Real world predictors of response and 24-month survival in high-grade TP53-mutated myeloid neoplasms

Current therapies for high-grade TP53-mutated myeloid neoplasms (≥10% blasts) do not offer a meaningful survival benefit except allogeneic stem cell transplantation in the minority who achieve a complete response to first line therapy (CR1). To identify reliable pre-therapy predictors of complete response to first-line therapy (CR1) and outcomes, we assembled a cohort of 242 individuals with TP53-mutated myeloid neoplasms and ≥10% blasts with well-annotated clinical, molecular and pathology data. Key outcomes examined were CR1 & 24-month survival (OS24). In this elderly cohort (median age 68.2 years) with 74.0% receiving frontline non-intensive regimens (hypomethylating agents +/- venetoclax), the overall cohort CR1 rate was 25.6% (50/195). We additionally identified several pre-therapy factors predictive of inferior CR1 including male gender (P = 0.026), ≥2 autosomal monosomies (P < 0.001), −17/17p (P = 0.011), multi-hit TP53 allelic state (P < 0.001) and CUX1 co-alterations (P = 0.010). In univariable analysis of the entire cohort, inferior OS24 was predicated by ≥2 monosomies (P = 0.004), TP53 VAF > 25% (P = 0.002), TP53 splice junction mutations (P = 0.007) and antecedent treated myeloid neoplasm (P = 0.001). In addition, mutations/deletions in CUX1, U2AF1, EZH2, TET2, CBL, or KRAS (‘EPI6’ signature) predicted inferior OS24 (HR = 2.0 [1.5–2.8]; P < 0.0001). In a subgroup analysis of HMA +/-Ven treated individuals (N = 144), TP53 VAF and monosomies did not impact OS24. A risk score for HMA +/-Ven treated individuals incorporating three pre-therapy predictors including TP53 splice junction mutations, EPI6 and antecedent treated myeloid neoplasm stratified 3 prognostic distinct groups: intermediate, intermediate-poor, and poor with significantly different median (12.8, 6.0, 4.3 months) and 24-month (20.9%, 5.7%, 0.5%) survival (P < 0.0001). For the first time, in a seemingly monolithic high-risk cohort, our data identifies several baseline factors that predict response and 24-month survival.

Beyond assessing complete response to first-line therapy (CR1) and allogeneic hematopoietic stem cell transplantation (allo-SCT) in CR1, there are no well-established pre-therapy prognostic indicators in this cohort of patients [10] who are automatically assigned to the adverse group per ELN2022 risk stratification [9].Furthermore, there is a significant heterogeneity in response to available therapies (intensive chemotherapy vs. hypomethylating agent (HMA)-based therapy) or agents used in clinical trials (for e.g., APR-246 [11]) with few long-term survivors even in those receiving allo-SCT [12].Consequently, there is a pressing need to explore better means to identify disease characteristics influencing therapeutic response to optimally select frontline treatments and identify patients who are likely to achieve durable benefit from allo-SCT in CR1.

MATERIALS AND METHODS Cohort case selection and sample procurement
We identified patients with TP53 MUT MN carrying ≥1 TP53 mutation at a VAF ≥ 3% diagnosed between 2014 and 2023 across four US centers with largely similar treatment practices.We excluded individuals with any of the following: a known germline TP53 mutation, known TP53 MUT precursor states (CHIP, CCUS), MDS with mutated TP53 (<10% blasts) [6], and core binding factor-altered AMLs.

Cytogenetic studies
Chromosome analysis was performed following standard cytogenetic laboratory clinical protocol.Fluorescence in situ hybridization (FISH) testing was performed using probe sets targeting most recurring abnormalities in myeloid neoplasms on bone marrow or malignancy-involved peripheral blood.Using previously published criteria, over 90% of the cohort cases met the classification of monosomal karyotype (MK) [1].Therefore, we applied a revised definition of MK restricting to only unique autosomal monosomies (0−1 vs. 2+) without considering other structural alterations (Supplementary S1).

Next generation sequencing
Somatic next-generation sequencing (NGS) data was available in all cases.To maintain consistency in our analysis, we only included genes that were tested across two or more centers (Supplementary S2).For missense TP53 mutations, we also examined the evolutionary action score for p53 (Supplementary S3) [14].Additionally, germline testing data (Supplementary S4) was available in a subset of cases.
Allelic status.TP53 multi-hit (TP53 MH ) allelic state was designated per ICC 2022 schema [6].However, single-hit (TP53 SH ) designation used an  Numbers reported in therapies and response may not add up to cohort total since some patients were either untreated or if treated, response could not evaluated due to various reasons (active second malignancy, early treatment-emergent adverse effects, transferred care elsewhere or early mortality).c In pts with multiple mutations, case was designated TP53 NMIS if ≥1 TP53 mutation was a TP53 NMIS mutation.
expanded VAF cutoff including cases with VAF between 3 and 10%.Additionally, we modeled any non-missense TP53 mutations (TP53 NMIS ) as a binary predictor.

Data analysis
The primary endpoint was 24-month overall survival (OS24) from the time of diagnosis of TP53 MUT MN to death or last follow-up, censoring patients alive at 24 months.Response to first-line therapy was also evaluated.We employed non-parametric Kaplan-Meier methods, along with Cox proportional hazards (P-H) regression models [15] and, where suitable, flexible parametric models [16] (Supplementary S5).

Cohort summary
Table 1 depicts the baseline data of all patients with therapy/ response data of all patients receiving definitive therapy stratified by TP53 allelic state at diagnosis of TP53 MUT myeloid neoplasm.Individuals accrued after 2018 were significantly older with 48.3% surpassing 70 years of age as opposed to 30.6% in the pre-2018 period (P = 0.015).Additionally, post-2018 individuals frequently received lower-intensity therapies including HMA + /-venetoclax or CPX-351.Furthermore, individuals with AML (blasts ≥20%) were significantly more likely to receive VEN-based regimens compared to patients with MDS/AMLs (P = 0.003) at diagnosis.
TP53 NMIS mutations are frequently associated with multi-hit allelic state and autosomal monosomies Mutational analysis identified 306 pathogenic mutations (200 unique mutations) among the 245 patients comprising mostly hot-spot DNA-binding domain TP53 MIS mutations (Fig. 1D).A total of 29.8% (73/245) harbored non-missense TP53 mutations, either singly or as part of multiple TP53 mutations.Individuals with single TP53 mutations were significantly more likely to harbor loss of chromosome 17p either due to monosomy 17 or structural losses of 17p (44.5% vs. 26.8% in multiple TP53 mutations; P = 0.019).These files were examined using the RCytoGPS package for R, treating them as a binary matrix, to create the ideogram [36].D Summary of TP53 mutations in the cohort.Somatic variants in TP53 visualized using lollipop plot generated via the ProteinPaint web-based application [37].Majority comprised hot-spot DNA-binding domain missense mutations.Chromosomal position coordinates were culled from the IARC database of TP53 mutations.A few complex mutations seen in a few cases are not depicted and numbers may not match up with that depicted in the results section.
We constructed an 'EPI6' signature based on alterations cooccurring in ≥4 individuals and defined by the presence of mutations or deletions in at least one of six key genes associated with methylation (TET2, EZH2), mitogen-activated protein kinase (KRAS), signaling (CBL), splicing (U2AF1) pathways or CUX1.We selected these genes based on their frequency of occurrence and impact in initial exploratory univariable analysis.This EPI6 signature was present in 24.8% (60/242) of individuals with a significantly higher proportion harboring TP53 VAF > 25% (88.3% vs. 74.7%with VAF > 25% in individuals lacking EPI6; P = 0.027).We used this EPI6 signature going forward in our analyses of treatment outcomes and long-term survival.
Response data: CUX1 alterations and ≥2 autosomal monosomies predict inferior CR1 First-line treatment and response information.Among the 234 patients with available treatment information, low-intensity regimens (mostly HMA-based) were used in 148 (63.2%), intensive chemotherapy in 22.2% (N = 52), CPX-351 in 12 (5.1%)while 22 (9.4%) received only best supportive care.Treatment information was not available in 8 since these patients were either lost to follow up or experienced early mortality before commencing Fig. 2 Co-mutations and structural alterations in patients with any co-mutations.Only variants occurring at least twice within the cohort are represented.Notably, structural alterations/losses were common with certain genes, particularly CUX1, EZH2, and APC.
treatment.Among HMA-based therapies (N = 144), HMA (with or without magrolimab) was used in 60 (41.7%) while HMA + VEN was used in 58.3% (N = 84).A total of 14.5% (35/242) went on to receive allogeneic stem cell transplantation with 69.7% receiving myeloablative conditioning regimens.First-line response was evaluable in 195 patients with 25.6% achieving CR, 24.1% with partial response, and 50.3% with non-response/stable-progressive disease.A small subset of patients could not be evaluated for treatment response because of treatment discontinuation due to treatment-emergent adverse effects, developed active second non-hematopoietic malignancies, or were lost to follow-up shortly after diagnosis.These patients were excluded from all response analyses but were included in the analysis of 24-month overall survival (OS24).

Baseline outcome data
The median duration of follow up from diagnosis of TP53 MUT MN to study exit was 6.1 months (range: 0.2-72.8months) with 202 deaths in 242 patients and a death rate of 87.0 per 1000 patientyears with a 24-month survival of 16% (95% CI = 11.3-21.5%).
While all TP53 VAF cutoffs (>10%, >25%, >40%, and >50%) predicted inferior OS24, we selected VAF > 25% (HR = 1.8 [1.2-2.6];P = 0.002) for all subsequent analysis based on the balance of cases across both groups for an appropriately powered analysis while also assessing the Youden index that reflected maximizing separation (Fig. 3D).Intriguingly, the beneficial impact of a lower VAF within each of these binary cut-point was restricted only to male gender when examined separately by gender (Supplementary Fig. 2) although the analysis was slightly underpowered.
TP53 Risk Score (TP53RS) for TP53 MUT myeloid neoplasms for patients treated with HMA-based therapies Multivariable model for entire cohort.Based on the predictors relevant in univariable analysis, a multivariable model for the entire cohort was constructed using 5 predictors including antecedent treated myeloid neoplasm, TP53 VAF, TP53 sjm , monosomies and EPI6 signature.The models include regular Cox, 45-day landmark Cox, and a competing risk model modeling only leukemia-specific deaths in the competing risk model (See Table 2 and Fig. 5A for details).
Multivariable model and Risk score for HMA-treated subgroup.Since HMA-based therapies are the most predominant frontline choice in this cohort, we developed a separate risk score for this subgroup excluding TP53 VAF and monosomies (which were relevant only in intensively-treated subgroups as noted earlier) (See Table 3

for details).
This TP53 risk score (TP53RS) for HMA-treated individuals delineated three risk groups comprising intermediate (0 factors, TP53RS0), intermediate-poor (1 factor, TP53RS1), and poor (2+ factors, TP53RS2) risk groups with significantly different median survival (12.8 vs. 6.0 vs. 4.3 mos.; P Log-rank < 0.001) as well as 24-Fig.3 Univariable K-M plots of key pre-therapy adverse predictors.A Antecedent myeloid neoplasm, B Autosomal monosomies and D. Highest TP53 VAF ≥ 25%, all adversely impact overall survival (OS24).C TP53 splice junction mutations were significantly associated with male gender and were associated with worse survival compared to all other classes of TP53 mutations.Patients with antecedent treated myeloid neoplasm included individuals receiving HMA as well as supportive care (transfusions).Only TP53 VAF and EPI6 remained relevant in all three models with monosomies retaining relevance in the regular and 45-day landmark models while TP53sjm and antecedent treated myeloid neoplasm were less relevant although adverse in multivariable analysis too despite not being statistically significant.Importantly EPI6 had independent prognostic value beyond all predictors in the multivariable model.Hazard ratio (HR) for competing risk model denotes Sub-Hazard Ratio.***p < 0.01, **p < 0.05, *p < 0.1.
Fig. 4 Univariable adverse impact of EPI6 genes and outcome.A Forest plot of the hazard ratios of each of the six "EPI6" genes in the OS24 analysis with numbers and frequencies of occurrence depicted on the right.There were gender specific differences in some of the genes and hence these genes were included in the EPI6 signature despite borderline adverse significance in the combined analysis.B Patients carrying co-mutations/alterations in any of the 6 "EPI6" genes including CBL, CUX1, EZH2, TET2, KRAS, or U2AF1 experienced shorter median and overall survival.CUX1 was the most influential gene in the EPI6 signature with overall survival of CUX1 altered individuals approaching 0% by 12 months after diagnosis.
Sensitivity analysis excluding low TP53 VAF patients.Since inclusion of cases with TP53 VAF < 10% could potentially inflate the impact of the proposed score, we performed an additional sensitivity analysis excluding 17 (7.0%)such individuals.TP53RS retained its prognostic value in this last analysis as well (HR = 1.8 [1.3-2.5];P < .001;N = 133).

DISCUSSION
While most prior studies focused on comparing the impact of pretherapy predictors (viz.any TP53 mutations or TP53 VAF) in Fig. 5 Impact of TP53 risk score with relevant predictors on OS24 by flexible parametric analysis in entire cohort and HMA subgroup.
A Entire cohort analysis utilizing TP53 VAF and monosomies in addition to antecedent treated myeloid neoplasm, TP53 sjm and EPI6 separates three prognostic groups corresponding to risk score cutoffs of 0 factors, 1-2 factors, and 3+ factors.B TP53 risk score for HMA-treated patients using parsimonious 3-parameter model in sensitivity analysis agnostic to monosomies and VAF information.
myeloid neoplasms across all risk strata, our cohort is remarkable for being the first to look within just the ELN 2022 high-risk subgroup.We identified several novel factors that reliably predict poor response to first-line therapy (monosomal karyotype, multihit TP53 allelic state, CUX1 deletions) and inferior survival including a novel EPI6 co-alteration signature including CUX1.We specifically focused on developing a model/risk score solely with pre-therapy predictors rather than well-known therapy-related factors (such as CR1 and transplant).While the proportion of patients treated with chemotherapy was still small in our cohort, it is remarkable that TP53 allelic state, TP53 VAF and monosomies were all predictive for survival only in the intensively treated subgroup of patients.While the 25% TP53 VAF cutoff in our study was admittedly data-driven, it was nevertheless in line with the 23% cutoff proposed by Bahaj and coworkers [22] as well as the 25% cutoff used by Grob and coworkers [5] although this latter study was an exclusively chemotherapy-treated, and slightly younger cohort.In comparing the impact of VAF by treatment subgroups, we note as well that TP53 VAF is relevant only within the intensive chemotherapy treated subgroups, in line with prior studies [17,23].We were however surprised by the higher CR rates in females possibly attributable to less frequent TP53 MH in females.Whether additional fitness-related factors or better compliance in females play a role remains to be seen.
While TP53 allelic state did not impact overall survival, it nevertheless remains useful in assessing response to front-line therapy (more so in patients receiving non-intensive therapies).Looking at the different TP53 mutation classes, patients with any TP53 NMIS mutation face significant early mortality.Two prior studies [24,25] evaluating MDS demonstrated worse OS with TP53 NMIS mutations compared to TP53 MIS with the latter looking at posttransplant survival.However, an AML study from the German-Austrian AMLSG did not find any impact on OS based on mutation class (but with a trend towards worse event-free survival with TP53 NMIS ) in a cohort treated primarily with intensive chemotherapy [26].In our cohort, TP53 splice junction mutations were associated with the worst outcomes compared to all other mutations.While they occurred much more frequently in men without any impact on CR1 rates, their adverse impact on survival was much more pronounced in women.These findings are congruent with recent demonstration of gender specific differences in the frequencies of AML-associated genetic alterations and less frequent complex karyotype and mutation in genes related to the spliceosome machinery in women (including U2AF1 and SRSF2) [27].
On the other hand, the predictive relevance of autosomal monosomies is quite remarkable and admittedly somewhat surprising in our high-risk cohort with complex karyotype.While it has long been known that monosomal karyotype is adverse and often enriched in complex karyotype AMLs [4,28] with frequent del(17p) [29], these cohorts comprised a heterogeneous mix of AML patients of all risk groups without specifically focusing on TP53-mutated individuals.Considering the high frequency of complex karyotype in our cohort, we restricted the analysis to monosomies only while ignoring structural alterations which were near-universal.Importantly, monosomy 7 was more relevant than monosomy involving chromosomes 5 or 17.We observed that the predictive value of 2+ monosomies for inferior CR1 was restricted only to the subgroup receiving non-intensive therapies underscoring its relevance in an increasingly HMA-treated elderly cohort.However, neither monosomies, nor TP53 VAF nor allelic state impacted overall survival in the HMA-treated subgroup.
In the context of the EPI6 signature, CUX1 emerged as the most pivotal gene.While previous studies [30,31] have highlighted the detrimental effects of CUX1 deletions and mutations in AML, their significance in TP53 MUT MN has not been evaluated.Notably, in our cohort, CUX1 losses (which were more prevalent than mutations) were observed in 10% of cases.These losses identified a subgroup with an abysmal response rate (0%) to frontline treatment with significantly worse 2-year survival (Fig. 4A) particularly among females.It is important to emphasize that while CUX1 alterations frequently co-occurred with losses of chromosome 7/7q [32], isolated losses at this genomic locus were still associated with adverse clinical outcomes, underscoring their independent relevance beyond karyotypically detectable -7/7q.Our observations of RAS pathway co-alterations (CBL and KRAS) in the EPI6 signature aligns with the documented activation of this pathway in CUX1-altered MNs, as recently demonstrated in the context of 7q alterations in myeloid neoplasms [33].Consistent with a previous study by Badar and coworkers which identified a significant association between TP53 mutations and the Q157 variant in U2AF1-mutated MNs, our cohort also exhibited a marked predominance of the U2AF1 Q157 variant compared to the S34F variant [34].
We note as well that blast counts at diagnosis are irrelevant once there are more than 9% blasts, affirming that MDS/AML and AML with TP53 MUT are indeed a single biologic entity.That said, establishing a diagnosis of morphologic CR or morphologic leukemia-free state (MLFS) based on blast counts after therapy however is particularly problematic in TP53 MUT MN where significant erythroid-predominant leukemic hematopoiesis (often CD34-/strong p53 + ) is frequently observed.Post-therapy marrow biopsies frequently show less than 5% CD34-expressing blasts with a significant component of strong p53-expressing erythroid component (frequently corresponding to an underlying TP53 mutation) highlighting that neither blast count nor CD34 are good surrogates for pathologic CR or MLFS in TP53 MUT MNs despite significant cytoreduction.To this end, the fast turnaround time of IHC compared to all the other tests (NGS and karyotype) makes it an attractive surrogate for most if not all non-truncating TP53 mutations [35], particularly in elderly and therapy-related AMLs at diagnosis and after therapy.
Two major limitations of our study include 1] lack of molecular MRD in response assessment and 2] lack of an external validation cohort.To the second point, most published AML and MDS cohort lack public data on alterations (especially losses) at the CUX1 locus hampering their use as a validation cohort for the EPI6 signature.Furthermore, prior studies included higher proportions of intensively treated patients limiting their use as a representative validation cohort even if there was available NGS data.Lastly, we were unable to ascertain if some of the observed TP53 mutations in patients with low VAF merely corresponded to age-related clonal hematopoiesisassociated mutations unrelated to the main clone (see Supplementary Fig. 8 for analysis of the missense TP53 substitutions and TP53 domain relevance in the University of Chicago cohort).Importantly, our work underscores that not all classes of TP53 mutations are equal given the adverse outcomes with splice junction mutations and early mortality with any TP53 NMIS mutations.
In the real-world setting, all candidate components of the proposed risk score (NGS, FISH, karyotype) are typically available within two weeks of diagnosis.Furthermore, most candidate genes of EPI6 are common genes included in standard NGS panels, except perhaps CUX1, for which we recommend an algorithm in the NGS pipeline for calling CUX1 loss detection.Among HMA-treated individuals, patients in intermediate/TP53RS0 risk group with 0−1 monosomies enjoy a 63% CR1 (vs.16% CR1 with 2+ monosomies; P < 0.001) and are most likely to respond significantly to frontline therapies and have the best chance at alloSCT with durable post-transplant survival.However, most other patients in intermediate-poor/TP53RS1 and poor/TP53RS2 groups still face early mortality due to comorbidities, active second malignancies or treatment-emergent adverse effects including neutropenic sepsis.As a result, these latter groups derive little to no benefit with existing therapies, highlighting the need for the developing novel agents with better safety and efficacy profiles for these subgroups.
In conclusion, the several novel proposed pre-therapy biologic predictors (CUX1 deletions, allelic state, TP53 splice junction mutations, TP53 VAF, autosomal monosomies, and specific comutation patterns) as well as antecedent treatment (HMA exposure) for a myeloid neoplasm have differential prognostic utility depending on treatment subgroups (intensive vs. HMAbased).These data will better inform frontline response and aid in identifying the best candidates for alloSCT in this high-risk, elderly cohort.
Figures and manuscript.HVS.Visiting pathologist curated data from IARC and EAP53 server.SYY collated all TP53 mutations for lollipop plot generation, BA, MS, JLP, ABP and MPM contributed cases, did chart review and mutation curation of contributed cases.MT, JS and PW curated mutation data and examined pipeline for NGS data.AL and CF reviewed cytogenetic data.AP, MWD, CS and JEC, TK, provided patient care, accrued patients on clinical trials (HMA/Magrolimab, Gilead), designed some analyses, and edited the manuscript.MEM reviewed the data related to CUX1 biology and edited those portions of the manuscript.LW, EH, DA, JXC, SG, SBG, and GV reviewed pathology data and edited final version of manuscript.OL and PK analyzed the TP53 missense mutations in the University of Chicago subset reported in the supplementary data.PW and PW analyzed the CNA data.GV initiated, designed, supervised study, conducted all statistical analysis, & wrote the manuscript.All authors edited the manuscript.

Fig. 1
Fig. 1 Summary of cytogenetic alterations and TP53 mutations.A Frequencies of autosomal monosomies (percentages represent frequencies in patients with available data on karyotype).The most frequent autosomies involved chromosomes 7, 17 and 18. B Cooccurrence matrix of autosomal monosomies.Counts indicate frequencies of co-occurrence.The most frequently co-occurring monosomies with loss of 17 were monosomies 7 and 16, followed by monosomies of chromosomes 5, 6 and 12. C Ideogram of global losses (orange) and gains (green) across the genome.Recurrent losses are enriched in chromosomes 5, 7 and 17.Furthermore, we observed additional recurrent losses on 12p, besides 16q, 18p, and 18q.Chromosomal gains were prevalent, particularly on chromosomes 8 (trisomy 8), as well as on chromosomes 1, 9, 11, 21, and 22. Clean karyotypes were batched parsed in CytoGPS to create .jsonfiles with loss, gain & fusion information.These files were examined using the RCytoGPS package for R, treating them as a binary matrix, to create the ideogram[36].D Summary of TP53 mutations in the cohort.Somatic variants in TP53 visualized using lollipop plot generated via the ProteinPaint web-based application[37].Majority comprised hot-spot DNA-binding domain missense mutations.Chromosomal position coordinates were culled from the IARC database of TP53 mutations.A few complex mutations seen in a few cases are not depicted and numbers may not match up with that depicted in the results section.

Table 1 .
Baseline characteristics of all 242 patients at diagnosis of TP53 MUT Myeloid Neoplasm stratified by TP53 allelic state at diagnosis (see methods section for definition of allelic state).
a These included treated or untreated MDS (Low and High-risk), MDS/MPN and MPN without any TP53 MUT up until evolution/progression to a TP53 MUT myeloid neoplasm.b

Table 2 .
Multivariable models fitted incorporating the four relevant pretherapy predictors adjusted for age at diagnosis and a clustered sandwich estimator for standard errors accounting for clustering across contributing centers.

Table 3 .
Multivariable (Cox, 45-day landmark Cox and competing risk) subgroup analysis in HMA+/-Ven treated individuals including only three relevant baseline predictors adjusted for age at diagnosis.