# paper.pdf ## Metadata - PDFFormatVersion=1.4 - IsLinearized=false - IsAcroFormPresent=false - IsXFAPresent=false - IsCollectionPresent=false - IsSignaturesPresent=false - Title=Dunn_et_al_Supplement_revised_240126_FINAL - Producer=macOS Version 14.6.1 (Build 23G93) Quartz PDFContext - Author=William Dunn - Creator=Word - CreationDate=D:20260223152647Z00'00' - ModDate=D:20260223152647Z00'00' ## Contents ### Page 1 1Supplementary Information12The prevalence and clinical significance of clonal monocytosis3William G. Dunn1,2,3, Michael C Sachs4, Matteo Maggi5, Muxin Gu1,2, Pedro M. Quiros6,7, Kiran Batta8,4Christen Lykkegaard Andersen9,10, Margarete A. Fabre2,3,11, Timothy Chevassut12, Irina Mohorianu1,5Daniel H. Wiseman† 7, George S. Vassiliou† 1,2,36† Co-corresponding authors78Affiliations:91. Cambridge Stem Cell Institute, University of Cambridge, Cambridge, UK.102. Department of Haematology, University of Cambridge, Cambridge, UK.113. Department of Haematology, Cambridge University Hospitals NHS Trust, Cambridge, UK.124. Section of Biostatistics, Department of Public Health, University of Copenhagen, Denmark.135. Department of Molecular Medicine, University of Pavia, Pavia, Italy.146. Department of Biochemistry and Molecular Biology, Instituto Universitario de Oncología15(IUOPA), Universidad de Oviedo, Oviedo, Spain.167. Instituto de Investigación Sanitaria del Principado de Asturias, ISPA, Oviedo, Spain.178. Epigenetics of Haematopoiesis Laboratory, Division of Cancer Sciences, University of18Manchester, Manchester, UK.199. Research Unit for General Practice, Department of Public Health, University of Copenhagen,20Denmark.2110. Department of Hematology, University Hospital Copenhagen, Rigshospitalet, Denmark.2211. Centre for Genomics Research, Discovery Sciences, Biopharmaceuticals R&D, AstraZeneca,23Cambridge, UK.2412. Department of Clinical and Experimental Medicine, Brighton and Sussex Medical School,25Brighton, UK.26 27Correspondence to:28Professor George Vassiliou gsv20@cam.ac.uk29Dr Daniel Wiseman daniel.wiseman@manchester.ac.uk30 31 ### Page 2 2Table of Contents32Supplementary Methods Page 3-633Supplementary Results Page 734Supplementary Figures Pages 8-2035Supplementary Tables Pages 21-3536References Page 363738 ### Page 3 3Supplementary Methods39United Kingdom Primary Care Cohort and the Copenhagen Primary Care Laboratory40In addition to the UKB, we examined data from a UK primary care cohort to examine trends in41monocyte indices versus age and sex. This dataset was derived from all CBC results for NHS England42primary care patients aged 1-100 years captured over a 23-day period in July 2013, from 339,322 males43and 535,716 females, as previously described1. Separately, to externally validate the performance of44our machine learning classifier to infer the presence of SRSF2 mutations amongst those with45monocytosis (see “Supervised machine learning” subsection), we examined data from 625,32846participants without missing data from the Copenhagen Primary Care Laboratory (“CopLab”). The47CopLab database and its activities are approved by the Danish Data Protection Agency through the joint48notification of The Faculty of Health and Medical Sciences at The University of Copenhagen (journal49no. 2015-57-0121). According to Danish legislation, no ethical approval or patient consent was required50because the patients were not approached at any time during the conduct of the study. Data are stored51at a server at University of Copenhagen, and information on data access from the CopLab cohort is52provided by the steering group upon request (www.publichealth.ku.dk/research/databases-for-53collaboration/coplab/).5455Identification of clonal haematopoiesis from whole exome sequencing data56Clonal haematopoiesis driver mutations were identified from UKB WES of blood DNA as described57previously2. Briefly, Mutect2 was executed in tumour-only mode targeting a panel of 38 genes (the list58of which has been published2) associated with CH. Germline variants were filtered using a panel-of-59normals. Candidate somatic variants were filtered further using FilterMutectCalls, and any variants60initially marked as “germline” were retained if they appeared at least five times in the set of candidate61somatic variants. The final set of driver mutations was curated by applying thresholds for alternate read62count (³2), presence on forward and reverse strands, minimum read depth (³7/³10 for SNVs/indels63respectively) and minor allele frequency (<0.001) in gnomAD3. Subsequent filtering of germline64variants and artefactual calls followed the criteria established by Vlasschaert et al.4 To ameliorate65previously reported issues with mapping at the U2AF1 locus5, U2AF1 mutations were identified66separately using Samtools mpileup6 to identify single nucleotide variants (SNVs) at known hotspots,67retaining variants with support from ³3 reads at a variant allele frequency (VAF) of >10%. In addition68to identifying CH driver mutations, we obtained data on mosaic chromosomal alterations (mCA) from69UKB Application 19808 Return 30947.7071Statistical modelling72Differences in proportions across monocytosis strata were evaluated using the Chi-squared test for73equality of proportions with continuity correction. Post-hoc pairwise comparisons were performed74 ### Page 4 4using the same test, with Holm correction for multiple comparisons. To model the effect of specific75driver mutations on monocyte indices with respect to age, we fitted a generalised additive model (GAM)76with age at recruitment as a smooth term and driver mutation as a categorical term.7778Amongst individuals with CMUS/CCMUS, a Fine-Gray subdistribution hazard model8 was fitted to79model covariate effects on the cumulative incidence of myeloid neoplasia in the presence of death as a80competing risk, using the finegray() function from the survival R package. Gray’s test was used to81evaluate the null hypothesis of equality of cause-specific cumulative incidence functions between82groups using the cuminc() function from the cmprsk package in R. We performed forward stepwise83regression to identify variables associated with progression from CMUS/CCMUS to MN, beginning84with a base model that included only age and age2 as covariates. Additional predictors were then added,85with Akaike Information Criteria (AIC) used to determine the predictor chosen at each step, and this86process continued iteratively until additional predictors no longer led to a decrease in AIC. The87candidate pool of predictors included demographic parameters (sex, smoking status), the88presence/absence of cytopenia(s), the presence/absence of driver mutations in 18 genes present in at89least 5 individuals with CMUS/CCMUS (ASXL1, BCORL1, CALR, CBL, DNMT3A, GNAS, GNB1,90IDH2, JAK2, KRAS, NRAS, PPM1D, RAD21, SF3B1, SMC3, SRSF2, STAG2, TP53, U2AF1), or the91presence/absence of mosaic chromosomal alterations that were present in at least 5 individuals with92CMUS/CCMUS (9p gain/CN-LOH, 14q loss/CN-LOH, 22q loss/CN-LOH). Since TET2 driver93mutations are i) highly prevalent in clonal myeloid disorders, ii) may be biallelic and iii) are associated94with copy neutral loss of heterozygosity (CN-LOH) or deletions of chromosome 4q leading to loss of95both wild type alleles, we combined TET2 and chromosome 4q events and included them as a single96categorical predictor with three levels (no TET2 or 4q loss/CN-LOH; single TET2 mutation or isolated974q loss; multiple TET2 mutations or TET2 mutation plus 4q loss/CN-LOH).9899To determine whether the presence of an absolute monocytosis was predictive of progression100independent of genomic factors, we performed Fine-Gray regression using the final set of predictors101derived from forward stepwise regression, in addition to the monocyte count as a binary variable102(<1/³1). This “alternative” model was then compared with the “base” model using a likelihood ratio103test. In addition to performing stepwise regression on the full cohort of CMUS/CCMUS cases, we also104repeated this process in males only, to assess predictors that are restricted to males (presence/absence105of ZRSR2 mutations and mosaic loss of the Y chromosome).106107To maximise interpretability, associations between CMUS/CCMUS and cardiovascular/renal disease108were examined by performing event history regression using a pseudo-observations framework, using109the cumincglm() function from the eventglm package in R9. This approach facilitates calculation of110estimands for the change in the ten-year cumulative incidence of cardiovascular/renal disease associated111 ### Page 5 5with each predictor, improving comprehensibility. However, as AIC is not defined for such models we112did not use this for performing stepwise regression as described for factors associated with progression113to MN. Associations between CMUS/CCMUS and microalbuminuria were examined using logistic114regression. For all regression analyses modelling associations between CMUS/CCMUS and115cardiovascular/renal disease, the following were included as covariates: age, sex, smoking status,116systolic blood pressure and glycated haemoglobin (HbA1c). Blood lipids (triglycerides, LDL117cholesterol, HDL cholesterol) were also included in analyses of association with cardiovascular disease.118119ICD-10 codes used to define myeloid neoplasms, cardiovascular and renal diseases for the time-to-120event analyses are listed in Supplementary Table 1. All statistical analyses were performed in R v4.3.1.121122Supervised machine learning123UKB participants were labelled as having either SRSF2-CMUS (cases) or non-clonal monocytosis124(controls); individuals with CMUS bearing other drivers were not used for training. As individuals with125monocytosis were much more likely to be controls than cases, random down-sampling was performed126to achieve a 1:1 ratio of cases:controls in the input data, to enhance model training and convergence;127this down-sampling process was repeated ten times iteratively. Each of the ten down-sampled datasets128were subsequently partitioned into an 80:20 train:test ratio. All models were built using ten repeats of129ten-fold cross-validation setups; a grid-search approach was used to tune hyperparameters, with the130number of trees in each Random Forest model held constant at n=1000. To avoid introducing bias from131the down-sampling step, a modified cross-validation approach was adopted, iteratively training and132evaluating each Random Forest model ten times, each time using a different random down-sample of133the majority (control, i.e. non-clonal monocytosis) class, thereby quantifying the robustness and134stability of each model to variation in the subset of control samples or train/test partition.135136We next selected the model that most closely approximated the median performance from ten repeats137of model building for further evaluation. Performance was assessed in the unseen test set, to which we138added unseen controls (non-clonal monocytosis, CMUS bearing non-SRSF2 driver mutations) such that139the prevalence of SRSF2-CMUS matched that of the whole UKB monocytosis cohort. We subsequently140measured performance using metrics that are independent (area under the receiver operating141characteristic curve, AUC) and dependent (positive and negative predictive value, PPV and NPV142respectively) on the prevalence of SRSF2-CMUS. The trade-off between sensitivity and positive143predictive value was used to ascertain the optimal threshold (cut-off) used to predict the presence of144SRSF2 mutations. To enable interpretability, variable importance was computed from the mean145decrease in node impurity from splitting on each feature (measured by Gini index), averaged across all146trees and across each of the ten repeats of model-building, using the importance() function from the147randomForest package in R.148 ### Page 6 6149To assess performance on an external cohort, we applied the classifier to CBC data for all participants150in the CopLab primary care dataset who had monocytosis on their first blood sample. As genetic data151were unavailable in this primary care cohort, we used incident MN as a surrogate marker of the presence152of SRSF2, since these mutations are associated with a high-risk of progression from CMUS to MN. The153median follow up for CopLab participants with monocytosis was 6.4 years.154155Receiver operating characteristic curve-based methods to derive monocyte thresholds156To explore alternative methods to derive new sex-specific monocyte thresholds than our method of157equalizing the proportions between sexes, we used receiver operating characteristic (ROC) curves to158determine the optimal threshold for monocyte count and percentage that best discriminated those with159CH who would later develop MN. Optimal sex-specific values were defined as those values that160maximized Youden’s J statistic, corresponding to the point on the ROC curve that is farthest from the161line of no discrimination.162163 ### Page 7 7Supplementary Results164ROC-based methods to derive sex-specific monocyte thresholds165In addition to our method of deriving sex-specific monocyte thresholds that equalize the proportion of166males/females defined as having monocytosis, we explored an alternative method using ROC curves to167determine the optimal monocyte thresholds that discriminate males/females that later develop MN (see168Supplementary Methods). Using this approach, we derived alternative sex-specific monocyte thresholds169(³0.8 x 109/L and ³11.4% of white blood cells in males, ³0.6 x 109/L and ³9.6% of white blood cells170in females). Notably, neither monocyte count nor percentage were strongly discriminative of incident171MN in isolation (Supplementary Figure 15). We next compared these ROC analysis-derived thresholds172with our previously proposed revised sex-specific thresholds. The ROC curve-derived thresholds173expectedly reduced the number of UKB participants who were classed as NCM, CMUS-REL and174CCMUS (Supplementary Table 15).175176Despite reducing the number of cases classed as CMUS-REL and enriching for cases that were closer177to CMUS-ABS in terms of their indices, we noted that the rate of progression for CMUS-REL was178similar using our previously proposed, more inclusive thresholds (Supplementary Figure 16). CCMUS179cases defined using the more stringent ROC curve-based thresholds showed a higher rate of progression,180but of note a substantial proportion of those reclassified as not CCMUS based on ROC-based thresholds181progressed to MN (11/69, 15.9%). By substantially reducing the number of CCMUS cases, these182stringent thresholds downgrade these demonstrably high risk cases to CCUS without monocytosis,183despite the rate of progression being almost three-fold higher than our previous observations for CCUS184without monocytosis (Figure 2A). Moreover, these ROC curve-based thresholds did not enrich for MN185progression amongst CMUS-REL cases, despite blurring the distinction between CMUS-REL and186CMUS-ABS in males. Consequently, we instead adopted our simple sex-specific thresholds derived187from equalizing proportions between the sexes, as described in the main text.188 189190 ### Page 8 8Supplementary Figures191Supplementary Figure 1192193194195Supplementary Figure 1 shows the relationship between age at sampling (years) and monocyte196count x 109/L amongst UK Biobank participants with clonal haematopoiesis, stratified by driver197mutation. Those with SRSF2 mutations have significantly higher monocyte counts with respect to age198(p < 2 x 10-16).1992000.4 0.5 0.6 0.7 0.8 40 50 60 70 Age (years) Monocyte count x 109/L Driver ASXL1 DNMT3A non-R882 DNMT3A R882 PPM1D SF3B1 SRSF2 TET2 TP53 ### Page 9 9Supplementary Figure 2201202CH without monocytosis (n=19,109)203204205206Supplementary Figure 2 shows a waterfall plot of driver mutations amongst those with clonal207haematopoiesis without monocytosis. Arrows highlight non-R882 DNMT3A mutations (blue), TET2208mutations (red) and SF3B1/SRSF2 splicing factor mutations (green). Here, non-R882 DNMT3A209mutations are the most prevalent driver.210211 ### Page 10 10Supplementary Figure 3212213CMUS-REL (n=1305)214215216217Supplementary Figure 3 shows a waterfall plot of driver mutations amongst those with CMUS-REL.218Arrows highlight non-R882 DNMT3A mutations (blue), TET2 mutations (red) and SF3B1/SRSF2219splicing factor mutations (green). Non-R882 DNMT3A mutations remain the most prevalent driver220within this group.221222 ### Page 11 11Supplementary Figure 4223224CMUS-ABS (n=196)225226227228Supplementary Figure 4 shows a waterfall plot of driver mutations amongst those with CMUS-ABS.229Arrows highlight non-R882 DNMT3A mutations (blue), TET2 mutations (red) and SF3B1/SRSF2230splicing factor mutations (green). In contrast to Supplementary Figures 2 and 3, TET2 mutations are231now the most prevalent driver mutations, and the proportion of individuals with SRSF2 mutations is232enriched.233234235 ### Page 12 12Supplementary Figure 5236237CCMUS (n=245)238239240241Supplementary Figure 5 shows a waterfall plot of driver mutations amongst those with CCMUS.242Arrows highlight non-R882 DNMT3A mutations (red), TET2 mutations (blue) and SF3B1/SRSF2243splicing factor mutations (green). Compared with Supplementary Figures 2 to 4, TET2 and SRSF2244driver mutations are enriched, and a proportion of individuals (n=13) have TET2/SRSF2 co-mutated.245246 ### Page 13 13Supplementary Figure 6247248249250251Supplementary Figure 6 shows the cumulative incidence of myeloid neoplasia for CCUS without252monocytosis versus CCMUS-REL and CCMUS-ABS.2532540.0 0.1 0.2 0.3 0.4 0 5 10 Years since CBC measurement Cumulative incidence CCUS−without−monocytosis CCMUS−REL MN (current monocyte thresholds) 0.0 0.1 0.2 0.3 0.4 0 Cumulative incidence MN ( 0 5 10 Years since CBC measurement CCUS−without−monocytosis CCMUS−REL CCMUS−ABS N (current monocyte thresholds) 0.0 0.1 0.2 0.3 0.4 0 5 10 Years since CBC measurement Cumulative incidence MN (sex-specific monocyte thresho ### Page 14 14Supplementary Figure 7255256257258Supplementary Figure 7 shows the platelet monocyte ratio amongst CMUS/CCMUS cases stratified259by driver gene, versus NCM. A reduced platelet:monocyte ratio is notable amongst SRSF2-CMUS,260but not CMUS bearing other driver mutations.2612620 500 1000ASXL1 CALR DNMT3A otherDNMT3A R882JAK2 PPM1D SF3B1 SRSF2 TET2 TP53Non-clonal monocytosis Platelet−monocyte ratio DriverASXL1 CALR DNMT3A non R882DNMT3A R882JAK2 PPM1D SF3B1 SRSF2 TET2 TP53Non-clonal monocytosis Platelet−monocyte ratio versus driver gene amongst CMUS/CCMUS and NCM cases ### Page 15 15Supplementary Figure 8263264265Supplementary Figure 8 Performance of MoSAIC amongst individuals with monocytosis266(n=32,261) in the Copenhagen Primary Care Laboratory (CopLab). Panels A-C shows the267cumulative incidence of myeloid neoplasia (left) and death without myeloid neoplasia (right) in those268predicted positive by the MoSAIC classifier (Panel A) versus using simple monocyte count cut-offs269of >1 x 109/L and >1.5 1 x 109/L for Panels B and C respectively.270 ### Page 16 16Supplementary Figure 9271272273274Supplementary Figure 9: Cumulative incidence of AML using MoSAIC and different monocyte275thresholds. Comparison of the performance of MoSAIC and different monocyte cut-offs for276predicting subsequent development of AML (left hand panels), all other forms of myeloid neoplasia277(MN, middle panels) and death without myeloid neoplasia (right hand panels).278279 ### Page 17 17Supplementary Figure 10280281282Supplementary Figure 10 shows the proportions of CH cases that meet criteria for CMUS and283CCMUS, stratified by driver gene, and compared with the prevalence of monocytosis amongst those284without CH. For most common driver genes, with large clone size, there is either an appreciable285increase in the proportion of CMUS/CCMUS cases or an increase in the proportion with the higher286risk entities CMUS-ABS and CCMUS, though this is not evident for CH bearing DNMT3A mutations.287288VAF <10% VAF ≥10%ASXL1 CALR DNMT3A non-R882DNMT3A R882JAK2 PPM1D SF3B1 SRSF2 TET2 TP53No CHMonocyte subtypeNo monocytosis NCM CMUS−REL CMUS−ABS CCMUSPrevalence of monocytosis versus driver and clone size ### Page 18 18Supplementary Figure 11289290291Supplementary Figure 11 summarizes the diagnostic transitions between CMUS/CCMUS entities292based on the current monocyte thresholds and driver mutation inclusion criteria, versus our proposed293criteria (excluding isolated DNMT3A mutations and introducing sex-specific monocyte thresholds).294295CCMUS CMUS−ABS CMUS−REL NCM (orDNMT3A with monocytosis) CCMUS CMUS−ABS CMUS−REL No monocytosis 0 500 1000 1500 Current CMUS Classification Proposed Revised CMUS Classification Number of Cases Sex Female Male CMUS diagnostic transitions ### Page 19 19 296 297 298299Supplementary Figure 12 shows ROC curves for monocyte count and threshold to determine the300optimal threshold distinguishing cases who later develop MN.3010.0 0.2 0.4 0.6 0.8 1.0 0.0 0.4 0.8ROC curve with annotated monocyte percentage thresholds (Males)1 − Specificity Sensitivity 4.67 5.23 5.63 5.97 6.31 6.62 6.957.277.58 7.888.23 8.6 8.99 9.46 10.03 10.84 12.35 Best = 11.360.0 0.2 0.4 0.6 0.8 1.0 0.0 0.4 0.8ROC curve with annotated monocyte thresholds (Males)1 − Specificity Sensitivity 0.32 0.42 0.5 0.6 Best = 0.8 0.70.0 0.2 0.4 0.6 0.8 1.0 0.0 0.4 0.8ROC curve with annotated monocyte percentage thresholds (Females)1 − Specificity Sensitivity 4.034.48 4.85 5.16 5.445.74 5.996.23 6.5 6.79 7.08 7.39 7.778.19 8.69 9.37 10.66 Best = 9.610.0 0.2 0.4 0.6 0.8 1.0 0.0 0.4 0.8ROC curve with annotated monocyte thresholds (Females)1 − Specificity Sensitivity 0.26 0.34 0.42 0.5 0.66 0.74 Best = 0.58 ### Page 20 20Supplementary Figure 13302303304Supplementary Figure 13 shows the cumulative incidence of MN across monocytosis subtypes305using our sex-specific monocyte thresholds derived by equalizing the proportion of males/females306with monocytosis (left), versus those derived using the ROC-based method described in307Supplementary Results (right).308 3090.0 0.1 0.2 0.3 0 5 10 Years since CBC measurement Cumulative incidence CMUS−REL CMUS−ABS CCMUS Sex-specific monocyte thresholds 0.0 0.1 0.2 0.3 0 5 10 Years since CBC measurement Cumulative incidence No monocytosis NCM CMUS−REL CMUS−ABS CCMUS ROC-based sex-specific monocyte thresholds ### Page 21 21Supplementary Tables310311312Supplementary Table 1: ICD-10 codes used to define myeloid neoplasms, atherosclerotic313cardiovascular disease, and renal diseases. Note that chronic kidney disease has been defined using314the disease codes pertaining to chronic kidney disease stage 3-5, or not otherwise specified.315316Description ICD-10 codesAcute myeloid leukaemia C92.0, C92.3, C92.4, C92.5, C92.6, C92.8, C93.0, C94.0, C94.2, C94.3, C94.4 Myelodysplastic syndrome D46.0, C93.2, C93.9 Chronic myelomonocytic leukaemia C93.1 Myeloproliferative neoplasms C92.2, D45, D47.0, D47.1, D47.3, D47.4, D47.5 Atherosclerotic heart disease I25.0,I25.1,I25.8,I25.9,I21.0,I21.1,I21.2,I21.3,I21.4,I21.9,I20.0,I20.8,I20.9,I24.0,I24.8,I24.9 Chronic kidney disease N18.3, N18.4, N18.5, N18.9 Acute kidney injury N17.0, N17.1, N17.2, N17.8, N17.9 ### Page 22 22 317Supplementary Table 2: Incident myeloid neoplasia versus monocytosis strata. Individuals who318developed an incident diagnosis of a myeloid neoplasm at any point during follow up have been319included in calculations. Note that where an individual had ³2 different diagnoses, they were included320in the numbers/proportions for all diagnoses.321322Monocytosis category n AML Percent AML CMML Percent CMML MDS Percent MDS MPN Percent MPN Any MN Percent MNNo monocytosis 398933 450 0.1 51 0.01 453 0.1 828 0.2 1637 0.4 NCM 30852 26 0.1 11 0.04 51 0.2 56 0.2 136 0.4 CMUS-REL 1305 23 1.8 7 0.5 9 0.7 20 1.5 42 3.2 CMUS-ABS 196 6 3.1 6 3.1 3 1.5 9 4.6 20 10.2 CCMUS 245 18 7.3 14 5.7 25 10.2 12 4.9 50 20.4 ### Page 23 23 323Supplementary Table 3: Estimates, confidence intervals and p values for all covariates in the324Fine-Gray model of cumulative incidence of myeloid neoplasia for all UK Biobank participants325with CMUS/CCMUS. Estimates, upper and lower bounds represent the subdistribution hazard ratio.326The model was developed using forward stepwise regression to iteratively add covariates from a pool327of candidate variables.328329Term Estimate Lower Upper p valueAge 0.97 0.54 1.74 0.92 Age² 1.00 1.00 1.01 0.83 ¹SRSF2 6.70 4.08 10.99 5.63 x 10⁻ ⁴DNMT3A 0.25 0.12 0.53 2.52 x 10⁻⁴ Cytopenia 2.70 1.76 4.12 4.49 x 10⁻⁶ 9p LOH/gain 26.00 8.66 78.11 6.39 x 10⁻⁹CALR 12.48 5.16 30.19 2.11 x 10⁻⁸JAK2 5.54 2.04 15.07 7.97 x 10⁻⁴SF3B1 3.19 1.32 7.70 0.01KRAS 4.68 1.41 15.56 0.01IDH2 2.12 0.99 4.54 0.05U2AF1 7.41 1.00 54.82 0.05 Single TET2 or 4q loss 1.15 0.71 1.86 0.56 Multiple TET2 or 4q loss/LOH 2.00 1.05 3.81 0.03 ### Page 24 24 330Supplementary Table 4: Estimates, confidence intervals and p values for all covariates in the331Fine-Gray model of cumulative incidence of myeloid neoplasia for male UK Biobank participants332with CMUS/CCMUS. Estimates, upper and lower bounds represent the subdistribution hazard ratio.333The model was developed using forward stepwise regression to iteratively add covariates from a pool334of candidate variables, including those collinear with male sex (ZRSR2 mutations and mosaic loss of335the Y chromosome, mLOY).336337Term Estimate Lower Upper p valueAge 0.70 0.37 1.32 0.27 Age² 1.00 1.00 1.01 0.24SRSF2 7.58 4.19 13.69 2.04 x 10⁻¹¹DNMT3A 0.23 0.10 0.55 9.00 x 10⁻⁴ Cytopenia 2.43 1.42 4.14 1.16 x 10⁻³JAK2 12.42 4.36 35.38 2.37 x 10⁻⁶ZRSR2 12.66 4.40 36.46 2.53 x 10⁻⁶ mLOY 0.25 0.08 0.79 0.02CBL 6.93 1.65 29.11 8.21 x 10⁻³CALR 6.79 1.58 29.19 0.01SF3B1 2.91 0.98 8.59 0.05 Single TET2 or 4q loss 1.09 0.62 1.91 0.76 Multiple TET2 or 4q loss/LOH 2.60 1.20 5.67 0.02IDH2 2.17 0.90 5.22 0.08 ### Page 25 25 338Supplementary Table 5: Estimates, confidence intervals and p values for all covariates in the339event history regression model of the cumulative incidence of atherosclerotic heart disease. Note340that an age:cholesterol interaction term has been included to account for the non-linear correlation341between age and LDL cholesterol observed amongst UK Biobank participants, attributable to statin use.342Estimates can be interpreted as the additive change in the 10-year cumulative incidence of ASCVD.343344Term Estimate Lower Upper p valueAge at recruitment -0.76 -0.90 -0.63 <2 x 10-¹⁶ Elevated cholesterol 8.95 7.69 10.21 <2 x 10-¹⁶ Age² 0.01 0.01 0.01 <2 x 10-¹⁶ Male sex 5.12 4.96 5.29 <2 x 10-¹⁶ Previous smoker 1.42 1.24 1.59 <2 x 10-¹⁶ Current smoker 3.30 3.01 3.59 <2 x 10-¹⁶ NCM-REL 0.31 -0.05 0.67 0.09 NCM-ABS 1.89 0.73 3.05 1.44 x10⁻³ CMUS-REL 1.10 -0.69 2.90 0.23 CMUS-ABS 8.21 2.49 13.94 3.63 x 10⁻³ CCMUS 1.74 -2.47 5.94 0.42 Hypertension (Stage 1+) 0.37 0.20 0.54 1.11 x 10⁻⁵ Elevated HbA1c 8.50 7.88 9.12 <2 x 10-¹⁶ Age/cholesterol interaction term -0.16 -0.19 -0.14 <2 x 10-¹⁶ ### Page 26 26 345Supplementary Table 6: Estimates, confidence intervals and p values for all covariates in the346event history regression model of the cumulative incidence of acute kidney injury (AKI). Estimates347can be interpreted as the additive change in the 10-year cumulative incidence of AKI.348349Term Estimate Lower Upper p valueAge at recruitment -0.88 -0.98 -0.79 <2 x 10-¹⁶ Age² 0.01 0.01 0.01 <2 x 10-¹⁶ Male sex 1.46 1.34 1.57 <2 x 10-¹⁶ Previous smoker 0.55 0.44 0.67 <2 x 10-¹⁶ Current smoker 2.39 2.18 2.60 2.74 x 10⁻⁶ NCM-REL 0.02 -0.21 0.26 0.85 NCM-ABS 2.17 1.27 3.06 2.10 x 10⁻⁶ CMUS-REL 0.13 -1.06 1.32 0.83 CMUS-ABS 6.08 1.53 10.62 8.74 x 10⁻³ CCMUS 10.44 5.83 15.04 8.85 x 10⁻⁶ Hypertension (Stage 1+) 0.20 0.09 0.32 3.37 x 10⁻⁴ Elevated HbA1c 7.68 7.17 8.20 <2 x 10-¹⁶ ### Page 27 27 350Supplementary Table 7: Estimates, confidence intervals and p values for all covariates in the351event history regression model of the cumulative incidence of chronic kidney disease (CKD). Note352CKD has been defined as that coded as stage 3, 4 or 5 CKD or not otherwise specified. Estimates can353be interpreted as the additive change in the 10-year cumulative incidence of CKD.354355Term Estimate Lower Upper p valueAge at recruitment -1.27 -1.36 -1.18 <2 x 10-¹⁶ Age² 0.01 0.01 0.01 <2 x 10-¹⁶ Male sex 0.31 0.21 0.42 9.22 x 10⁻⁹ Previous smoker 0.50 0.39 0.62 <2 x 10-¹⁶ Current smoker 0.74 0.57 0.92 <2 x 10-¹⁶ NCM-REL 0.06 -0.16 0.29 0.58 NCM-ABS 1.67 0.85 2.49 6.56 x 10⁻⁵ CMUS-REL -0.28 -1.36 0.81 0.62 CMUS-ABS 4.00 0.02 7.97 0.05 CCMUS 5.64 1.84 9.44 3.60 x 10⁻³ Hypertension (Stage 1+) 0.27 0.16 0.37 7.75 x 10⁻⁷ Elevated HbA1c 7.90 7.39 8.41 <2 x 10-¹⁶ ### Page 28 28 356Supplementary Table 8: Estimates, confidence intervals and p values for all covariates in the357logistic regression model of the presence/absence of microalbuminuria at recruitment to the UK358Biobank. Microalbuminuria was defined as a urinary albumin:creatinine ratio (ACR) >30mg/g.359360Term Estimate Lower Upper p valueAge at recruitment 0.88 0.85 0.90 <2 x 10-¹⁶ Age² 1.00 1.00 1.00 <2 x 10-¹⁶ Male sex 1.05 1.02 1.08 1.96 x 10⁻³ Previous smoker 1.09 1.06 1.13 1.15 x 10⁻⁷ Current smoker 1.51 1.44 1.58 <2 x 10-¹⁶ NCM-REL 0.93 0.87 0.99 0.02 NCM-ABS 1.69 1.48 1.93 5.63 x 10⁻¹⁵ CMUS-REL 1.28 1.02 1.60 0.03 CMUS-ABS 2.51 1.65 3.83 1.96 x 10⁻⁵ CCMUS 1.33 0.81 2.19 0.26 Hypertension 1.67 1.61 1.73 <2 x 10-¹⁶ Elevated HbA1c 4.46 4.26 4.67 <2 x 10-¹⁶ ### Page 29 29 361Supplementary Table 9: Confusion matrix of MoSAIC classifier performance. Columns are362ground truth labels and rows are predicted labels using MoSAIC with a probability threshold of 0.82363for predicting the positive class (SRSF2-CMUS). Performance is in the unseen test set using the median364model.365366No SRSF2 SRSF2Predicted no SRSF2 6103 8 Predicted SRSF2 50 6 ### Page 30 30 367Supplementary Table 10: Complete blood count indices versus MoSAIC classification amongst368CopLab participants with monocytosis. MoSAIC positive individuals are those predicted to have a369SRSF2 mutation by the classifier, using the previously determined optimal prediction threshold.370371CBC parameter MoSAIC negative (Median counts) MoSAIC positive (Median counts)Basophils 0.03 0.02 Eosinophils 0.16 0.02 Haemoglobin 13.90 12.40 Lymphocytes 1.70 1.70 Monocytes 0.70 1.30 Neutrophils 3.40 2.90 Platelets 259.00 160.50 White blood cells 6.30 7.20 ### Page 31 31 372Supplementary Table 11: Performance of the MoSAIC classifier versus simple monocyte373thresholds in the CopLab cohort. Specificity, sensitivity, positive predictive value (PPV) and negative374predictive value (NPV) for identifying individuals who developed myeloid neoplasia (MN) within ten375years, based on prediction by MoSAIC based on complete blood count indices, versus using simple376monocyte cut-offs of >1 or >1.5 x 109/L.377378Developed MN in 10y? MoSAIC predicted Monocytes >1 Monocytes >1.5Sensitivity 38.00% 45.80% 27.20% Specificity 98.80% 98.90% 98.50% PPV 14.2% (141/993) 4.9% (170/3440) 11.1% (101/914) NPV 98.70% 98.70% 98.50% ### Page 32 32Category Persistent Non-persistent Total Any MN 170 89 259No MN 6935 12367 19302Total 7105 12456 19561379Supplementary Table 12: Persistence of monocytosis and MN risk in the CopLab dataset380The risk of developing MN (i.e. AML, CMML, MDS or MPN) amongst 19,561 CopLab participants381(8817 women and 10,753 men) with at least two CBC measurement was ~3.5x higher amongst those382with persistent vs non-persistent monocytosis (2.4% vs 0.7%; p<0.00001, Chi Square). Persistent383monocytosis was defined as monocytosis identified on at least two independent CBC measurements.384Non-persistent monocytosis was identified on a single CBC, but not on any subsequent CBC.385 ### Page 33 33 386 Supplementary Table 13: Current CMML, CMUS and CCMUS definitions. Only the ICC has defined CMUS and CCMUS to date. Oligomonocytic387CMML requires additional supporting criteria in both the WHO and ICC definitions.388 389Diagnostic entity CMML CMUS CCMUS Classification board ICC WHO 5th Edition ICC ICC Definition of mono- cytosisPeripheral blood monocytes 0.5 ⇥ 109/L and 10% of white cellsBlast threshold Blasts < 20% of cells in the peripheral blood and bone marrow Not stated Not statedCytopenia(s) re- quiredYes No No YesClonality definition Cytogenetic abnormality and/or presence of at least one myeloid neoplasm–associated mutation at VAF 10% Acquired clonal cytogenetic or molecular abnormality At least one myeloid neoplasm–associated mutation at VAF 2%Bone marrow find- ingsHypercellularity due to a myeloid prolifer- ation, often with increased monocytes Dysplasia involving 1 myeloid lineage(s) (one of the supporting criteria) No significant dysplasia, increased blasts, or morphological findings of CMMLOligomonocytic ad- justmentsYes, oligomonocytic cases must have evi- dence of clonality Yes, oligomonocytic cases must meet more supporting criteria No NoExclusions Diagnostic features consistent with another neoplasm (e.g. AML, CML, MPN), other conditions associated with monocytosis, or myeloid/lymphoid neoplasms with tyrosine kinase fusions Criteria for another myeloid or haematopoietic neoplasm are fulfilledDiagnostic require- ments Cases with evidence of clonality:Presence of monocytosis, cytopenia, blast threshold, and bone marrow findings; ex- clusion criteria must be absent.Cases without evidence of clonality:Monocytosis threshold raised to 1.0 ⇥109/L and presence of increased blasts ormorphological dysplasia or abnormal im- munophenotype required. Persistent monocytosis and blast count cri- teria with absence of alternative neoplasm. Three supporting criteria are specified: 1) dysplasia involving 1 myeloid lineage(s); 2) evidence of clonality; 3) abnormal parti- tioning of peripheral blood monocyte sub- sets. Oligomonocytic cases require criteria 1 & 2; cases with monocytes 1.0⇥109/L require only one supporting criterion. Presence of all criteria: persistent mono- cytosis, absence of cytopenia, presence of driver mutation, no significant dysplasia on bone marrow examination, no evidence of a haematopoietic neoplasm, and absence of exclusion criteria. Presence of all criteria: persistent mono- cytosis, presence of cytopenia, presence of driver mutation, no significant dysplasia on bone marrow examination, no evidence of a haematopoietic neoplasm, and absence of exclusion criteria. ### Page 34 34390Supplementary Table 14: Diagnostic entities according to our proposed revised classifications. “CHIP” cases are those not meeting criteria to be called391CCUS without monocytosis, CMUS-REL, CMUS-ABS and CCMUS.392393CHIP CCUS without mono- cytosis CMUS-REL CMUS-ABS CCMUS Definition Somatic driver mutation at VAF 2%, in the ab- sence of evidence of a haematological neoplasm As per CHIP, plus the presence of one or more cytopenia(s) Non-DNMT3A driver mutation at VAF 2%, plus relative monocytosis< 1 ⇥109 /L (0.6 and11.5% of white blood cells in males, or 0.5 and 10% in females) Non-DNMT3A driver mutation at VAF 2%, plus absolute monocy- tosis 1 ⇥109 /L (and11.5% of white blood cells in males or 10% in females) Non-DNMT3A driver mutation at VAF 2%, plus relative or absolute monocytosis (0.6 and11.5% of white blood cells in males, or 0.5 and 10% in females)Prevalence (UK Biobank) 4.4%a 0.3% 0.09% 0.02% 0.03%Most common mutations DNMT3A (53.8%), TET2(22.2%)DNMT3A (47.7%), TET2(26.9%)TET2 (50.3%), ASXL1(18.9%)TET2 (49.5%), ASXL1(22.2%), SRSF2 (13.1%)TET2 (49.3%), SRSF2(26.4%), ASXL1 (16.4%)10-year cumulative incidence of MN 1.8% 5.4% 6.0% 14.7% 25.8%10-year cumulative incidence of death 7.0% 13.3% 8.4% 21.1% 14.7%a CHIP prevalence excludes individuals meeting criteria for CCUS without monocytosis, CMUS, or CCMUS. ### Page 35 35No monocytosis NCM CMUS-REL CMUS-ABS CCMUS Originally proposed sex- specific monocyte thresholds414,134 16,404 667 146 180ROC-based sex-specific monocyte thresholds418,374 12,394 492 152 119394Supplementary Table 15: Numbers of each diagnostic entity using different sex-specific395monocyte thresholds.396397 ### Page 36 36References3983991. Phillips, R., Wood, H., Weaving, G. & Chevassut, T. Changes in full blood count parameters with400age and sex: results of a survey of almost 900 000 patient samples from primary care. British401Journal of Haematology 192, e102–e105 (2021).4022. Gu, M. et al. Multiparameter prediction of myeloid neoplasia risk. Nat Genet 55, 1523–1530403(2023).4043. Karczewski, K. J. et al. The mutational constraint spectrum quantified from variation in 141,456405humans. Nature 581, 434–443 (2020).4064. Vlasschaert, C. et al. A practical approach to curate clonal hematopoiesis of indeterminate407potential in human genetic data sets. Blood 141, 2214–2223 (2023).4085. Miller, C. A. et al. Failure to Detect Mutations in U2AF1 due to Changes in the GRCh38409Reference Sequence. The Journal of Molecular Diagnostics 24, 219–223 (2022).4106. Danecek, P. et al. Twelve years of SAMtools and BCFtools. Gigascience 10, giab008 (2021).4117. Loh, P.-R., Genovese, G. & McCarroll, S. A. Monogenic and polygenic inheritance become412instruments for clonal selection. Nature 584, 136–141 (2020).4138. Fine, J. P. & Gray, R. J. A Proportional Hazards Model for the Subdistribution of a Competing414Risk. Journal of the American Statistical Association 94, 496–509 (1999).4159. Sachs, M. C. & Gabriel, E. E. Event History Regression with Pseudo-Observations:416Computational Approaches and an Implementation in R. Journal of Statistical Software 102, 1–41734 (2022).418419