Patients’ characteristics
The main characteristics of all patients (N = 177) are in Table 1. The median age of patients at the time of diagnosis was 62 years (range 24–89). Most patients presented with FIGO stage III (82%), grade G3 (85%), and HGSC subtype (84%). About one-third of patients (32%) underwent preoperative chemotherapy, and half of patients (50%) had disease residuum left after surgical tumor debulking. The vast majority of patients (96%) received platinum-based chemotherapy regimens in an adjuvant setting, two received taxane monotherapy, four did not receive any adjuvant treatment due to poor performance status, and for six patients the information about therapy was not available. The median PFI and OS were 25 and 48 months, respectively. Patients with FIGO stage III or IV, residuum after surgery (R1 or R2), or with PFI < 12 months had significantly poorer OS than the rest of the patients (p < 0.001 for all) (Supplementary Fig. S1A-C).
Somatic genetic variability
All six KRAS variants found previously by exome sequencing (n = 50) were also detected by Sanger sequencing in the confirmation part of the study (n = 50). In the extended validation part (n = 125, two samples not assessed due to the lack of DNA), variants in a further six samples were observed (Table 2).
Table 2
Molecular characteristics of EOC patients
Gene | Number of patients | Percentage |
KRAS mutation status* | | |
KRAS wild-type | 163 | 93 |
KRAS mutated | 12 | 7 |
KRAS mutation spectrum | | |
p.Gly12Asp | 5 | 42 |
p.Gly12Val | 3 | 26 |
p.Gly12Cys | 1 | 8 |
p.Gly12Ala | 1 | 8 |
p.Gln61Arg | 1 | 8 |
p.Gln61His | 1 | 8 |
TP53 mutation status | | |
TP53 wild-type | 92 | 52 |
TP53 mutated | 85 | 48 |
TP53 mutation spectrum | | |
Hotspots | | |
p.Arg175His | 8 | |
p.Arg273His | 5 | |
p.Arg248Gln | 5 | |
p.His214Arg | 3 | |
p.Tyr220Cys | 3 | |
p.His179Gln | 2 | |
p.Arg248Trp | 2 | |
p.Asp259Tyr | 2 | |
p.Arg282Trp | 2 | |
p.Glu198Ter | 3 | |
p.Arg213Ter | 3 | |
Private missense mutations | 27 | |
Private frameshift or nonsense mutations | 13 | |
Private splice site mutations with pathogenic features | 4 | |
TP53 mutation functional consequences# | | |
Loss-of-function | 37 | 56 |
Gain-of-function | 29 | 44 |
Not classified | 19 | ̶̶ |
Dominant-negative effect (DNE) & loss-of-function (LOF) properties# | |
DNE_LOF | 54 | 83 |
notDNE_notLOF | 6 | 9 |
notDNE_LOF | 5 | 8 |
Not classified | 20 | ̶̶ |
Transactivation function# | | |
non-functional | 51 | 34 |
functional or partially functional | 97 | 66 |
Not classified | 29 | ̶̶ |
DNA binding loop affected# | | |
yes | 43 | 24 |
no | 134 | 76 |
Footnotes: |
*Result for two samples not available due to DNA of low quality/quantity.
#Evaluated using The TP53 database of NCI (https://tp53.isb-cgc.org/) and The Clinical Knowledgebase (CKB) powered by The Jackson Laboratory database (https://ckb.jax.org/) and literature cited therein.
All variants were missense single nucleotide substitutions in exon 2 (n = 10) or 3 (n = 2). Representative chromatograms are in Supplementary Fig. S2A, B.
As for TP53, the confirmation set showed exactly the same variants compared to exome sequencing, i.e., 39 mutated and 10 wild-type patients, except in one sample where originally the variant p.Pro75fs was detected in exon 4, but the exon was then not covered by the direct sequencing approach. In the validation set (n = 127), an additional 45 mutated samples were identified (Table 2). Representative chromatograms are presented in Supplementary Fig. S3A-J.
Functional classifications enabled the distribution of TP53 variants to several categories: i/ missense (n = 62), out of which 27 were single private mutations and the rest mutational hotspots detected in two or more patients, ii/ two hotspot nonsense variants present in three patients each, and iii/ private frameshifts or nonsense variants (n = 13). The last category was splice site variants with pathogenic features, which were all private (n = 4). The TP53 database of NCI and The Clinical Knowledgebase (CKB) enabled more detailed stratification of variants into loss-of-function (n = 37) versus gain-of-function (n = 29) variants. Most of the somatic variants were classified as having the following properties: dominant-negative effect or loss-of-function (n = 54), non-functional transactivation (n = 51), and affecting DNA binding loop (n = 43) by these databases (Table 2).
Three patients carried mutations in both TP53 and KRAS (co-mutations). One patient with the HGSC subtype had the combination of TP53-Arg282Trp with KRAS-Gln61His, chemoresistant status, and OS of 16 months. The second patient had a clear cell subtype, TP53-Arg248Gln with KRAS-Gln61Arg, chemoresistant status, and extremely short OS of 7 months. The third patient with the mucinous subtype had TP53-Arg213Ter with KRAS-Gly12Asp mutation combination, chemoresistant status, and OS of 19 months. These three available cases suggest that carriage of TP53-KRAS co-mutations could be associated with chemoresistance and poor patient prognosis.
All subsequent clinical genomic analyses were performed using the combined confirmation and validation cohorts (N = 177).
Intratumoral KRAS and TP53 transcript levels
To provide additional functional evidence, we analyzed by qPCR the TP53 and KRAS transcript levels in all available tumor samples together with genetic information. Five samples could not be determined due to low RNA quantity or quality and no tissue left. No extreme outliers were observed.
The carriage or type of KRAS mutations did not significantly associate with the KRAS transcript level (p > 0.05). On the other hand, a significantly lower TP53 transcript level in tumors bearing nonsense, frameshift, or splice site types of variants compared to wild-type TP53 was observed (p < 0.001, Fig. 1A). In contrast, tumors with missense TP53 variants had significantly higher transcript levels than wild-type ones (p < 0.001, Fig. 1A). Higher TP53 transcript level was found in tumors with TP53 variants classified as gain-of-function compared to loss-of-function (p = 0.018), non-functional vs. functional transactivation (p = 0.001), or DNA binding loop affecting vs. other (p < 0.001) (Supplementary Fig. S4A-C).
Carriage of co-mutated TP53-KRAS did not affect transcript expression (p = 0.224 for TP53 and p = 0.204 for KRAS).
Interestingly, the normalized intratumoral TP53 and KRAS transcript levels were mutually significantly correlated (ρ = 0.384, p < 0.001, Fig. 1B).
Figure 1
Associations of TP53 and KRAS normalized transcript levels in tumors with characteristics of EOC patients
(A) TP53 normalized transcript level with TP53 mutation type. (B) Mutual correlation between KRAS and TP53 transcript levels. (C) KRAS normalized transcript level with EOC subtype.
HIGH means nonsense, frameshift, or splice site functional variant classification.
Associations of somatic genetic variability and transcript levels with clinical data of patients
Afterward, we performed statistical analysis of associations between transcript levels, mutational status, spectra, and functional classifications of both genes and clinical data of patients.
Patients with FIGO stage I or II had significantly more frequently mutated KRAS compared to stage III or IV patients (p = 0.007, Table 3). On the other hand, patients with the HGSC tumor subtype had significantly less frequently mutated KRAS (p < 0.001, Table 3), and they had significantly higher KRAS transcript levels (p = 0.004, Fig. 1C) compared to those with other EOC subtypes. KRAS mutation status, spectra, or transcript level were not significantly associated with the rest of the clinical parameters (age, grade of tumor, surgical radicality, chemosensitivity status, or OS, all p > 0.05), and this was true for the association between transcript level and stage as well. Patients with nonHGSC subtypes had significantly more often less advanced stages I/II than HGSCs (p < 0.001, Supplementary Table S2) and thus less aggressive disease. However, only the PFI of patients with clear cell subtype (n = 10) was any better than that of HGSC (n = 134), while for mucinous (n = 9) or LGSC (n = 5), it was not, and endometrioid patients had worst PFI (n = 2) (Supplementary Fig. S5A). However, for OS the difference between subtypes was not that apparent (Supplementary Fig. S5B).As for TP53, its transcript level, mutation status, spectra, or functional classifications were not significantly associated with any of the clinical parameters (age, stage, grade of tumor, subtype, surgical radicality, chemosensitivity status, or OS, all p > 0.05).
Table 3
Associations between KRAS mutational status and stage or tumor subtype of EOC patients
Characteristics | KRAS wild-type* | KRAS mutated* | p-value |
Stage I/II | 17 | 5 | 0.007 |
Stage III/IV | 139 | 6 | |
HGSC | 140 | 2 | < 0.001 |
other subtypes | 19 | 9 | |
Footnotes: |
*Numbers of patients; for some patients clinical data or KRAS mutation status were not available.
We further performed patient stratification into HGSC and nonHGSC subgroups, given the importance of the EOC subtype in previous analyses. No significant associations with clinical data were identified for KRAS or TP53 transcript levels, mutations, or their functional classifications in the HGSC subgroup (n = 143). However, patients with nonHGSC subtypes (n = 28) bearing any TP53 mutations had non-significantly poorer PFI than patients with the wild-type (p = 0.062, Supplementary Fig. S6), and patients with TP53 missense variants disrupting the DNA binding loop had significantly poorer PFI than patients without these alterations, including wild-type carriers (p = 0.011, Fig. 2A). No association was found for OS or other clinical data, including chemosensitivity status.
Patients with co-mutated TP53-KRAS had significantly worse PFI and OS than wild-type patients or those with a single gene mutation (p < 0.001 for both, Fig. 2B, C).
Figure 2
Associations between patient survival and carriage of TP53 or KRAS mutations
(A) Platinum-free interval stratified by carriage of TP53 DNA binding domain mutations in EOC patients with nonHGSC subtype. (B) Platinum-free interval and (C) overall survival in TP53-KRAS co-mutated patients compared to wild-type or single gene mutated EOC patients of all subtypes.
Validation using external datasets
Finally, we attempted to validate our findings using the largest and most up-to-date publicly available EOC dataset within the GENIE project (n = 2210).
The TP53 and KRAS mutation analysis confirmed the overrepresentation of KRAS mutations in nonHGSC compared to HGSC cases. In our dataset, 32% of nonHGSC patients harbored KRAS mutations, while only 1.4% of HGSC cases had such alterations (Table 3). Size of the GENIE dataset allowed the analysis of the distribution of KRAS mutations across all major nonHGSC subtypes. The frequency of KRAS mutations raised in the trend HGSC (1.2%) < < clear cell (13%) < endometrioid (27%) = LGSC (28%) < mucinous (67%). Even more interesting was the trend in the ratio of TP53/KRAS mutability among subtypes, where mucinous and clear cell cases had a 1/1 ratio, while endometrioid and LGSC subtypes had more KRAS than TP53 mutations. Patients with the HGSC subtype had a ratio close to 100/1 in favor of TP53. Most interestingly, analysis of the GENIE dataset revealed a considerable fraction of TP53-KRAS co-mutated patients, again with a high heterogeneity across subtypes. Almost half (46%) of patients with the mucinous subtype had both genes mutated. The other subtypes had a much lower proportion of such events, 4% for endometrioid and 1.5% for clear cell EOC. The occurrence of this phenomenon in LGSC and HGSC was comparable and less than 1% in both cases (Fig. 3A).
As GENIE does not contain expression data, we used the TCGA-OV dataset (n = 374) for the assessment of transcript levels. The comparison of TP53 transcript levels with main mutation classification groups confirmed the trend observed in our study, i.e., significantly higher level in tumors harboring missense mutations (p = 0.002) and lower in those with nonsense, frameshift, or splicing mutations of (p = 0.009) compared to wild-type (Fig. 3B). For KRAS, no significant association of transcript expression with mutation spectra was found (p > 0.05, Fig. 3C), perhaps due to the low number of observations (n = 3 mutated samples). A weak, non-significant correlation was observed between TP53 and KRAS transcripts (p = 0.052, Fig. 3D). In terms of available clinical data, neither grade (G1 or G2 versus G3 or G4) nor stage (stage I or II versus III or IV) were significantly associated with KRAS or TP53 transcript level (p > 0.05, data not shown).
Only two KRAS-TP53 co-mutations were found in the TCGA dataset. One patient (TCGA-29-1696-01A) had KRAS Gly12Arg with TP53 frameshift co-mutation, stage IIIC, G2, and died 34 months after diagnosis. The second patient (TCGA-61-2009-01A) had KRAS Glu61Leu with TP53 missense co-mutation, stage IIIC, G3, and was alive 40 months after diagnosis. Thus, external data do not seem to corroborate our observation of the considerably poorer prognosis of the three EOC patients with such co-mutations.
Due to the absence of survival data in the public version of GENIE and the lack of histopathologically confirmed subtype stratification in the TCGA dataset (however, all tumors were serous), we could not attempt the validation of our prognostic associations (Fig. 2).