Prognostic implications of TET2 variant allele frequency in chronic myelomonocytic leukemia: a multicenter study

Key Points

• •This is, to our knowledge, the first large-scale quantitative characterization of TET2 VAF as a continuous prognostic variable in CMML.
• •A per-integer VAF adjustment refines three contemporary CMML prognostic tools; a freely accessible web calculator implements the framework.

Visual Abstract

Abstract

Chronic myelomonocytic leukemia (CMML) is a heterogeneous myeloid neoplasm with highly variable prognosis. TET2 is mutated in ∼60% of patients, previously linked to favorable outcomes. However, no study has quantitatively characterized TET2 variant allele frequency (VAF) as a continuous, nonlinear prognostic variable, nor leveraged it to refine contemporary CMML risk stratification systems. We examined the prognostic role of TET2 VAF in 1206 patients across a UK discovery cohort (DC; TET2-mutated, 433/645 [67%]), an international validation cohort (validation cohort 1 [VC1]; TET2-mutated, 141/291 [48%]), and a French CD14⁺ monocyte-sequenced cohort (VC2; TET2-mutated, 176/270 [65%]). Segmented regression of TET2 VAF distributions reproducibly identified two inflection points (DC, 32%/55%; VC1, 33%/54%; VC2, 36%/56%), implying consistent subdivision of TET2 mutation burden across CMML of epidemiological (and potentially biological) relevance. Restricted cubic spline Cox regression revealed a nonlinear relationship between TET2 VAF and survival, with risk lowest near VAF 39% (the empirical first quartile), whereas patients with TET2 wild-type status had hazards comparable to patients with high VAF. Capitalizing on this continuous relationship, we constructed TET2-VAF–adjusted CMML-specific Prognostic Scoring System–Molecular, Groupe Francais des Myelodysplasies model, and Mayo Molecular Model scores by adding a per-integer VAF adjustment to each patient's original score. Across cohorts and prognostic tools, the adjusted scores improved discrimination (Harrell C-index gain in 16/18 comparisons; integrated discrimination improvement significant in all 18 comparisons; P < .01) and group-level stratification for overall survival and acute myeloid leukemia–free survival. These findings suggest that TET2 VAF represents a continuous prognostic variable in CMML and provide a robust, reproducible framework for refining standard risk stratification systems.

Introduction

Chronic myelomonocytic leukemia (CMML) is a clonal myeloid neoplasm with overlapping features of myelodysplasia and myeloproliferation with persistent monocytosis.^1,2 Pathogenesis results from the cumulative impact of somatic mutations in epigenetic, splicing, and signaling genes, with ∼30% of patients destined to transform to acute myeloid leukemia (AML). Although prognosis is often poor, clinical outcomes are highly variable, with some patients experiencing indolent disease courses and long-term survival. Predicting who is destined to progress rapidly and for whom transplant or active therapy is thereby justified remains a major clinical challenge.

Numerous prognostic scoring systems have been developed for CMML, incorporating overlapping but distinct sets of clinicopathologic, laboratory, and other parameters.¹ Direct comparison of 7 systems revealed similar performance, with no clear winner, but none providing optimal predictive performance.³ Notably, only 3 tools (CMML-specific Prognostic Scoring System–Molecular [CPSS-Mol]; Groupe Francais des Myelodysplasies [GFM] model; and Mayo Molecular Model [MMM]) incorporate somatic mutations,4, 5, 6 with 2 including ASXL1 alone.^4,5 Recently, the International CMML Prognostic Scoring System (iCPSS) extended molecular contribution to 9 genes based on comprehensive machine learning on a large retrospective cohort (n = 3013) and was validated prospectively (n = 516).⁷

Among the new genes of prognostic relevance in the iCPSS is TET2, which encode an epigenetic modifier and is the most common recurrent mutation in CMML (present in ≥60% of patients).¹ Previous studies had suggested a putative protective effect on prognosis for patients with TET2-mutated (TET2-mut) status compared with those with TET2 wild-type (TET2-WT) status.8, 9, 10, 11, 12, 13, 14 Accordingly, the iCPSS confers favorable risk weighted to TET2-mut status in its prognostic model. However, to date, most studies have considered binary TET2 mutation status only (some in compound conjunction with ASXL1 status), generally ignoring mutation number and allelic status. A recent exception, a study by Csizmar et al,¹⁴ tested TET2 variant allele frequency (VAF) as a linear continuous variable in Cox regression and reported no association with prognosis, motivating the nonlinear approach presented here.

Complete loss of TET2 activity leads to a more pronounced phenotype than that observed in patients with one intact allele, a gene dosage effect that reflects graded loss of TET2 dioxygenase activity.¹⁵ This effect was recently confirmed at single-cell resolution.¹⁶ This observation is supported by Coltro et al,⁹ who reported that TET2 mutations were associated with a survival advantage in CMML, which was most pronounced in cases with multiple mutations and truncating variants located upstream of the catalytic domain (ie, early truncations that ablate both the N-terminal and catalytic regions). Additionally, biallelic TET2 mutations are associated with poor prognosis in AML¹⁷; the clonal hierarchy of TET2 mutations is shaped by founder acquisition and subsequent biallelic events¹⁸; and a TET2 VAF ≥32% predicts adverse outcomes in myelodysplastic syndromes (MDS).¹⁹

Despite these observations, the prognostic role of TET2 VAF, including biallelic mutations and loss of heterozygosity (LOH), remains unexplored in CMML, in which TET2 is the most frequently mutated gene. TET2 mutation burden (beyond status alone) therefore likely harbors prognostic nuance, hitherto unstudied. Here, we investigated the prognostic impact of TET2 mutations, incorporating the percentage VAF as a measure of mutation dosage (including as a surrogate marker of allelic status) in 1206 patients with CMML from 7 institutions (across 6 countries), to evaluate its potential to refine and enhance existing prognostic systems.

Methods

Study cohorts

Our UK discovery cohort (DC) comprised 645 consecutive patients with CMML diagnosed and treated across North West England between 2009 and 2023 and for whom extensive clinical, molecular, and outcome data were available. A full karyotype was available for 598 patients (93%). Separately, we pooled 291 patients with CMML diagnosed and treated at the National Taiwan University Hospital (Taipei, Taiwan),²⁰ Vanderbilt University Medical Center (Nashville, TN), Cleveland Clinic (Cleveland, OH),²¹ Hospital Morales Meseguer (Murcia, Spain),²² and Vilnius University Hospital Santaros Klinikos (Vilnius, Lithuania), together forming validation cohort 1 (VC1). Next-generation sequencing (NGS) was performed on diagnostic bone marrow (BM) samples. An additional 270 patients with CMML from Gustave Roussy Cancer Center (Villejuif, France) were recruited as VC2. NGS data for VC2 were derived from CD14⁺ monocytes, sorted from mononuclear cells (MNCs) using CD14 microbeads, as described previously.²³ All diagnoses conformed to 2022 World Health Organization classification criteria.² This study was approved by the institutional review boards of each participating hospital and conducted in accordance with the Declaration of Helsinki.

Targeted mutation sequencing

In the UK DC, targeted gene panels and sequencing platforms varied across institutions, with 27 core shared myeloid neoplasm–relevant genes (supplemental Table 1). NGS was performed as previously described.²⁴ Similarly, in VC1 panels and NGS, platforms varied across institutions,^21,22,25 with 27 common mutations profiled across centers (supplemental Table 1). The NGS panel for VC2 is presented in supplemental Table 1.²³ Variants were annotated using international databases of normal alleles and pathologic mutations, with variants of uncertain significance excluded from analysis. For multiple TET2 mutations, hits were ordered by VAF values. Recognizing that TET2 mutations can manifest in various clonal states, we focused on the mutation with highest VAF for outcome analyses. Because of limited data on copy number alterations and confirmed LOH, these were not directly considered (although LOH was inferred by high VAF percentage, as described in “Results”).

Statistical analyses

Categorical variables were compared using the χ² or Fisher exact tests, and continuous variables were compared using the Mann-Whitney U or Kruskal-Wallis tests. Pearson correlation was used to evaluate associations between TET2 VAF and continuous clinical variables. Survival outcomes (overall survival [OS]; AML-free survival [LFS]) were analyzed by the Kaplan-Meier (KM) method with log-rank comparison. Univariate and multivariable analyses used Cox proportional hazards models. Distribution of TET2 VAF was characterized by segmented regression to detect inflection points, applied independently in DC, VC1, and VC2. To assess the prognostic relationship of TET2 VAF as a continuous variable, we fitted restricted cubic spline (RCS) Cox regression models for both LFS and OS. Hypomethylating agent (HMA) treatment response was assessed in the DC using the International Working Group consensus response criteria for myelodysplastic/myeloproliferative neoplasms.²⁶

Development of the TET2-VAF–adjusted scoring systems

A per-integer VAF adjustment value was generated for each scoring system (CPSS-Mol, GFM, and MMM) by averaging the log hazard ratios (HR) from the RCS Cox models for OS and LFS at each integer VAF value (1%-100%) relative to the reference and then dividing by the corresponding system-specific univariable Cox coefficient. For patients with TET2-WT status, a single fixed adjustment was derived from a binary Cox model comparing TET2-WT to mutated status within the reference zone, again averaged for OS and LFS and normalized by the system-specific coefficient. The TET2-VAF–adjusted score for each patient was calculated as follows: original score + (VAF-derived or WT adjustment). Risk category cut points for the TET2-VAF–adjusted scores were rederived using a likelihood ratio–optimized (LR-opt) approach, scanning candidate cut points to maximize the joint OS and LFS log-rank χ² statistic; the number of risk categories was preserved from each parental system (CPSS-Mol, 4; GFM, 3; MMM, 4).

Performance evaluation

Discrimination was assessed using the Harrell C-index and time-dependent area under the curve (AUC; paired DeLong). Reclassification was assessed using integrated discrimination improvement (IDI) and net reclassification improvement (NRI), both with bootstrapped 95% confidence intervals (CIs; 1000 resamples). Group-level discrimination was further compared between the original and TET2-VAF–adjusted classifications using the log-rank χ² statistic. Calibration was assessed visually using calibration plots. A P value <.05 was deemed significant. Analyses were performed using R version 4.4.0 and Prism 10.

Use of primary human tissue was in compliance with the ethical and legal framework of the UK Human Tissue Act 2004. Primary samples were obtained from the Manchester Cancer Research Centre’s Tissue Biobank (instituted with the approval of the South Manchester Research Ethics Committee [18/NW/0092] and licensed by the Human Tissue Authority [license number 30004]). Use of the samples was authorized following ethical review by the Tissue Biobank’s scientific subcommittee (approval 17_KIBA_01). The National Taiwan University Hospital (NTUH) Research Ethics Committee approved the study (number 201709072RINC).

Results

Patient characteristics and mutational landscapes

The median age of the 645 UK DC patients was 75 years (range, 24-98), with a male predominance (63%). According to World Health Organization 2022 criteria, the cohort included 410 patients (63.6%) with dysplastic and 235 patients (36.4%) with proliferative CMML; 523 (81.1%) had CMML-1, and 122 (18.9%) had CMML-2. Cytogenetic and molecular characteristics, alongside the equivalent details of VC1 and VC2, are summarized in Table 1 and supplemental Table 1. TET2 was the most frequently mutated gene (433/645 [67.1%]), followed by SRSF2 (47%), ASXL1 (38.6%), RUNX1 (19.5%), and NRAS (15.3%). As expected, co-occurrence analysis revealed significant co-occurrence of TET2/SRSF2 and IDH2/SRSF2 mutations (both P < .001; Figure 1A), as well as mutual exclusivity of SRSF2 with other splicing factor genes (SF3B1, U2AF1, and ZRSR2) and EZH2 (all P < .001).

Table 1.

Clinical and laboratory features at diagnosis of patients with CMML in the DC, VC1, and VC2

Variable	DC (n = 645)	VC1 (n = 291)	VC2 (n = 270)
Age, y∗	75 (24-98)	71 (21-96)	74 (42-93)
Sex, n (%)
Female	236 (36.6)	95 (32.6)	89 (33)
Male	409 (63.4)	196 (67.4)	181 (67)
Laboratory data∗
Leukocytes, ×109/L	12.4 (2.1-171.5)	9.3 (1.2-110)	14.2 (2-161)
Hemoglobin, g/dL	10.4 (5.8-15.4)	10.3 (3-17)	11.5 (4.8-17.2)
Platelets, ×109/L	83 (13-802)	98 (1-1277)	109 (9-996)
Circulating blast, %	0 (0-14)	0 (0-17)	0 (0-18)
BM blasts, %	3 (1-18)	4 (0-19)	5 (0-19)
WHO subgroup, n (%)
Dysplastic	410 (63.6)	175 (60.1)	111 (41.1)
Proliferative	235 (36.4)	116 (39.9)	159 (58.9)
CMML-1	523 (81.1)	224 (77)	199 (73.7)
CMML-2	122 (18.9)	67 (23)	71 (26.3)
TET2 mutation number
0	212	150	94
1	215	71	80
2	200	67	75
3	16	3	15
4	2	0	5
5	0	0	1
Risk stratification†,‡, n (%)
MMM risk
Low	118 (18.3)	26 (9.1)	65 (24.1)
Intermediate-1	206 (31.9)	84 (29.3)	69 (25.6)
Intermediate-2	197 (30.5)	90 (31.4)	69 (25.6)
High	124 (19.2)	87 (30.3)	67 (24.8)
GFM risk
Low	236 (36.6)	123 (42.4)	96 (35.6)
Intermediate	259 (40.2)	114 (39.3)	96 (35.6)
High	150 (23.3)	53 (18.3)	78 (28.9)
CPSS-Mol risk
Low	21 (3.3)	48 (16.6)	29 (10.7)
Intermediate-1	113 (17.5)	61 (21.1)	51 (18.9)
Intermediate-2	297 (46)	55 (19)	107 (39.6)
High	167 (25.9)	125 (43.3)	83 (30.7)
Cytogenetic abnormalities‡,§, n (%)
Low risk	489 (81.8)	207 (71.6)	206 (76.3)
Intermediate risk	57 (9.5)	45 (15.6)	29 (10.7)
High risk	52 (8.7)	37 (12.8)	35 (13.0)

Statistically significant if P value is <.05.

WHO, World Health Organization.

^∗Presented as median (range).

^†Overall, 598 patients in the DC had cytogenetic data available for calculation of CPSS-Mol risk scores.

^‡In VC1, 289 patients had available CPSS-Mol data, 290 had available GFM data, and 287 had available MMM data.

^§Low risk was defined as normal karyotype and −Y; high risk was defined as trisomy 8, a complex karyotype, or abnormalities of chromosome 7; intermediate risk was defined as abnormalities other than those in low- and high-risk categories.

Figure 1.

Mutations profiles and VAFs of TET2 within 645 patients with CMML in the DC. (A) Mutation co-occurrence plot represents the co-occurrence and mutual exclusivity of mutations in the targeted NGS panel for 645 patients with CMML in the DC. The color scale indicates the significance of the co-occurrence or mutual exclusivity (measured as −log10(P value)), with shades of green indicating co-occurrence and shades of brown indicating mutual exclusivity. Asterisks (∗) indicate highly significant associations (P < .001), whereas dots (•) denote significant associations (P < .05). (B) Proportions of patients harboring different numbers of TET2 mutations (upper) and stacked bars illustrating the distribution of different variant types of TET2 within patient groups bearing different numbers of TET2 mutations (lower). (C) Violin plots depicting VAFs of TET2 mutations. The dotted line indicates the median value and the upper and lower quartile values. (D) Histogram showing the distribution of TET2 VAF among patients with TET2-mut status. The dashed line indicates the median VAF. (E) Landscape plot displaying individual TET2 VAFs among patients with TET2-mut status (red line), sorted in ascending order. Inflection regression revealed inflection points at VAFs of 32% and 55%. (F) RCS Cox regression analysis of TET2 VAF as a continuous variable for LFS (left) and OS (right) in the DC. The solid line represents the smoothed HR, with the shaded area indicating the 95% CI. The dashed horizontal line represents the HR for TET2-WT patients. The reference VAF was set at 39%, the empirical first quartile of VAFs. ID, identity.

A total of 671 TET2 mutations were identified in 433 patients (Figure 1B), with 218 (50.3%) carrying multiple TET2 mutations: 200 (46.2%) with 2; 16 (3.7%) with 3; and 2 (0.5%) with 4. Frameshift was the dominant variant type (45.9%), followed by nonsense (32.9%), missense (18%), splice site (2.7%), and in-frame deletion (0.5%) (Figure 1B). Median VAFs of the first, second, third, and fourth TET2 hits were 45.9%, 33.9%, 7.7%, and 6.1%, respectively (Figure 1C). The histogram of dominant-hit TET2 VAFs revealed a bimodal distribution, with a primary peak at 40% to 50% and a secondary peak at 90% to 100%, consistent with cases of TET2 LOH (Figure 1D). Pearson correlation showed a very modest positive correlation between dominant-hit TET2 VAF and white blood cell count (R² = 0.108; P < .001) but no significant correlation with age, hemoglobin, platelets, blasts, or total mutation number (supplemental Table 2). To address the distribution of TET2 VAF among patients with TET2 mutilations with co-occurring driver mutations, we compared dominant-hit TET2 VAF across mutation status for each of 21 frequently mutated genes (supplemental Table 3). Four genes showed nominal differences: CBL (higher VAF in mutated cases; P = .002), DNMT3A (lower; P = .018), NRAS (higher; P = .044), and SRSF2 (higher; P = .044), whereas only CBL remained significant after Bonferroni correction for 21 comparisons; the remaining signals should be considered exploratory. The other 17 comutations, including ASXL1 and RUNX1, showed no association.

Over a median follow-up of 21.1 months, 120 patients (19%) experienced AML transformation, and 391 patients (61%) died. The median LFS and OS were 27.8 months and 28.8 months, respectively. As expected, stratification of LFS and OS by the 3 established prognostic models was robust (supplemental Figure 1A-C). Patients harboring ≥2 TET2 mutations exhibited significantly better LFS and OS than those with a single mutation or TET2-WT (TET2 mutation number, ≥2 vs 1 vs 0; LFS, 38.1 vs 28.8 vs 17.9 months [P < .001]; OS, 39.3 vs 30.9 vs 21.8 months [P < .001]; supplemental Figure 2A), in line with prior observations.⁹ Among TET2-mut cases, those harboring any truncating mutation (n = 384) demonstrated a trend toward better LFS than those with nontruncating mutations (n = 48; median, 37 vs 24 months; P = .056),⁹ although no significant difference in OS was observed (median, 37 vs 24 months; P = .457). Overall, the DC reflects a typical CMML population with broadly representative clinical, genomic, and outcome features, albeit with slightly higher rates of TET2 mutation than in some other reports.^4,6,9 Furthermore, CPSS-Mol low-risk cases were relatively less represented than in the original CPSS-Mol derivation study (14%-19%)⁶ and a recent Mayo Clinic/MD Anderson cohort (9.8%).¹⁴

TET2 VAF distribution and inflection points

We next sought to explore the prognostic dimension beyond mutation number, namely, the VAF of dominant-hit TET2 mutations. Examination of the distribution of dominant-hit TET2 VAFs across the 433 patients with TET2-mut status revealed two clear inflection points in the landscape plot (Figure 1E). Segmented regression confirmed inflections at 32% and 55% VAFs (supplemental Table 4), the latter consistent with biallelic TET2 inactivation by LOH^21,27,28; supporting an LOH origin, the proportion of single-TET2 mutation cases sharply increased at VAF ≥55% (59/72 [82%]) compared with VAFs of 32% to 55% (122/302 [40%]) and <32% (34/59 [58%]; 3-group χ² P < .0001), indicating that high VAF arises predominantly from copy-neutral LOH in single-hit cases rather than independent acquisition of multiple mutations (supplemental Table 5). In the small subset (n = 5) with paired karyotype/microarray data, 4 had VAF of ≥55%. We therefore use VAF of ≥55% as a working threshold for inferred biallelic TET2 inactivation throughout, while emphasizing that direct copy number confirmation through dedicated techniques remains the gold-standard approach.

Phenotypic correlates of TET2 VAF groups

We next examined CMML subtype distribution across the 4 inflection-defined TET2 VAF groups: TET2-WT (n = 212), TET2-low (<32% VAF; n = 59), TET2-Intermediate (Int) (32%-55%; n = 302), and TET2-high (≥55%; biallelic inferred; n = 72). Proliferative CMML was substantially less frequent among patients with TET2 mutations (31.4%) than among patients with TET2-WT status (46.7%; χ² P = .0002), with the lowest proportion in TET2-low (20.3%) and similar levels in TET2-Int (33.1%) and TET2-high (33.3%); the global 4-group difference was significant (χ² P = .0005). Among patients with TET2-mut status, ASXL1 co-mutation tracked with a proliferative phenotype: 27.1% of patients with TET2-mut/ASXL1-WT status vs 40.4% of those with TET2-mut/ASXL1-mut status (P = .005). The proportion of patients with a proliferative phenotype did not differ between those with single vs multiple TET2 mutations (33.0% vs 29.8%; P = .48) or between those with biallelic-inferred (≥55%) and monoallelic status (<55%; 33.3% vs 31.0%; P = .70). Detailed crosstabulations are summarized in supplemental Table 6.

TET2 VAF is a nonlinear continuous prognostic variable

Having established phenotypic correlates of the inflection-defined groups, we next evaluated TET2 VAF as a continuous prognostic variable. Treated as a continuous variable, Cox regression revealed that higher TET2 VAF correlated with inferior LFS (HR, 1.009; 95% CI, 1.003-1.016; P = .003) and OS (HR, 1.008; 95% CI, 1.002-1.015; P = .012) among patients with TET2-mut status. This remained significant even after adjusting for number of TET2 mutations (LFS [HR, 1.008; 95% CI, 1.002-1.015; P = .012]; OS [HR, 1.007; 95% CI, 1.001-1.014; P = .029]). A formal interaction test further confirmed no significant interaction between TET2 VAF and TET2 mutation number (supplemental Table 7).

To systematically evaluate the prognostic significance of TET2 VAF without imposing arbitrary cutoffs, we fitted RCS Cox regression models, using the empirical first quartile of TET2 VAF in patients with mutations (39%) as the reference (HR, 1.0). Hazards for both LFS and OS varied nonlinearly with VAF, rising as VAF increased toward the biallelic LOH threshold and plateauing at higher VAFs (Figure 1F). Patients with TET2-WT status had hazards well above the reference (HR, ∼1.86 [LFS] and ∼1.70 [OS]; Figure 1F dashed line). Notably, patients with TET2-mut status carrying 1 or ≥2 mutations displayed near-identical VAF-hazard curves for both LFS and OS (supplemental Figure 2B), indicating that the prognostic effect of TET2 VAF operates similarly regardless of mutation number. Together, these observations establish TET2 VAF as a robust continuous prognostic variable and motivated a dose-aware adjustment to the existing risk scores.

To address potential confounding by co-occurring mutations, we performed multivariable Cox proportional hazards regression in patients with TET2-mut status, adjusted for putative mutations including ASXL1, SRSF2, RUNX1, and age at diagnosis (supplemental Table 8). Continuous TET2 VAF (per 1% increase) remained independently associated with both inferior LFS (HR, 1.008; 95% CI, 1.002-1.014; P = .012) and OS (HR, 1.007; 95% CI, 1.001-1.014; P = .024). Among the comutations, only ASXL1 retained an independent adverse effect (LFS [HR, 1.30; 95% CI, 1.01-1.68; P = .044]; OS [HR, 1.34; 95% CI, 1.04-1.74; P = .025]); SRSF2 and RUNX1 were not significant in this multivariable model.

Development of TET2-VAF–adjusted scoring systems

Capitalizing on the nonlinear continuous VAF-survival relationship, we generated VAF-adjustment lookups (Figure 2A) translating any TET2 VAF and TET2-WT status into a per-system numeric adjustment for CPSS-Mol, GFM, and MMM (Figure 2B; supplemental Table 9). Across the 1% to 100% VAF range, adjustment values were minimal at low VAFs (<25%), became positive between ∼40% and ∼70%, and then gently declined; patients with TET2-WT status received the largest fixed adjustment (CPSS-Mol, +1.78; GFM, +2.68; MMM, +2.96) (supplemental Table 9). The TET2-VAF–adjusted scores yielded higher time-dependent AUCs and Harrell C-indices than the parental scores for both LFS and OS (Figure 2C; supplemental Table 17). All IDI estimates were highly significant (range, 0.30-0.56; all P < .001), and NRI was significant across all 3 systems (range, 0.091-0.105; P ≤ .017; supplemental Table 17).

Figure 2.

Development of the VAF-adjusted scoring system. (A) Heat maps showing 4-year OS (left) and LFS (right) rates stratified by original risk categories of CPSS-Mol, GFM, and MMM scoring systems (rows) and TET2 VAF categories (columns). The color scale ranges from green (high survival) to red (low survival), with WT patients shown separately. (B) Formula for the VAF-adjusted score. For patients with TET2-mut status, the adjustment is derived from the mean of the log HRs for OS and LFS at each integer VAF value, normalized by the respective model coefficients. For patients with TET2-WT status, a fixed adjustment is applied based on the HRs relative to the reference zone (VAF, 33%-45%). (C) Time-dependent receiver operating characteristic (ROC) AUC plots comparing the original scores (blue) vs VAF-adjusted scores (red) for OS (upper row) and LFS (lower row) across CPSS-Mol (left), GFM (middle), and MMM (right) in the DC. Mean AUC values and P values from paired comparisons are indicated. ref, reference for hazard ratio derivation.

Subsequently, LR-opt cut points were derived for each system (CPSS-Mol, 1.5176/3.4174/6.3246; GFM, 5.4826/10.3339; MMM, 1.2515/3.8602/5.0226), preserving the original number of risk tiers (Figure 3). The resulting TET2-VAF–adjusted risk groups robustly stratified survival in DC, with monotonically decreasing OS and LFS across categories (eg, CPSS-Mol low/Int-1/Int-2/high median OS, 109.8/39.6/25.8/13.5 months; all log-rank P < .001; Figure 4; supplemental Table 10). Compared with the parental classifications, the TET2-VAF–adjusted groups yielded log-rank χ² gains of +17.2 to +29.7 across the 3 systems and 2 end points (supplemental Table 17).

Figure 3.

Risk reclassification by the VAF-adjusted scoring systems in the DC. (A) CPSS-Mol; risk nomogram (left) depicting the interaction between the CPSS-Mol risk score (y-axis) and TET2 VAF (x-axis), with colors indicating the VAF-adjusted risk categories (low, Int-1, Int-2, and high). The Sankey diagram (right) illustrates patient reclassification from original to VAF-adjusted groups, with the number of patients in each group and the proportion reclassified indicated. Cut points for the VAF-adjusted score, defined using LR-opt methodology, are shown below. (B) GFM: risk nomogram showing the GFM risk score landscape and reclassification with 3 risk categories (low, Int, and high). (C) MMM: similar layout showing the MMM risk score landscape and reclassification with 4 risk categories (low, Int-1, Int-2, and high). adj, adjusted.

Figure 4.

Kaplan-Meier survival curves for the VAF-adjusted scoring systems in the DC. Kaplan-Meier curves for LFS (left) and OS (right), stratified by VAF-adjusted risk groups for CPSS-Mol (A), GFM (B), and MMM (C). med, median.

Independent validation of the prognostic significance of TET2 VAF

Validation in cohort 1 (VC1)

VC1 (n = 291) had a median age of 71 years, with a male predominance (67.4%) and a slightly higher proportion of patients with proliferative CMML (39.9%) and CMML-2 (23%) than the DC (Table 1; supplemental Figure 3). TET2 was again the most frequently mutated gene (48.5%); 141 patients carried 214 TET2 mutations. The histogram demonstrated a similarly bimodal VAF distribution (Figure 5A). There was no significant correlation between TET2 VAF and clinical variables (supplemental Table 11). Segmented regression identified inflection points at 33% and 54%, closely mirroring those in the DC (Figure 5B; supplemental Table 12). Higher VAF as a continuous covariate remained associated with inferior LFS (HR, 1.018; P = .005) and OS (HR, 1.018; P = .006), retaining the same nonlinear pattern (Figure 5C; supplemental Figure 4).

Figure 5.

Validation of the VAF-adjusted scoring systems in VC1. (A) Histogram of TET2 VAF distribution in VC1. (B) Landscape plot of individual TET2 VAFs among patients with TET2-mut status, sorted in ascending order. VAF inflection points were identified at 33% and 54%. (C) RCS Cox regression analysis of TET2 VAF for LFS (upper) and OS (lower). (D-F) Time-dependent ROC AUC plots comparing the original scores (blue) with the VAF-adjusted scores (red) for LFS (upper) and OS (lower) in CPSS-Mol (D), GFM (E), and MMM (F). (G-I) Kaplan-Meier curves for LFS (upper) and OS (lower), stratified by VAF-adjusted risk groups for CPSS-Mol (G), GFM (H), and MMM (I). adj, adjusted; ID, identity.

Applying the DC-derived per-VAF adjustment lookup, the TET2-VAF–adjusted scores outperformed the original CPSS-Mol, GFM, and MMM scores by AUC (Figure 5D-F) and C-index (supplemental Table 17). All IDIs were significant (P ≤ .005); NRIs were significant for all 3 systems (P ≤ .004). Cohort-specific LR-opt cut points (CPSS-Mol, 1.0350/2.8118/5.7672; GFM, 7.0052/9.5370; MMM, 3.0568/6.2592/6.6786) defined risk groups with strong OS and LFS stratification (Figure 5G-I; all log-rank P < .001; supplemental Table 13).

Validation in cohort 2 (VC2)

The median age of VC2 (n = 270; Table 1; supplemental Figure 5) was 74 years; proliferative (58.9%) and CMML-2 (26.3%) cases were proportionally more frequent than in DC. In total, 176 patients carried 300 TET2 mutations. No correlation between clinical variables and TET2 VAFs were observed (supplemental Table 14). Despite the difference in cell source, the bimodal distribution of TET2 VAFs (Figure 6A) and VAF inflection points (36% and 56%; Figure 6B; supplemental Table 15) were concordant with those in the DC and VC1, supporting the fidelity of TET2 VAF measurements between BM MNCs across CD14⁺ monocytes. RCS analysis demonstrated a broadly comparable nonlinear pattern of VAF-related effects on LFS and OS (Figure 6C; supplemental Figure 6). Notably, in VC2, patients with TET2-WT status showed a higher HRs for both LFS and OS (5.617 and 4.868, respectively; Figure 6C) relative to the reference VAF.

Figure 6.

Validation of the VAF-adjusted scoring systems in VC2. (A) Histogram of TET2 VAF distribution. (B) Landscape plot of individual TET2 VAFs. VAF inflection points were identified at 36% and 56%. (C) RCS Cox regression analysis of TET2 VAF for LFS (upper) and OS (lower). (D-F) Time-dependent ROC AUC plots comparing the original scores with the VAF-adjusted scores for LFS (upper) and OS (lower) in CPSS-Mol (D), GFM (E), and MMM (F). (G-I) Kaplan-Meier survival curves stratified by VAF-adjusted risk groups for CPSS-Mol (G), GFM (H), and MMM (I). adj, adjusted; ID, identity.

The DC-derived adjustment lookup similarly improved CPSS-Mol, GFM, and MMM in VC2 (Figure 6D-F; supplemental Table 17; C-index gains, +0.007 to +0.018; IDI, 0.25-0.47; all P ≤ .006). NRI did not reach significance in VC2 (P ≥ .39), potentially reflecting smaller adjusted-category transitions or the monocyte-derived VAF input. Nevertheless, group-level KM stratification remained strong (Figure 6G-I; all log-rank P < .001; supplemental Table 16).

Crosscohort summary

Across all 3 cohorts, 3 scoring systems, and 2 end points (Figure 7; supplemental Table 17), the TET2-VAF–adjusted scores consistently improved discrimination (C-index gains in 16/18 comparisons; Figure 7A), reclassification (IDI significant in 18/18; P < .01; Figure 7B; NRI significant in DC and most VC1 comparisons; Figure 7C), and group-level stratification (log-rank χ² gains, +9 to +30; Figure 7D-F). The per-VAF adjustment lookup derived in DC generalized to 2 methodologically and demographically distinct cohorts without refitting, supporting the robustness of TET2 VAF as a continuous prognostic input across CMML populations and NGS sources.

Figure 7.

Summary of prognostic performance improvement with VAF-adjusted scoring systems across all cohorts. (A) Scatterplots of C-index values for VAF-adjusted scores (y-axis) vs the original scores (x-axis) for LFS (upper) and OS (lower) across all cohorts (DC, VC1, and VC2) and scoring systems (CPSS-Mol, GFM, and MMM). Points above the diagonal indicate improved discrimination with the VAF-adjusted score. (B) Forest plot of IDI values with 95% CIs for each scoring system and cohort combination for OS and LFS. Significance levels are indicated (∗∗∗P < .001; ∗∗P < .01; ∗P < .05). (C) Forest plot of NRI values with 95% CIs. (D-F) Bar charts comparing log-rank χ² statistics between the original (blue) and VAF-adjusted (red) scores for OS and LFS in the DC (D), VC1 (E), and VC2 (F), across CPSS-Mol, GFM, and MMM. Percentage changes are indicated above the bars.

TET2 VAF and HMA response

Finally, to explore whether TET2 VAF was associated with treatment response, we examined HMA response in the DC in the subset of patients for whom uniform response data were available. Of 37 first-line HMA-treated patients, 32 were evaluable according to International Working Group (IWG) MDS/MPN consensus criteria. Overall response rates were similar between patients with TET2-mut and those with TET2-WT status (16/24 [67%] vs 5/8 [63%]; supplemental Table 18) and, among patients with TET2-mut status, did not differ between patients with monoallelic status (VAF <55%) and those with inferred biallelic (VAF ≥55%) status (overall response rate, 65% vs 71%; complete remission, 35% vs 43%). Consistent with this, TET2 VAF was near identical in responders and nonresponders (median, 49% vs 50%; P = .78). Thus, although TET2 VAF is prognostically informative, we found no evidence that it is predictive of response to HMA therapy.

Discussion

In this multi-institutional study, we demonstrated TET2 VAF as a robust continuous, nonlinear prognostic variable. Unbiased segmented regression of dominant-hit TET2 VAFs reproducibly identified 2 inflection points at 32% and 55% across the DC and both validation cohorts (VC1 and VC2), whereas RCS Cox modeling formalized the resulting nonlinear relationship between TET2 VAF and survival. Building on this, we generated TET2-VAF–adjusted CPSS-Mol, GFM, and MMM scores by applying a per-integer VAF adjustment to each patient’s baseline score. These adjusted scores improved discrimination, reclassification, and group-level stratification across all cohorts, scoring systems, and end points (OS and LFS), without the need for refitting in the validation data sets.

The favorable prognostic impact of TET2 mutations in CMML has been consistently reported,8, 9, 10, 11, 12, 13, 14 with subsequent studies refining the contribution of mutation number, truncation status, and ASXL1/TET2 genotype combinations.^{9,12,14,29,30} However, most prior CMML-focused analyses have treated TET2 status as binary or approximated mutation burden using mutation count or arbitrary VAF thresholds. To our knowledge, this study represents the first large-scale quantitative evaluation of TET2 VAF as a continuous, prognostic variable in CMML, formally modeling its nonlinear relationship with survival and translating this into a clinically applicable, per-integer VAF adjustment framework. This extends recent observations by Csizmar et al,¹⁴ who reported no benefit from linear incorporation of TET2 VAF, by demonstrating that the underlying relationship is nonlinear and, when appropriately modeled, conveys clinically meaningful and additive prognostic information.

The nonlinear VAF-hazard relationship has a plausible biological interpretation. Hazards were favorable in the heterozygous lower-VAF range (<32%), rose progressively as VAF approached the biallelic LOH threshold (32%-55%), and then plateaued at ≥55%. The plateau at ≥55%, accompanied by a marked enrichment of single-hit cases (82% vs ∼40–58% at lower VAFs; P < .0001), is consistent with copy-neutral LOH as the predominant mechanism generating very high VAFs, aligning with prior chromosomal-microarray and single nucleotide polymorphism (SNP)-array evidence.^18,21,27,28 This stepwise biology (subdominant heterozygote → dominant heterozygote → biallelic LOH), paralleling the graded loss of TET2 dioxygenase activity described in MDS,¹⁵ provides a mechanistic frame for the 3-zone hazard structure and reinforces the case for modeling TET2 VAF as a continuous variable, rather than collapsing it into a single binary status. The earlier-reported counterintuitive “favorable dosage” effect of multihit TET2⁹ may, in this view, reflect that the multihit subset is enriched for the lower-VAF (heterozygous) range rather than the high-VAF (LOH) range. Consistent with this, Yousuf et al³⁰ recently reported that exactly 2 TET2 mutations, most likely representing a stable, functionally biallelic TET2-deficient state, conferred the longest survival in CMML, whereas single mutations and ≥3 mutations were associated with less benefit, the latter potentially reflecting clonal complexity rather than further loss of TET2 function.

Our findings have direct clinical implications. TET2-mut patients with CMML had consistently more favorable expected outcomes than patients with TET2-WT status, and the TET2-VAF–adjusted CPSS-Mol, GFM, and MMM scores allow more refined risk stratification by integrating the continuous TET2 VAF signal alongside conventional clinical and molecular variables. Patients reclassified into a higher risk tier by the adjusted score may warrant closer monitoring, earlier therapeutic intervention, and timely consideration of allogeneic transplantation, whereas those moving to a lower tier may be appropriate for less intensive management. Notably, the per-integer VAF adjustment lookup derived in the DC generalized to 2 independent validation cohorts, including one based on CD14⁺ monocyte sequencing, without refitting, supporting prospective application of the framework across diverse sequencing platforms and sample sources, provided quantitative TET2 VAF is available. To facilitate clinical use, we provide the TET2-VAF–adjusted scoring system as a freely accessible web-based calculator (https://tet2-vaf-adjusted-cmml-risk.netlify.app/), allowing clinicians to compute the adjusted CPSS-Mol, GFM, or MMM score and corresponding risk category from a patient's parental score and dominant-hit TET2 VAF. In exploratory analyses, TET2 VAF was not associated with response to HMA therapy, with comparable response rates observed across TET2 mutation and VAF strata (supplemental Table 18). Interpretation of these findings is limited by the modest number of HMA-treated cases in our cohort, reflecting the relatively restricted availability of HMA for CMML in the United Kingdom. Larger studies will be required to determine whether TET2 VAF has predictive value for HMA response and could inform treatment decision-making.

Strengths of this study include its scale (n = 1206), geographic and methodological diversity (United Kingdom, France, United States, Spain, Lithuania, and Taiwan; BM MNCs and sorted CD14⁺ monocytes; multiple targeted NGS platforms), and a fully data-driven analytical approach that avoided arbitrary VAF thresholds. The consistency of segmented-regression inflection points (32%/55% in DC, 33%/54% in VC1, and 36%/56% in VC2) across these heterogeneous cohorts further supports the robustness of the proposed framework.

Several limitations should be acknowledged. First, the DC was largely derived from a tertiary referral setting, with a relatively high TET2 mutation frequency (67%) and underrepresentation of CPSS-Mol low-risk cases compared with population-based series; validation in unselected cohorts will be important. Second, although the ≥55% threshold for inferred biallelic LOH is supported by both our data and prior studies,^21,27,28 direct copy-number assessment would allow for more precise classification. Third, follow-up in VC2 was shorter than in DC and VC1, which may have reduced statistical power for reclassification analyses and contributed to the attenuated NRI observed. Finally, the study is retrospective; prospective validation and incorporation into stratified clinical trials will be required to confirm clinical utility.

In summary, we provide, to our knowledge, the first large-scale quantitative characterization of TET2 VAF as a continuous, nonlinear prognostic variable in CMML. A continuous per-integer VAF adjustment applied to established CPSS-Mol, GFM, and MMM models consistently improves prognostic performance across independent cohorts. This TET2-VAF–adjusted framework offers a practical refinement of current CMML risk stratification and provides a basis for prospective evaluation of TET2 VAF as a quantitative biomarker in clinical decision-making.

Conflict-of-interest disclosure: The authors declare no competing financial interests.