that the enrichment of KRAS^G12R among NAC followed by resection patients is indicative of a higher probability of response than of presentation with borderline resectable disease.

Despite the fact that the association between NAC and KRAS^G12R was not statistically significant, we elected to stratify patients into upfront resection and neoadjuvant chemotherapy cohorts to control for this difference. We again observed persistent trends towards earlier overall AJCC stage in KRAS^G12R disease when limiting to patients receiving upfront resection (Figure 3H; Table S9) or when restricting to the NAC cohort (Figure 3I; Table S10), although these differences were not statistically significant. Importantly, stage I predominance appeared driven by a significant increase in node-negativity among both upfront resected and NAC patients, but without a difference in T-stage (Figure 3J). Analysis of these subsets thus indicated that KRAS^G12R was associated with increased frequency of node-negative pathology regardless of absence or presence of preoperative therapy.

KRAS^G12R mutant PDAC features increased tumor suppressor inactivation

We then evaluated if KRAS alleles were different with respect to other genomic alterations. Chromatin modifier mutations were not statistically different according to KRAS status (Figure 3K). In addition, DNA repair pathway, BRAF, MAPK, or PI3KCA mutations were relatively rare and comparably frequent across the KRAS alterations. Notably, we found that KRAS^G12R mutant tumors had a higher proportion of CDKN2A mutations as compared to KRAS^G12D (40% vs 22.1%; p=0.046) (Figure 3L). No differences were observed with respect to the frequency of TP53 or SMAD4 alterations, nor the rate of CDKN2A deletion (Figure S2C) nor actionable fusions (Figure S2D). However, KRAS^G12R mutant tumors were different in the frequency of the various combinations of KRAS, CDKN2A, and SMAD4, with alterations in two or more tumor suppressors (TP53, SMAD4, CDKN2A) more common than in the KRAS^G12D variant (51% vs. 30%; p=0.029) (Figure 3M).

Improved outcomes typify KRAS^G12R mutant PDAC

We then assessed if the different clinicopathologic and mutational features of KRAS^G12R disease were associated with a difference in clinical outcomes. We observed that patients with KRAS^G12R mutations appeared to have distinct first recurrence patterns as compared to those with the KRAS^G12D mutation (Figure 4A). For first distant recurrences, secondary organ involvement was not different between the KRAS alterations (Figure 4B). In a time-to-recurrence analysis, distant recurrence was less frequent in KRAS^G12R disease (2 year CIF: 24.2%; p=0.017), KRAS^WT disease (20.3%; p=0.026), and with other KRAS mutations (13%; p=0.023), as compared to KRAS^G12D (43.2%) (Figure 4C; 4E). Conversely, local recurrence was elevated in KRAS^G12R disease (2-year CIF 20.6% [KRAS^G12R] vs 10.6% [KRAS^G12D]; p=0.045) (Figure 4D–E).

Figure 4. Improved outcomes typify KRAS^G12R mutant PDAC.

A, Bar chart of frequency of first recurrence, by KRAS status. B, Donut chart of the frequency of distant recurrence site by KRAS status. C-D, Cumulative incidence of distant recurrence (DR) (C) and local recurrence (LR) (D) between KRAS alterations. E, Forest plot of cumulative incidence (2 year estimates with 95% confidence intervals) of distant and local recurrence by KRAS allele. Red circles indicate statistical significance in Dunnett’s test comparison to KRAS^G12D. F-G, Kaplan-Meier curves of overall survival (F) or recurrence-free survival (G) for each KRAS mutant and WT. H-I, Median OS (H) and RFS (I) estimates (with 95% confidence intervals) by each KRAS allele. Red circles indicate statistical significance in Dunnett’s test comparison to KRAS^G12D. J-L, Kaplan-Meier curves of overall survival for TP53 (J), CDKN2A (K), and SMAD4 (L) mutation versus wild-type. *p<0.05.

In Kaplan-Meier analysis for overall survival, we observed rare KRAS variants (classified as ‘other’) and KRAS^WT disease were associated with the best OS and KRAS^G12D was associated with the worst median OS (other KRAS 47 months, KRAS^WT 46, KRAS^G12R 39, KRAS^G12V 39, KRAS^G12D 30; p=0.026) (Figure 4F). Similarly, rare KRAS variants had the best median RFS followed by KRAS^WT, with KRAS^G12D having the worst RFS (other KRAS 38 months, KRAS^WT 23, KRAS^G12R 19, KRAS^G12V 14, KRAS^G12D 12; p=0.019) (Figure 4G). However, in individual comparisons, KRAS^G12R and KRAS^G12V were not statistically different from KRAS^G12D in either OS (Figure 4H) or RFS (Figure 4I) in our cohort.

With respect to tumor suppressors, we also were able to find that TP53 mutations were associated with worse OS as compared to wildtype (median, 33.7 vs 43.6 months, p=0.014) (Figure 4J), while alterations in SMAD4 and CDKN2A were not (Figures 4K–L).

KRAS^G12R and KRAS^G12V are associated with improved survival in external datasets

To support our findings, we used multiple external datasets representing 950 patients (Table S11), composed of the ICGC-PACA-AU (n=375), ICGC-PACA-CA (n=232), TCGA-Firehose Legacy (n=148), and Sausen (n=100) cohorts. There were 855 patients with genomic and survival data. Overall survival in the individual datasets was quite comparable (median OS: ICGC AU 19.0 months; ICGC-CA 19.3; Sausen 18.26; TCGA 19.55; p>0.95) (Figure 5A) but was different according to KRAS mutational status in the pooled dataset (Figure 5B). As before, overall survival was shorter in the KRAS^G12D as compared to KRAS^G12V or KRAS^G12R or KRAS^WT patients (median OS 15.6 months vs 22.4, 20.8, and 27 months, respectively; p=0.006). Significant differences according to KRAS mutation were observed in individual comparisons between KRAS^G12D and KRAS^G12R (p=0.021), KRAS^G12D and KRAS^G12V (p=0.004), and KRAS^G12D and KRAS^WT (p=0.003) (Figure 5C).

Figure 5. KRAS^G12R and KRAS^G12V are associated with improved survival in external datasets.

A, Kaplan-Meier curves of overall survival for ICGC-AU, ICGC-CA, Sausen and TCGA Firehose cohort. B, Kaplan-Meier curves of overall survival for all cohorts combined per KRAS alterations. C, Median OS estimates (95% confidence intervals from (B) by KRAS allele. Red circles indicate statistical significance. D, Kaplan-Meier curves of overall survival for all cohorts combined by number of tumor suppressors mutated (0 to 3). E, Univariate analysis by KRAS status using KRAS^G12D as reference or number of tumor suppressors mutated (0, 2 or 3) using 1 as reference. Hazard ratios with 95% confidence intervals are shown. F, Stratified multivariate analysis by KRAS status using KRAS^G12D as reference or number of tumor suppressors mutated (0, 2 or 3) using 1 as reference. Hazard ratios with 95% confidence intervals are shown. Red circles indicate statistical significance. *p<0.05; **p<0.01. See also Table S11.

We then examined tumor suppressor loss in the external datasets. As before, patients with KRAS mutations varied in survival outcome depending on the number of tumor suppressors mutated (Figure 5D). Specifically, patients with no tumor suppressor mutations had longer survival than those with at least one co-mutation of TP53, CDKN2A, or SMAD4 (median OS; 0 suppressors 28.2 months, 1 suppressor 19.6, 2 suppressors 17, 3 suppressors 17 months, p=0.013).

Finally, we performed univariate and multivariate analyses using KRAS mutational status and the number of tumor suppressors lost. In univariate analysis, KRAS^WT (HR 0.67 [0.49–0.91], p=0.011), KRAS^G12R (HR 0.77 [0.60–0.98], p=0.035), and KRAS^G12V (HR 0.71 [0.57–0.88], p=0.001) were all associated with improved survival relative to KRAS^G12D (Figure 5E). As before, the absence of additional tumor suppressor mutations was associated with improved OS (relative to mutation of one tumor suppressor) (HR 0.72 [0.56–0.92], p=0.009). There was no difference between 2 or 3 tumor suppressors compared to 1 (HR 1.08 [0.89–1.32], p=0.402; HR 1.02 [0.73–1.42], p=0.923). In a multivariate analysis stratified by data source, KRAS^G12R (HR 0.77 [0.60–0.99], p=0.038) and KRAS^G12V (HR 0.71 [0.58–0.88], p=0.02) status, as well as the lack of tumor suppressor inactivation (HR 0.74 [0.57–0.95], p=0.02), were all independent predictors of overall outcome (Figure 5F).

Spatial profiling identifies allele-specific differences between KRAS^G12R and KRAS^G12D PDAC

To understand the biological basis for divergent clinical outcomes, we performed spatial molecular imaging on the pancreata of 20 patients (2 cores each; 6 normal (3mm cores), 7 KRAS^G12R PDAC (1.5mm cores), and 7 KRAS^G12D PDAC (1.5mm cores) who underwent resection between 2018 and 2023. Normal pancreas was identified in the resection specimens of those with non-malignant or indolent pathology, including low-grade small pancreatic neuroendocrine tumors, metastatic renal cell carcinoma, or a splenule. Following pathology review, 40 cores from the 20 patients were assembled into two tissue microarrays for spatial analysis consisting of slide processing, image reconstruction, and cell segmentation (Figure 6A). A total of ~400 fields-of-view measuring 0.5 × 0.5mm were interrogated.

Figure 6. Spatial profiling of PDAC reveals allele-specific differences between KRAS^G12D and KRAS^G12R.

A, Schematic delineating spatial molecular imaging workflow using the CosMx. B, Representative images showing the localization of indicated proteins following image reconstruction. C, Representative sections of all 14 PDAC patients following cell typing. D-E, UMAP of all profiled cells across the remaining 13 patients (after exclusion of Patient R2) stratified by cell type (D) or KRAS allele (E). F, Bar plots indicating abundance of the indicated cell types across all normal, KRAS^G12D, and KRAS^G12R samples. G, Representative sections of all 14 patients showing the identified niches across the cohort. H, Bar plot of niche abundance across all normal, KRAS^G12D, and KRAS^G12R samples. I, Boxplot of the proportion of cells in each field-ov-view (FOV) for each indicated genotype (KRAS^G12D or KRAS^G12R) belonging to niche 3. Student’s t-test was performed. ***p<0.001. J, Bar plots indicating abundance of each cell type within the corresponding niche. K, Heatmap of differentially abundant proteins for normal, KRAS^G12D, and KRAS^G12R samples. See also Figure S3.

PDAC samples displayed considerable heterogeneity between patients but not between cores, and rich complex microenvironments featuring abundant desmoplastic stroma, immune infiltrates, and inflammatory markers (Figure 6B). One KRAS^G12R patient (Patient R2) showed distinct marker abundance with profound enrichment of NF-κB and a distinct tumor fraction that clustered separately from all other samples (Figure S3A–B). Review of the initial histology showed a high-grade carcinoma lacking characteristic gland formation in PDAC; molecular analysis also showed an exceedingly rare Il6 mutation. This sample was therefore excluded from subsequent analysis. Cell typing for each of the remaining 26 tumor samples was performed (Figure 6C; Figure S3C), and two-dimensional projection showed discrete tumor and non-tumor compartments (Figure 6D), with greater divergence between KRAS^G12R and KRAS^G12D in the tumoral portion than in cell types of the microenvironment (Figure 6E). Overall, cell type composition of KRAS^G12D and KRAS^G12R tumors was highly similar (Figure 6F), with similar degrees of desmoplasia and immune infiltration that was markedly different from normal tissues.

We then performed neighborhood analysis to identify ‘niches’ within the samples (Figure 6G). Niche composition was different between KRAS^G12D and KRAS^G12R with increased abundance of niche 3 and niche 8 in KRAS^G12D (Figure 6H–I). Niche 6 was only present in one patient and therefore not examined further. Typing of cells within niche 3 revealed that this neighborhood (along with niche 8) is T-cell rich (60% and 95%, respectively) but largely absent of tumor, suggesting that different KRAS mutant tumors interface differently with the immune system. Niche 2, wherein both T cells and tumor are present, was more abundant in KRAS^G12R. Finally, we then examined protein marker abundance between KRAS^G12R and KRAS^G12D tumors, observing that KRAS^G12R was enriched for cytokeratin expression, NF-κB (p65), and CD31 but depleted of EMT/mesenchymal markers (vimentin and fibronectin) and immune cell markers (CD45, CD3, CD8, CD15) (Figure 6K; Figure S3D), highlighting molecular and spatial differences between KRAS^G12D and KRAS^G12R tumors.

KRAS^G12R displays attenuated malignant features in human and mouse PDAC

Given our spatial observations suggesting a KRAS^G12R association with increased inflammation but decreased mesenchymal features and/or desmoplasia, we turned to prospective data from the COMPASS trial²⁴ and resected patients from the same group, wherein we identified 100 upfront resected patients with matched genomic information and bulk RNA-seq. In this cohort, in which there were 63 KRAS^G12D and 37 KRAS^G12R tumors, we observed that KRAS^G12R patients showed improved overall survival from the time of resection (Figure 7A), validating our prior findings. Gene set enrichment analysis (GSEA) of the expression data showed increased ‘KRAS signaling down’ and ‘TNFα signaling via NF-κB’ Hallmarks in KRAS^G12R and increased EMT, mTORC1 signaling and proliferative Hallmarks (E2F, G2M) in KRAS^G12D (Figure 7B). In addition, we observed negative enrichment for KRAS and PDAC-specific molecular signatures in KRAS^G12R tumors (Figure S4A).^25–26 This was not associated with a tendency for KRAS^G12R to be different in KRAS dosage (Figure 7C) or in molecular / transcriptional subtype (Figure 7D).

Figure 7. Kras^G12R displays attenuated malignant features in human and mouse PDAC.

A, Kaplan-Meier curves of overall survival for upfronted resected KRAS^G12D and KRAS^G12R PDAC patients in the COMPASS cohort. B, Gene set enrichment analysis (GSEA) for selected Hallmark pathways for the 100 KRAS^G12D and KRAS^G12R patients in (A). C, Bar plots indicating KRAS allelic imbalance across KRAS^G12D, KRAS^G12R, and KRAS^G12V patients in the COMPASS cohort. D, Bar plots indicating PDAC molecular subtype (as per Collisson, Moffitt, and Bailey classifications) for KRAS^G12D, KRAS^G12R, and KRAS^G12V patients in the COMPASS cohort. E, Schematic describing Kras^MUT; p53^KO mouse PDAC organoid generation. F, Growth over time of the indicated organoid lines. G, Cell viability after 72 hour incubation under standard growth conditions in Alamar blue assays. H, Representative brightfield images for the indicated time points and PDAC lines in a wound healing (migration) assay in vitro. I, Quantification of wound healing / migration assays conducted for the indicated organoids. n=3 independent experiments. J, Gene set enrichment analysis (GSEA) for selected Hallmark pathways for the Kras^G12D and Kras^G12R organoids. K, Kaplan-Meier curves of overall survival for mice orthotopically transplanted with the indicated organoid lines. See also Figure S4.

Next, we leveraged LSL-Kras^G12R and LSL-Kras^G12D mice from which have generated ex vivo-transformed PDAC organoids harboring both Kras mutation and loss of Trp53 (Figure 7E).²¹ These PDAC organoids were previously shown to be comparable in RAS-GTP levels and downstream signaling (as measured by phospho-Erk and phospho-Akt). We observed that these Kras^G12D and Kras^G12R organoids do not diverge in their growth (Figure 7F) nor proliferation (Figure 7G), but did show a robust difference in migration (Figures 7H–I), in keeping with the difference in EMT / KRAS signaling observed in the human setting. Indeed, analysis of RNA-seq of the mouse organoids²¹ showed many similar enriched / depleted pathways as observed in human tumors (Figure 7J), with specific enrichment of mutant KRAS PDAC signatures in Kras^G12D organoids as compared to Kras^G12R (Figure S4B). Orthotopic transplantation of Kras^MUT; p53^KO organoids into the pancreas of immunocompetent B6 mice showed prolonged survival for Kras^G12R (median OS 90 days versus 44 days for Kras^G12D; p<0.001) (Figure 7K), supporting a diminished oncogenicity for KRAS^G12R in the mouse and recapitulating the difference in survival outcomes found in human disease.

DISCUSSION

Pancreatic ductal adenocarcinoma features dismal overall survival, even in patients who undergo curative intent surgical intervention and (neo)adjuvant chemotherapy. Of these patients, early (stage I) disease is associated with the best outcomes. Given recent updates to the AJCC staging criteria^27–28 that standardize classification of these patients, we examined the clinical, pathologic, and genomic features of patients with early PDAC to determine if early diagnosis or distinct biology drives this subset of disease. In keeping with published reports^4,27,29, we show that early PDAC features substantially improved OS and RFS. Early PDAC also has a lower propensity for distant recurrences, but is similar to late PDAC in the likelihood of a local recurrence, suggesting that the N0 status of early PDAC is more central to improved outcome than decreased T-stage. We observed that surprisingly the local recurrence rate is remarkably similar between early and late disease despite the much higher frequency of microscopic positive margins (R1) in the latter group. Indeed, together with the various reports suggesting durable survival of a subset of patients with stage III T4 disease^30–32, these data underscore the finding that metastatic but not necessarily local involvement is the key determinant of long-term survival in this disease.

Importantly, we find that early disease is associated with specific features that suggest distinct biology, namely a tendency towards increased differentiation, less frequent lymphovascular and perineural invasion, and even differential localization within the pancreas. Broadly, however, genomic alterations occur at similar frequencies in early and late disease, both in terms of individual genes and overall number of mutations.

Prior data has demonstrated that genomic alterations of driver genes are associated with outcomes in patients with PDAC, as KRAS, specifically KRAS^G12D, and TP53 have been shown to be associated with worse survival.^9–10 Here we recapitulate those findings in our data and external datasets that TP53 and KRAS status strongly influence disease outcome. We show in particular that KRAS^WT, KRAS^G12R and KRAS^G12V mutations, and the degree of tumor suppressor inactivation are all independently associated with clinical outcomes, reinforcing the importance of mutational profiling to aid in prognostication. Why KRAS^G12V mutation is associated with improved OS and RFS, but is not associated with distinct clinical, pathologic, or genomic features (as we observed with KRAS^G12R) remains a key area for further investigation.

Importantly, we discovered here that KRAS^G12R mutations are specifically enriched in early-stage disease, tending to be node-negative and to have been treated with neoadjuvant chemotherapy. At our institutions, we largely utilize NAC in borderline resectable or locally advanced disease, such that the enrichment of KRAS^G12R tumors in those given NAC suggests this allele more frequently presents with vascular involvement and possibility responds more favorably to chemotherapy.

Close analysis of the KRAS^G12R patients unveiled distinct recurrence patterns for this disease, with a diminished tendency to disseminate to distant organs. In addition, we found an increased rate of co-mutation to tumor suppressors, including to CDKN2A. In terms of gene expression, KRAS^G12R shows diminished KRAS signaling, EMT, and proliferative transcriptional programs in human disease, findings that were validated in spatial profiling of KRAS^G12R and KRAS^G12D tumors. We observed in the tumoral compartment a tendency for KRAS^G12D to feature mesenchymal markers such as fibronectin and vimentin and for KRAS^G12R to show increased inflammatory signaling, findings that concord with prior observations suggesting an enrichment of KRAS^G12D (and absence of KRAS^G12R) in a subset of ‘desmoplastic’ tumors.³³ Together, these data support preclinical findings that Kras^G12R represents a ‘weaker’ oncogenic mutation with diminished abilities to generate tumors in mouse models²¹ and to activate PI3K signaling and macropinocytosis²², both key features of PDAC. We validated these biological observations further via generation of isogenic Kras^G12D; p53^KO and Kras^G12R; p53^KO PDAC organoids that show in the latter reduced KRAS-dependent gene expression²⁶ and a loss of Myc activation, with both KRAS and MYC central to PDAC progression.^34–37 In our model, Kras^G12R shows reduced migration such that transplanted mice show prolonged survival, findings that closely mirror the diminished distant recurrence and favorable survival outcomes in humans.

Based on these findings, we speculate the KRAS^G12R mutant represents a biologically distinct ‘path’, requiring additional inactivation of tumor suppressors – beyond that of either KRAS^G12D or KRAS^G12V – to overcome its signaling defect and to enable malignant progression. In addition, KRAS^G12R mutant tumors seem to also represent a distinct endpoint – microscopically indistinguishable from KRAS^G12D, but with persistent transcriptional differences, an altered microenvironment, and an impairment in its ability to disseminate.

To date, the treatment of resectable PDAC remains largely agnostic to KRAS status, with no consensus recommendation for molecular profiling prior to resection. Our findings suggest, however, there should be a greater benefit of resection (i.e. local control) in KRAS^G12R patients for whom distant recurrence is less likely. In addition, clinical early-stage KRAS^G12R disease, wherein distant recurrence is less likely, would not be expected to benefit as much from neoadjuvant chemotherapy. In our estimation, these are items for immediate consideration in the clinical arena, and highlight the importance of mutational profiling because of the impact on survival and site of recurrence, and the biological differences between tumors depending on genotype.

Over the last ten years, mutational analysis of PDAC has become increasingly vital to determining potential actionable targets. Our data define further scope for leveraging genomic information in PDAC to reveal biological heterogeneity among KRAS variants that are predictive of clinical outcome. These findings should warrant both further interrogation of the heterogeneity among KRAS alleles across multiple tumor types, as well as routine incorporation of KRAS mutational testing in all patients with pancreatic cancer, particularly as novel KRAS inhibitors undergo further preclinical development and clinical validation.

Limitations of the study:

Herein we examined all patients undergoing surgical resection for localized (stage I-III) disease. Whether allele-specific differences are recapitulated in unresectable stage III or metastatic stage IV disease was not explored. Second, our validation of KRAS^G12R-specific biology using spatial profiling of human tumors, bulk transcriptomic analyses of resected PDAC, and mouse organoids – while all internally consistent - should not be viewed as definitive across all mutational or biological contexts, given the small numbers of patient samples / organoid lines interrogated and the mechanistic implications still to be elucidated. Finally, how KRAS allele status impacts biological responses to conventional chemotherapy and/or radiation, or investigational treatments was not articulated here, and warrants further study.

STAR METHODS

Resource availability

Lead contact.

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Rohit Chandwani (roc9045@med.cornell.edu).

Materials availability

This study did not generate new unique reagents.

Data and code availability

Deidentified human PDAC targeted DNA sequencing have been deposited at www.cbioportal.org with study ID ‘pancreas_msk_2024’ and are publicly available as of the date of publication. Additional deidentified human PDAC RNA-seq have been deposited at the European Genome-Phenome Archive (EGA). They are publicly available as of the date of publication. Accession numbers are listed in the key resources table.

Key resources table.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
None
Bacterial and virus strains
None
Biological samples
Normal pancreas obtained from surgical resection	Weill Cornell Medicine
PDAC obtained at surgical resection	Weill Cornell Medicine
Chemicals, peptides, and recombinant proteins
Hank’s balanced salt solution	Corning	14025134
Collagenase V	Sigma-Aldrich	C9263-500MG
DMEM, high glucose	Gibco	11-965-118
Fetal bovine serum	Corning	35-017-CV
Advanced DMEM / F-12	Gibco	12-634-028
Penicillin / streptomycin	GIbco	15140122
L-Glutamine	Gibco	25030081
N-Acetyl-L-cysteine	Sigma-Aldrich	A9165-5G
B-27 supplement	Gibco	17504044
Matrigel (growth factor reduced), phenol red free	Corning	356231
[Leu15]-Gastrin I human	Sigma-Aldrich	G9145-.1MG
Animal Free Recombinant Human Epidermal growth factor (EGF)	PeproTech	AF-100-15
Recombinant Human Noggin	PeproTech	120-10C
Recombinant Human FGF-10	PeproTech	100-26
Y-27632 dihydrochloride (Rock inhibitor)	Sigma Aldrich	Y0503-5MG
Nicotinamide	Sigma-Aldrich	N0636-500G
CHIR99021 (GSK-3 inhibitor)	Cayman Chemical Company	13122
TrypLE Express Enzyme, phenol red	Gibco	12605028
Lipofectamine 2000 Transfection Reagent	Invitrogen	11668019
Nutlin-3	Selleck Chemicals	S1061-5MG
Rat tail collagen I	Gibco	A1048301
alamarBlue Cell Viability Reagent	Invitrogen	DAL1100
Doxycycline hyclate	Sigma-Aldrich	D9891-1G
Mitomycin C	Sigma-Aldrich	M0503-5X2MG
Critical commercial assays
None
Deposited data
Human PDAC targeted DNA sequencing	This paper	https://www.cbioportal.org/study/summary?id=pancreasmsk2024
Mouse PDAC organoid RNA-seq	Zafra et al., 202021	SRA: PRJNA578549
Human PDAC RNA-seq	This paper	EGA: EGAD0001004548 and EGAD00001009409
Human PDAC whole exome sequencing: Johns Hopkins	Sausen et al., 201545
Human PDAC whole exome sequencing: TCGA	Raphael et al., 201746	http://www.cbioportal.org
Human PDAC whole exome sequencing: ICGC (PACA-AU; PACA-CA)	Bailey et al., 20168
Experimental models: Cell lines
Mouse: KrasMUT; Trp53KO organoids	Zafra et al., 20 2021 This paper
Mouse: KrasMUT; Trp53KO cell lines	This paper
Experimental models: Organisms/strains
Mouse: C57BL/6	Charles River Laboratories
Oligonucleotides
None
Software and algorithms
Mutect	CIbulskis et al., 201339	https://github.com/broadinstitute/mutect
Pindel	Ye et al., 200940	https://github.com/genome/pindel
MSIsensor	Niu et al., 201442	https://github.com/ding-lab/msisensor
FACETS	Shen et al., 201644	https://github.com/mskcc/facets
AtoMx Spatial Imaging Platform	Nanostring He et al., 202248
Harmony	Korsunsky et al., 201949	https://portals.broadinstitute.org/harmony/
Squidpy	Palla et al., 202250	https://squidpy.readthedocs.io/
Scanpy	Wolf et al., 201851	https://scanpy.readthedocs.io/
Seurat	Hao et al., 202452	https://www.satijalab.org/seurat
R (v4.2.3 and v.3.5.1)	GNU project
Python (v3.10)	Python Software Foundation
DittoSeq	Bunis et al., 202053	https://bioconductor.org/packages/release/bioc/html/dittoSeq.html
STAR (v2.7.4a)	Dobin et al., 201355	https://github.com/alexdobin/STAR
Picard (v2.21.4a)	Broad Institute	https://broadinstitute.github.io/picard/
Stringtie (v2.0.6)	Pertea et al., 201556.
fGSEA	Subramanian et al., 200557	http://gsea-msigdb.org
DESeq2	Love et al.58	https://bioconductor.org/packages/release/bioc/html/DESeq2.html
SAS (v9.4)	SAS Institute
GraphPad Prism	GraphPad Software	https://www.graphpad.com/
Adobe Illustrator	Adobe	https://www.adobe.com/
Biorender	Biorender
Other
Incucyte	Sartorius Biosciences
CosMx Spatial Molecular Imager	Nanostring

This paper does not report original code.

Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Experimental model and study participant details

Patient cohorts.

A prospectively maintained database was queried for consecutive patients who underwent pancreatic resection for pancreatic ductal adenocarcinoma from April 2004 to April 2019, identifying 1360 study participances. All patients included underwent R0/R1 resection, and a subset (397) had genomic sequencing of their primary tumor using Memorial Sloan Kettering-Integrated Mutation Profiling of Actionable Cancer Targets (MSK-IMPACT) (NCT01775072). Patients were excluded if they had metastatic disease at the time of operation, underwent R2 resection or had inadequate follow-up. Demographic, clinical, and pathologic information were obtained from the database and the electronic medical record. Approval for this study was obtained by the Institutional Review Board at Memorial Sloan Kettering Cancer Center (MSKCC) (IRB# 12–245).

Early-stage disease was defined as T1N0 or T2N0 disease based on the AJCC Staging 8^th Edition, whereas late-stage disease was defined as T3, T4 or N1/2 tumors.²⁷ Cross-sectional imaging was obtained during follow-up at the discretion of the treating physician. Recurrence patterns were assessed on postoperative imaging studies. Local recurrence was defined as radiographic evidence of recurrent disease in the remnant pancreas or surgical bed, whereas distant recurrences were defined as disease at all other sites.

Additional study participants included patients enrolled in the COMPASS trial (NCT02750657) and resected patients. Patients were enrolled from multiple centres (Princess Margaret Cancer Centre (Toronto, Ontario, Canada), McGill University Health Centre (MUHC, Montreal, Quebec, Canada), and Kingston General Hospital (Kingston, Ontario, Canada)), with written informed consent prior to enrollment. The COMPASS trial consists of patients with advanced pancreatic ductal adenocarcinoma who had biopsies prior to chemotherapy with modified FOLFIRINOX (mFFX) or Gemcitabine/nab-Paclitaxel (GnP). All patients were of good performance status (ECOG: 0 to 1) and choice of treatment was at the discretion of the treating medical oncologist. Treatment response was assessed radiologically (RECIST 1.1). Patients were enrolled from December 2015 until June 2020 with follow-up till October 1, 2021. The COMPASS trial and analysis of patients undergoing resection were approved by the Institutional Ethics boards at the University Health Network (Toronto, Ontario, Canada); MUHC Centre for Applied Ethics (Montreal, Quebec, Canada); and Queen’s University Health Sciences and Affiliated Teaching Hospitals Research Ethics Board (Kingston, Ontario, Canada).

Additional study participants included patients undergoing resection for localized pancreatic ductal adenocarcinoma. Patients with KRAS^G12D and KRAS^G12R PDAC (n=7 each), as well as ‘normal’ patients (n=6) undergoing resection for non-exocrine pathology (metastatic renal cell carcinoma, splenule, pancreatic neuroendocrine tumors) were retrospectively identified for spatial analysis. Analysis of these patients pancreatic tissue was approved by the Institutional Review Board of Weill Cornell Medicine (IRB#1306014012).

Animal models.

Mice were housed in a pathogen-free facility at Weill Cornell Medicine (WCM) under standard housing conditions. All manipulations were performed under the Institutional Animal Care and Use Committee (IACUC)–approved protocol (2017–0038)

Method details

Genomic sequencing.

A subset of patients underwent next generation sequencing (NGS) of their primary tumor. All samples were obtained from a resection specimen or pretreatment biopsy of the primary tumor, which were subsequently reviewed by a gastrointestinal pathologist and macrodissected from formalin-fixed, paraffin-embedded blocks. Matched DNA from blood samples were used to call somatic tumor mutations. Targeted genomic sequencing of all exons and selected introns of 351, 410, or 468-genes was completed using MSK-IMPACT.³⁸ Analysis was completed as previously described – MuTect³⁹, Pindel⁴⁰ and Somatic Indel Detector were used to call single-nucleotide variants and insertions/deletions,⁴¹ MSIsensor was used to evaluate microsatellite instability,^42–43 and FACETS was used to determine copy number alterations.⁴⁴

External datasets.

A total of four external dataset cohorts were included in this study. Somatic mutational and clinical data for the Sausen cohort (n = 100) of PDAC primary tumor samples from patients with resectable stage I disease with at least 10% of viable tumor cell content were obtained.⁴⁵ Somatic mutational and clinical data of PDAC tumor samples of the TCGA-Firehose Legacy cohort (n = 148) were downloaded from cBioPortal (https://www.cbioportal.org/).⁴⁶ Somatic mutation and clinical data of the ICGC-PACA-CA cohort (n = 232) and the ICGC-PACA-AU cohort (n = 375) were downloaded from the International Cancer Genome Consortium (ICGC) database (https://www.icgc.org) and combined with published data for the ICGC-PACA-AU cohort.^8,9,11 For both ICGC cohorts, samples were restricted to primary operable, non-pretreated pancreatic ductal adenocarcinoma. In all datasets, patients were excluded if they had metastatic disease at the time of operation and/or have non-PDAC histology. For all studies, all KRAS mutations that were not KRAS^G12D, KRAS^G12R, or KRAS^G12V were combined into an “Other” category.

Organoid derivation.

Isolation of normal pancreatic ducts was done modifying previously described protocols.²¹ Briefly, pancreata were minced and washed in Hank’s Balanced Salt Solution (Corning), and then incubated for 30 minutes at 37°C with Collagenase V to release the ducts. After washing twice with DMEM/10% FBS media, ducts were resuspended in basal media [advanced DMEM/F12 (Corning) containing 1% penicillin/streptomycin, 1% glutamine, 1.25 mmol/L N-acetylcysteine (Sigma Aldrich A9165-SG), and B27 Supplement (Gibco)], and mixed 1:10 with growth factor–reduced (GFR) Matrigel (BD Biosciences). Forty microliters of the resuspension was plated per well in a 48-well plate and placed in a 37°C incubator to polymerize for 10 minutes. To culture ductal pancreatic organoids, the basal media described above was supplemented with 10 nmol/L Gastrin (Sigma), 50 ng/mL EGF (PeproTech), 10% RSPO1-conditioned media, 100 ng/mL Noggin (PeproTech), 100 ng/mL FGF10 (PeproTech), and 10 mmol/L nicotinamide [Sigma; note: culture freshly isolated organoids in pancreatic organoid media (POM) containing 10 μmol/L Rock inhibitor (Y-27632) during 48–72 hours]. For subculture and maintenance, media were changed on organoids every two days and they were passaged 1:3 every 5 days. To passage, the growth media was removed and the Matrigel was resuspended in cold basal media and transferred to a 15-mL Falcon tube. Organoids were mechanically disassociated using a P1000 and pipetting 40 times. Five milliliters of cold PBS were added to the tube and cells were then centrifuged at 1,200 rpm for 5 minutes and the supernatant was aspirated. Cells were then resuspended in GFR Matrigel and replated as above.

In order to generate Kras^mut / Trp53 loss (KP) organoids normal organoids were cultured in complete media as described above plus GSK-3 inhibitor and Y-27632 for 48 hours before transduction. A single cell suspension from 1 well of a 48 well plate of organoids was generated by incubating in TrypLE for 10 minutes and mechanically dissociating by pipetting 50x every 5 minutes. The organoids were then washed with ice-cold DPBS and transfected with plasmid containing sequences encoding Cre recombinase, gRNA (against Trp53), and Cas9. And using Lipofectamine (Gibco). Organoids were cultured in supplemented basal media with Y-27632 for a week before selecting with 10 umol/L nutlin.

Generation of 2D-organoid derived lines.

Twelve-well plates were coated for at least an hour with rat tail collagen I (Gibco Cat: A1048301) diluted 1:100 in DPBS. Organoids were mechanically disassociated and seeded directly on collagen coated plates and allowed to grow for 48–72 hours in complete media. Cells were then passaged 2 more times directly onto 12 well plate with no collagen coating and switched to Basal Media as described above before expanding.

In vitro organoid studies.

For growth studies, cells were plated at low-density in 24 well plates and in an IncuCyte Live-Cell Analysis System (Essen Biosciences, Ann Arbor, MI). Images were acquired daily for 7 days. For proliferation, similarly plated cells were incubated for 72 hours and the alamarBlue Viability Reagent was added. For migration studies, cells of interest were pretreated with doxycycline (Dox) for 2–3 days. Subsequently, the cells were trypsinized and counted. Inserts were placed in the middle of each well in a 12-well plate, ensuring similar placement between wells. Cells were then added at a high density, ranging from 30,000 to 70,000 cells per well, with a typical experiment using 55,000 cells in 110 μL per well. Two hours before the experiment, cells were treated with Mitomycin C at a working concentration of 5 μg/mL. Inserts were carefully removed using tweezers, and wells were then filled with appropriate media. Imaging was conducted at 4x magnification at various intervals, depending on the cell type, with five images taken per well. The same localization spot was imaged at each interval to maintain consistency. Imaging was performed every 12 hours.

Orthotopic transplantation models.

All experiments were conducted according to standard protocols outlined using standard husbandry and housing conditions. For orthotopic mouse models, 7 to 9-week-old wild-type (WT) C57BL/6 (stock 027) mice were purchased from The Charles River Laboratories. All mice were from a C57BL/6 genetic background. Female mice were used for orthotopic injections of Kras^G12D/+; Trp53^KO organoids and Kras^G12R/+; Trp53^KO organoids, as described previously.⁴⁷ Mice were monitored biweekly and animals were euthanized if they displayed severe signs of distress.

Spatial molecular imaging.

We performed high-plex spatial profiling on pancreatic and pancreatic cancer tissue microarrays (TMA) using NanoString CosMx SMI platform. The method for this technology is detailed in He et al. (2022).⁴⁸ Each slide was processed for multiplex in situ spatial transcriptomics (1000-plex) and proteomics (67-plex), enabling measurements of RNA and proteins at subcellular resolution. Briefly, formalin-fixed, paraffin-embedded (FFPE) pancreatic and pancreatic cancer tissue microarrays (TMA) were utilized. The blocks were first subjected to standard FFPE tissue preparation protocols, including deparaffinization, rehydration, and antigen retrieval. Following tissue preparation, RNA targets were exposed using a combination of heat-induced epitope retrieval.

The spatial molecular imaging (SMI) chemistry relies on in situ hybridization (ISH) probes and fluorescent readout probes called reporters. Each ISH probe consists of a target-binding domain (35–50 nt) specific to the RNA target and a readout domain (60–80 nt) comprising four consecutive sequences for reporter binding. For each gene, five distinct RNA-detection oligonucleotide probes were designed to ensure high detection sensitivity, even in highly fragmented RNA from FFPE samples. The slides were hybridized with ISH probes followed by the introduction of fluorescent bead-based fiducials fixed to the tissue for optical reference during cyclic imaging. Each reporter construct was conjugated with one of four fluorophores (Alexa Fluor-488, ATTO 532, Dyomics-605, or Alexa Fluor-647), allowing detection as one of four colors in the imaging system. Then, tissue samples were assembled into flow cells and placed within the SMI instrument. For RNA assay readout, tissue was hybridized with 16 sets of fluorescent reporters sequentially. Each reporter set contained four single-color reporter pools, and high-resolution Z-stacked images were acquired after each hybridization round. Post hybridization, the PC linkers on the reporters were cleaved using UV illumination, and the fluorophores were washed away before proceeding with the next set of reporters. Primary image processing transformed fluorescence signals in raw images into digital outputs comprising detected spots localized in x-, y-, and z-dimensions.

AtoMx spatial informatics platform (v1.3) was used as a custom image analysis algorithm, to process 3D multichannel image stacks and to reduce the multidimensional image data to a list of individual reporter-binding events. This platform was also used to decode algorithms that identify unique xyz locations, constructing neighborhoods for gene-specific barcode searches. Two passes of the data with varying search radii ensured comprehensive detection while minimizing error calls. Subsequent filtering steps removed potential duplicate calls and ensured high-confidence transcript identification. After RNA decoding, cell segmentation was performed using tissue morphology markers. The tissue was stained with DAPI and labeled with antibodies for specific markers such as CD298, PanCK, CD45, and CD68. Segmentation accuracy was enhanced using a combination of image preprocessing and machine learning algorithms, particularly the optimized Cellpose neural network available via AtoMx, to define cell boundaries accurately.

Followed by quality control, data normalization was done only for protein using total intensity normalization to correct for technical variations, ensuring comparability across different samples. RNA data was not incorporated into the current analysis. Batch effects were corrected using the Harmony algorithm, which integrates data by aligning cells based on principal components and correcting batch-related variations while preserving biological signals.⁴⁹ Visualization of spatial omics data was performed using several bioinformatics tools, including Squidpy⁵⁰ (for spatial analysis and visualization capabilities, Scanpy⁵¹ (for count-level single-cell analysis and clustering), Seurat⁵² (for R-based data integration and clustering), and DittoSeq⁵³ (for visualization). The analyses were conducted using R version 4.2.3 and Python version 3.10.

RNA-sequencing.

Tumour samples underwent laser capture microdissection for tumor enrichment followed by RNA-seq, as described previously.⁵⁴ Briefly, reads were aligned to the human reference genome (hg38) and transcriptome (Ensembl v.100) using STAR v.2.7.4a⁵⁵, with duplicated reads marked using Picard v.2.21.4 (https://github.com/broadinstitute/picard). Gene expression was calculated in transcripts per million reads mapped using the stringtie package v.2.0.6.⁵⁶ All RNA-seq data has been deposited in the European Genome-phenome Archive (EGA) at EGAD0001004548 and EGAD00001009409. Mouse organoid RNA-seq performed previously²¹ was also analyzed (SRA: PRJNA578549).

Gene set enrichment analysis.

Using normalized read counts of RNA-seq data, the fgseaMultilevel function from the fgsea package with the following parameters were used: minGSSize = 15; maxGSSize = 500 MSigDB Hallmark, and specific custom gene sets were tested for all analyses.⁵⁷ Genes were ranked based on DESeq’s wald statistic (stat), which takes into account the log-fold change and its standard error.⁵⁸

Quantification and statistical analysis

Continuous data are expressed as median and interquartile range (IQR) and compared between groups using Wilcoxon rank sum test. Categorical variables are shown as frequency and percentage and compared between groups using Fisher’s exact test. False discovery rate (FDR) was used to correct for multiple comparisons in gene mutation analysis.

Overall survival (OS) was defined as time from surgery to date of death or censored at last follow up. Recurrence free survival (RFS) was defined as time from surgery to recurrence, death, or censored at last CT without evidence of disease (NED). For cumulative incidence functions (CIF), local recurrence free survival (LRFS) was defined as time until a local only recurrence and competing risk included any distant recurrence or death without local recurrence. Distant recurrence free survival (DRFS) was defined as time until a distant only recurrence and competing risks included any local recurrence or death without distant recurrence. Simultaneous recurrence free survival (SRFS) was defined as time to a local and distant recurrence simultaneously competing risk were any single recurrence or death without recurrence. OS and RFS were analyzed using Kaplan Meier (KM) methods and compared between groups using log rank test. CIF for LRFS, DRFS, and SRFS were analyzed using competing framework and compared between groups using Fine and Gray test. Dunnett’s test was used to adjust for multiple comparisons against our control group KRAS^G12D.

All tests were 2-sided, and p<0.05 was considered significant. SAS (version 9.4; SAS institute, Inc., Cary, North Carolina) or R (version 3.5.1; R Foundation for Statistical Computing, Vienna, Austria) were used for all analyses.

Highlights.

• Early-stage (stage I) pancreatic cancer is enriched for KRAS^G12R mutations

• KRAS^G12R patients have reduced nodal disease and distant recurrence and improved OS

• Spatial profiling reveals decreased oncogenic signaling in human KRAS^G12R tumors

• Mouse Kras^G12R PDAC organoids recapitulate transcriptional and survival differences