Single cell analysis reveals molecular traits of pediatric lymphoma resistant subclones

148 words 22 Full text: 11291 words (including main text, methods, and figure legends) 23 Figures: 7 figures and 10 supplementary figures, 2 tables 24

count: 108 25

pediatric lymphoma, reactive lymph nodes, single cell sequencing, VDJ clonal 26 analysis, spatial transcriptomics, resistant subclone, chemoresistance 27 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 2

28 The cause of refractory/relapse (R/R) in pediatric lymphoma is unclear. We hypothesized that 29 a stem-like, chemoresistance lymphoma subclone may contribute to R/R. Combining single cell 30 RNA sequencing (scRNA -seq) and immune receptor sequencing ( scVDJ-seq) on freshly 31 acquired pediatric non-Hodgkin lymphoma (pNHL n=10), Hodgkin lymphoma (pHL n=5), and 32 10 reactive lymph nodes from adults or children (a/pReLy), we discovered pediatric-specific 33 progenitor-like lymphocytes, whose cellular program was enriched in pNHL subclones 34 emerging late during clonal evolution, accompanied by loss of immune receptor expression. R-35 CHOP target gene expression indicated that these subclones may escape first-line treatment, 36 and suggested MSI2, an RNA binding protein, as a potential target. In pHL, the progenitor-like 37 program was found in the tumor microenvironment (TME) but not Hodgkin cells which, during 38 relapse, were myeloid-like and accompanied by CD74highCCL5+ CD8 T cells. In summary, we 39 discovered R/R associated lymphoma subclones in pediatric lymphoma and potential ways to 40 eradicate them. 41 42 Key messages 43 • Progenitor-like lymphocytes are uniquely found in pediatric reactive lymph nodes, but 44 not in adults. 45 • Resistant sub clones from pNHL acquire progenitor -like program and share MSI2 46 expression. 47 • Relapsed Hodgkin cells are monocyte-like and recruit CD74highCCL5+ CD8 T cells. 48 49 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 3

50 51 Pediatric l ymphoma accounts for 10-15% of all childhood cancer with ~60% non -Hodgkin 52 lymphoma (pNHL, B cell ~40%, T cell ~20%), and 40% Hodgkin lymphoma (pHL) (1-3). 53 Though the difference between pediatric and adult lymphoma is not fully underst ood, 54 accumulating evidence underscores pediatric lymphoma being a unique entity from its adult 55 counterpart in prevalence, presentation, molecular traits, and treatment response (4-6). While 56 treatment strategies differ across lymphoma subtypes, first -line therapy for aggressive pNHL 57 includes agents overlapping with the CHOP regimen (cyclophosphamide, doxorubicin, 58 vincristine, and prednisone) widely used in adults, with the addition of the anti-CD20 antibody 59 rituximab (R) in mature B -cell disease. In Europe, first -line therapy of classical pHL is also 60 CHOP based (7), while other regimens are considered outside Europe 61 (https://classic.clinicaltrials.gov/ct2/show/NCT02684708 ). Although resulting in a cure-rate 62 up to 70 -95%, long term adverse effects exist such as infertility, cardiovascular disease, and 63 endocrine conditions (8). Moreover, the events of relapses and drug resistance (R/R) are often 64 highly aggressive and associated with poor outcome (9, 10). Novel insight into how R/R occurs 65 may benefit future treatment. 66 67 Acquisition of stemness is key to cancer progression events such as epithelial mesenchymal 68 transition, metastasis and drug resistance (11, 12), and often as a part of cancer evolution (13). 69 Although the existence of cancer stem cell is widely acknowledged (14, 15), lymphoma stem 70 cells is still debatable (15). Most lymphoma, to various degrees, retain their original cellular 71 trait (16). For example, clonal expansion of the immune receptor, collectively known as VDJ 72 (Variable, Diversity and Joining) gene segments, and specifically called B cell receptor (BCR) 73 and T cell receptor (TCR), is used for diagnostic in B and T cell NHL lymphoma (17, 18). 74 However, to what extent these cell traits are kept at single cell level is unknown. Recent studies 75 identified stem-like single cells in leukemic/lymphoma samples (19, 20 ), supporting the 76 scenario where pre-existing chemoresistant lymphoma subclones being selected by treatment, 77 eventually contribute to R/R (21, 22). 78 79 Cell type annotation in physiological single cell RNA sequencing (scRNA-seq) data has seen a 80 tremendous improvement over the years (23, 24), but precise identification of cancer single 81 cells relies on methods customized to cancer type and conditions (25), including unsupervised 82 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 4 clustering, marker genes, copy number alteration /variation (CNA/CNV) (26, 27 ), and for 83 lymphoid malignancies specifically, hyperexpanded immune receptor (28). Reliable cancer cell 84 annotation with CNV inference allows reconstruction of a cross-sectional phylogenetic tree to 85 reveal key diversification events within a sample (29, 30 ). However, comparisons of 86 evolutionary traits between lymphoma subtypes remain scarce. 87 88 In pursuit of a R/R related lymphoma subclone, we here combined scRNA -seq with scVDJ -89 seq, inferred copy number variation (inferCNV, https://github.com/broadinstitute/inferCNV ), 90 single-cell somatic mutations (31), and marker gene expression to reconstruct a phylogenetic 91 tree for each lymphoma sample. We discovered a progenitor -like, chemoresistant status as a 92 common trait of pNHL evolution, representing a compartment that had long been over-looked 93 by conventional methods, while Hodgkin cells in R/R became myeloid-like and were associated 94 with CD74 highCCL5+ CD8 T cells infiltration in the TME. These results pave the way to 95 mechanistically understand pediatric lymphoma and ultimately to develop novel treatment that 96 prevent R/R. 97 98 99

100 101 Characterization of pediatric lymphoid landscape. 102 We prepared scRNA-seq and scVDJ-seq data from fresh lymphoma and reactive lymph node 103 samples collected at Karolinska University Hospital between 2021 to 2025, including non-104 cancerous pediatric reactive lymph nodes (pReLy, n=8, including 1 revisiting: pReLy -01 and 105 pReLy-01b), adults reactive lymph nodes (aReLy, n=2), Burkitt Lymphoma (pBL; n=3), T cell 106 lymphoblastic lymphoma (pTLBL; n=5), classical Hodgkin Lymphoma (pHL; n=5), anaplastic 107 large cell lymphoma (pALCL; n=1), and primary mediastinal B cell lymphoma (pPMBCL; 108 n=1) (Figure 1A, Table 1). After quality control (see methods), 21,851 and 60,032 single cells 109 remained for pReLy and aReLy , with mean unique gene count per sample ranging between 110 2,615~6,587 and 3,819~8,297, and mean unique feature count 1,114~1,987, and 1,375~2,636, 111 respectively. 21 distinct cell types were found by combining a machine learning algorithm 112 trained by public data sets (16, 32) with expert knowledge. To avoid patient specific result in 113 unsupervised clustering, the Ig heavy and light chain loci and TCR alpha/beta loci was excluded 114 from scRNA-seq (33). 115 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 5 116 Figure 1. Determining the lymphoid landscape in pediatric specimens: A. Schematics of the study. 23 freshly 117 isolated reactive and cancerous pediatric (<18 y.o.) lymph nodes, and two reactive adult lymph nodes were 118 included. Samples were prepared for single cell RNA sequencing (scRNAseq), single cell VDJ sequencing 119 (scVDJseq, BCR and TCR), flow cytometry and, for Hodgkin lymphoma (pHL), spatial transcriptomics. 120 (Schematics created in https://BioRender.com ) B. Cell type annotation of integrated reactive lymph node 121 (pediatric and adult) represented with 2D UMAP . C. (Left) Dot plot highlighting the top 2 marker genes for each 122 identified cell type. Size of the dot indicates percentage of expression, variation in color specifies expression. 123 Figure 1 D CD45.high.DZGCB CD45.low.DZGCB Cytotoxic.T Double.neg.T LZGCB Mono/mDC Naive.B Naive.CD4.T Naive.CD8.T NK Non.switch.memB pDC Plasma Progenitor.B Progenitor.T Proliferation.T Switched.memB T.helpers T.exhaust Tfh Treg Pediatric (n=21851) vs Adult (n=21851) + Modified Expression 0.85 0.95 1.05 1.15 Reactive Lymph Node 1st CNVref 2nd CNVref Chr1 Chr2 Chr3 Chr4 Chr5 Chr6 Chr7 Chr8 Chr9 Chr10 Chr11 Chr12 Chr13 Chr14 Chr15 Chr16 Chr17 Chr18 Chr19 Chr20 Chr21 Chr22 Genomic Region 0.00 0.25 0.50 0.75 CNV Score Prog.B Prog.T Cell Frequency Modified Expression Sum Chr 19 2nd CNVref ~20% 1st CNVref ~80% A B C E F G GCB + Plasma NaiveB + MemB + ProgB T+NK Cell Type UMAP1 UMAP2 0.00 0.50 1.00 VDJ Prop. pReLy_01 pReLy_01b pReLy_02 pReLy_03 pReLy_04 pReLy_05 pReLy_06 pRely_07 aReLy_01 aReLy_02 0 20000 Count Sample CD45.high.DZGCB CD45.low.DZGCB Cytotoxic.T Double.neg.T LZGCB Mono/mDC Naive.B Naive.CD4.T Naive.CD8.T NK Non.switch.memB pDC Plasma Progenitor.B Progenitor.T Proliferation.T Switched.memB T.helpers T.exhaust Tfh Treg STMN1HMGB2TUBA1B CCL5GZMK MEF2B HIST1H4C CST3 GZMA LYZFCER2CD74NOSIPCCR7CD8BCD8AGNLYNKG7BANK1RALGPS2GZMBLILRA4JCHAINMZB1ARHGAP24MAML2CAMK4MS4A1ANXA1S100A4ITM2ACTLA4TOX2FOXP3 Cell Type Gene 0 25 50 75 100 2 1 0 -1 Pct. Exp. Avg. Exp. VDJ ClonoType no_VDJ BCR_Large (0.01 < X <= 0.1) BCR_Medium (0.001 < X <= 0.01) BCR_Small (1e-04 < X <= 0.001) TCR_Large (0.01 < X <= 0.1) TCR_Small (1e-04 < X <= 0.001) TCR_Medium (0.001 < X <= 0.01) Sample Prop. 0.0 0.1 0.2 0.3 CD45.high.DZGCBCD45.low.DZGCBCytotoxic.TDouble.neg.TLZGCBMono/mDC Naive.BNaive.CD4.TNaive.CD8.T NK Non.switch.memB pDC PlasmaProgenitor.BProgenitor.TProliferation.TSwitched.memBT.helpersT.exhaustTfhTreg Spliced Unspliced Ambiguous RNA Transcript LEE_DIFFERENTIATING_THADDAD_B_LYMPH_PROG 0.0 0.2 0.4 0.6 0.00 0.25 0.50 2 0 -2 2 0 -2 NES=1.747 p.adj=1.64e-08 Cell Type NES=1.745 p.adj=1.2e-09 RES Ranked Score 5000 10000 15000 5000 10000 15000 CD45.high.DZGCB CD45.low.DZGCB Cytotoxic.T Double.neg.T LZGCB Mono/mDC Naive.B Naive.CD4.T Naive.CD8.T NK Non.switch.memB pDC Plasma Progenitor.B Progenitor.T Proliferation.T Switched.memB T.helpers T.exhaust Tfh Treg UMAP1 UMAP2 Adult Pediatric 0.0 0.5 1.0Norm. Prop. 0.0 0.2 0.4 0.6 0.8 1.0 Proportion Pediatric sample Adult sample Reactive lymph node (pReLy; n=8; 60032 cells) Burkitt lymphoma (pBL; n=3; 11290 cells) T cell lymphoblastic lymphoma (pTLBL; n=5; 24367 cells) Hodgkin lymphoma (pHL; n=5; 57437cells) Anapastic large cell lymphoma (pALCL; n=1; 355cells) Primary mediastinal B cell lymphoma (pPMBCL; n=1; 1293 cells) Reactive lymph node (aReLy; n=2; 21851 cells) ValidationFlow cytometry Animal model IHC Single cell RNA seq Quality Control Spatial Transcriptomics on pHL (n=4) BCR &TCR seq κ/λ ratio VDJ clonality IgH/L chain mutation Downstream analysis DEG inferCNV mRNA splicing GSEA/ORA … DZGCB LZGCB Naive.CD4 pDC Plasma Tfh Texhaust Treg Cytotoxic. T memB NaiveB ProgTProgB Naive.CD8 NK T.helpers Prolif.T Mono/mDC ProgB ProgT (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 6 (Middle) Proportion of VDJ clonotype by cell type. (Right) Proportion of each cell type distribute across all sample 124 integrated in B. D. Transcript splicing status by cell types. E. Gene set enrichment analysis of progenitor B and T 125 cells against progenitor B and differentiating T cell gene sets, respectively. F. (Left) UMAP distribution of adult 126 lymph node cells (n= 21,851 cells) and randomly selected 21,851 pediatric cells. (Right) Proportion of sample 127 source by each cell type. Arrowheads indicate progenitor clusters. G. (left) Representative inferCNV heatmap of 128 reactive lymph node sample. Each row represents a randomly selected cell, and each column represents a gene 129 ordered by chromosomes. Color code indicates inferred chromosome gain/loss. (Top right) CNV score distribution 130 across reactive lymphoid landscape. (Lower right) separation of CNV reference based on chromosome 19 131 expression resulted in two compartments: 20% with high CNV score (2nd CNVref) and the rest 80% (1st CNVref). 132 133 To gain insights into how pediatric immune response differ from adults, w e first generated a 134 cell atlas of reactive lymph nodes consisting of pReLy and aReLy. Using the uniform manifold 135 approximation and projection (UMAP), three main clusters were observed; cluster 1) T and NK 136 cell cluster included different T cell populations ; n aïve CD4/CD8, mature helper T cells , 137 cytotoxic T cells , and NK cells , as well as a n unique, progenitor-like T cell cluster (ProgT); 138 cluster 2) naïve and memory B cell which also contained a unique compartment termed 139 progenitor-like B cells (ProgB); and cluster 3) germinal center B cells (light zone; LZGCB and 140 dark zone; DZGCB ) and plasma cell cluster captured highly proliferative germinal center B 141 cells and proliferating T cells (Figure 1B and Figure S1A). Smaller clusters included more rare 142 populations such as monocytes/dendritic cells (DC), and plasmacytoid DCs ( Figure 1B ). 143 Marker genes were in line with each mature cell type whereas ProgB/T shared MAML2, a 144 NOTCH signaling protein (Figure 1C). We also observed substantial drop of VDJ expression 145 among ProgB/T as compared to effector cell types (~40% Figure 1C). As expected, we did not 146 observe hyperexpanded VDJ clones in these samples (Figure 1C). Interestingly, ProgB/T were 147 found abundant in 6 out of 7 pediatric patients, but underrepresented among adults (Figure 1C), 148 suggesting progenitor -like B/T cells are unique to the pediatric lymphoid landscape . 149 Furthermore, ProgB/T cells contained ~10% more unspliced mRNA ( Figure 1D ) and 150 significantly enriched in progenitor cell gene sets (Figure 1E), strongly suggesting a progenitor 151 status. Detailed investigation of ProgB/T cells showed downregulation of translation related 152 genes and upregulation of NOTCH signaling gene sets in both (Figure S1B), and WNT pathway 153 genes in ProgT cells (Figure S1C ). Sample-size balanced comparison between pReLy and 154 aReLy revealed that pReLys had a larger population of naïve and ProgB/T cells (arrowheads, 155 Figure 1F ) while adult samples were dominated by mature, effector cells (Figure 1 F), 156 indicating ProgB/T compartment was associated with an overall naïve, inexperienced immune 157 landscape. 158 Chromosomal CNV inferred from transcriptome data has emerged as one of the mainstays for 159 single cancer cell detection, for which a reliable reference is essential. We used inferCNV to 160 explore whether the pReLy dataset is suitable for this purpose. We found two distinctive sets of 161 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 7 inferred CNV background with the majority (80%) near baseline, termed 1 st CNVref, and a 162 smaller set (~20% of cells) signified by chromosome 19 loss and short arm chromosome 10 163 gain (2nd CNVref) and high CNV score overall (Figure 1G). Interestingly, the CNV score was 164 concentrated in the ProgB/T clusters (Figure 1G), suggesting that the main contributors of 2nd 165 CNVref were ProgB/T cells. This was confirmed by overlapping of up/down regulated genes 166 of ProgB/T cells with the 2nd CNVref gain/loss region (Figure S1D), indicating that the 2nd 167 CNVref was transcriptomically driven and indeed represented as a different tran scriptomic 168 status. Moreover, the size of 2 nd CNVref compartment was proportional to ProgB/T in each 169 sample (Figure 1C, S1E-M), agreeing with the link between ProgB/T and 2nd CNVref. In short, 170 we created a reactive, non-cancerous single cell lymphoid atlas and found progenitor-like cells 171 as a unique trait of pediatric immune landscape that need to be considered separately when 172 using as reference in inferCNV . 173 174 A single-cell atlas of pediatric lymphoma reveals origin-specific traits. 175 To gain an overview of our cohort, we created a pediatric lymphoma atlas by integrating pReLy 176 with all lymphoma sample s. We found presence of non-cancerous cell types in all lymphoma 177 samples, while lymphoma exclusive clusters fall outside the three main clusters , suggesting 178 abnormal cell status (Figure 2A). Breakdown by cell origin showed single cells from B cell 179 originated lymphoma (pBL, pPMBCL) clustered more frequently with B cell than T cells, with 180 pPMBCL higher in myeloid and regulatory T (Treg) cell infiltration (Figure 2B, upper left). 181 For pHL, we observed an even distribution between T and B cell clusters but major expansions 182 in the myeloid cluster and lymphoma-exclusive region that linked plasma cells and progenitor 183 T cells ( Figure 2B , upper right ). For T cell originated lymphoma (pTLBL , pALCL ), a 184 lymphoma-exclusive cluster adjacent to progenitor T cells was detected (Figure 2B , lower 185 right). 186 To investigate how lymphoma single cells distributed throughout the atlas, we leveraged 187 scVDJ-seq data that identifies lymphoma single cells signified by possessing hyperexpanded 188 immune receptors. For BCR, we found 1 dominant clone (>90%) in each of the 3 pBL samples 189 that transcriptomically span across memory (MemB), light zone (LZGCB) and dark zone 190 germinal center B cells (DZGCB) , but not lymphoma-exclusive clusters (Figure 2C ), 191 suggesting pBL largely retain its canonical B cell program after a single transformation event. 192 A less dominant BCR clone was found in pPMBCL, but with IGHE as constant chain, pointing 193 to its post germinal center origin (Figure 2C). Interestingly, a hyperexpanded, class-switched 194 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 8 B cell clone was found in pALCL, a T cell lymphoma, indicating effector B cell infiltration 195 (Figure 2C). 196 197 Figure 2. A single-cell atlas uncovers lymphoma origin-specific characteristics. A. Integrated scRNAseq UMAP 198 over all pReLy and lymphoma (in grey) samples. B. (top) scRNAseq UMAP plotted by pathological origin. C-D. 199 Proportion of top 3 BCR and TCR clones per sample. Hyperexpanded BCR clones; pBL_01: 200 IGHV1−46.NA.IGHJ4.IGHM_IGLV2−11.IGLJ2.IGLC2, pBL_02: 201 IGHV3−74.NA.IGHJ6.IGHM_IGLV2−18.IGLJ3.IGLC3, pBL_03: 202 IGHV3−30.NA.IGHJ4.IGHM_IGKV2−24.IGKJ2.IGKC and hyperexpanded TCR clones; pTLL_02: 203 TRA V25.TRAJ45.TRAC_TRBV29−1.NA.TRBJ2−1.TRBC2; pTLL_03: NA_TRBV7−9.NA.TRBJ1−1.TRBC1, 204 pTLL_04: NA_TRBV20−1.TRBD1.TRBJ1−2.TRBC1). (Bottom) Contour plots of hyperexpanded (Proportion > 205 0.1) immune receptor distribution. Color code matches that of the bar plot. E. Proportions of predicted cell types 206 of single cells from lymphoma samples. 207 208 For TCR, we found hyperexpanded clone s existed only in 3 out of 5 of the pTLBL samples, 209 with the other 2 lacking hyperexpanded TCR clones, suggesting that pTLBL transformation 210 occurred during different stages of T cell development in the thymus (Figure 2D). Intriguingly, 211 cells from pTLBL occupied a lymphoma exclusive cluster proximal to progenitor T cells and 212 the DZGCB cluster , signifying pTLBL having a unique transcriptome and existence of a 213 UMAP1 UMAP2 CD45.high.DZGCB CD45.low.DZGCB Cytotoxic T Double.neg.T Lymphoma Sample LZGCB Mono/mDC Naive B Naive CD4 T Naive CD8 T NK Non.switch.memB pDC Plasma Progenitor B Progenitor T Proliferation T Switched.memB T helpers Texhaust Tfh Treg GCB+Plasma NaiveB+MemB+ProgB T+NK A B C D E B cell origin Hodgkin Reactive T cell origin Diagnosis ALCL Burkitt Hodgkin PMBCL Reactive TLBL Cell Type UMAP1 UMAP2 pReLy pBL pTLBL pHL pALCL pPMBCL pBL_01pBL_02pBL_03 0.00 0.25 0.50 0.75 1.00 Leading BCR Clones Proportion pReLy pHL pTLBL pALCLpPMBCL Heavy+Light Light only Hyperexpanded (Prop. >0.1) Alpha+Beta Beta only 0.00 0.25 0.50 0.75 Proportion pReLy pHL pALCLpPMBCL pBL pTLBL_01pTLBL_02pTLBL_03pTLBL_04pTLBL_05 Leading TCR Clones Hyperexpanded (Prop. >0.1) pALCL_01 pBL_01 pBL_02 pBL_03 pHL_01 pHL_02 pHL_03 pHL_04 pHL_05 pPMBCL_01 pTLBL_01 pTLBL_02 pTLBL_03 pTLBL_04 pTLBL_05 CD45.high.DZGCB CD45.low.DZGCB Cytotoxic T Double.neg.T LZGCB Mono/mDC Naive B Naive CD4 T Naive CD8 T NK Non.switch.memB pDC Plasma Progenitor B Progenitor T Proliferation T Switched.memB T helpers Texhaust Tfh Treg Unassigned Diagnosis Diagnosis ALCL Burkitt Hodgkin PMBCL TLBL Proportion 0.6 0.4 0.2 0.0 Predicted Cell Type 0 10000 20000 Count DZGCB LZGCB MemB Naive.B ProgB Mono/mDC Plasma Tfh NK pDC ProgT Cytotoxic.T Th Naive.CD8 Naive.CD8 Treg Texhaust Figure 2 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 9 proliferating compartment (Figure 2D). To investigate transcriptome changes across individual 214 samples, we trained a cell type predictor with pReLy data and applied on all lymphoma samples 215 (Figure 2 E). High plasma cell content was observed in pALCL , validating effector B cell 216 infiltration observed in Figure 2C . We observed that pBL was predicted to have a h igh 217 proportion of proliferating cells (DZGCB) followed by pTLBL, whereas pHL had less of 218 DZGCB cells. pTLBL had an increased proportion of naïve CD4 T cells and progenitor T cells 219 and had the largest enrichment of unassigned cells, agreeing with previous observation of a 220 large lymphoma-exclusive cluster. pHL was enriched in a wide range of cell types including 221 naïve B/T, exhausted T, T helper cells, progenitor T cells, and various degree of myeloid cells 222 (Figure 2E). In summary, we created a pediatric lymphoma atlas to reveal unique transcriptome 223 and cell type heterogeneity across different origins. 224 225 pBL acquires a progenitor-like, BCR negative program and upregulates MSI2. 226 In search of potential pBL resistant subclones, we integrated 11,290 single cells from pBL 227 samples with pReLy. Most of the single cells from pBL clustered with germinal center (GC) B 228 cells (Figure 3A), suggesting Burkitt lymphoma cells largely retain its GC status . This was 229 confirmed by scVDJ-seq data that revealed lymphoma cells, signified by possessing 230 hyperexpanded BCR, spanning across DZGCB, LZGCB , and memB status (Figure 3B ). 231 Detailed BCR characterization showed deviated light chain usage to either kappa or lambda 232 (Figure 3C), and lack of VDJ diversity among dominant clones in pBL samples (Figure 3D), 233 with pBL_01 and pBL_03 dominated by a single VDJ and CDR3 sequence (up to 90%) and no 234 diversity found in pBL_02 (VDJ and CDR3 100% identical, Figure S2), as confirmed by flow 235 cytometry (Figure S3A). This is clearly different from the extensive diversity exhibited by 236 pReLy (Figure 3C,D). The evidence indicated that pBL derived from a BCR single cell yet, 237 upon transformation, diversified transcriptomically. 238 To capture this diversification after transformation , we subclustered each pBL sample with 239 inferCNV (Figure 3E-G). We noticed that CNV for pBL_03 S5 resembled ProgB/T cells in the 240 CNV landscape, meaning the inference could have been skewed by obtaining the progenitor-241 like program. We revealed the true CNV of this cluster separately with the 2nd CNVref data set 242 (Figure S1C, S4C). Combining several key factors describing each subclone including VDJ 243 clonotype, MYC expression, CNV score and transcriptome landscape, we were able to annotate 244 these subclones as either Burkitt or Support (defined as non-cancerous cells from the lymphoma 245 sample) and reconstructed phylogenetic trees based on CNV accumulation ( Figure 3F,G). 246 Since inferCNV is gene count-sensitive only, we validated the phylogenetic tree structures by 247 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 10 a sequence-sensitive approach: we detect ed transcriptome-wide, expressed single nucleotide 248 variation (SNV) at single-cell level via SComatic and calculated Jaccard distances between pBL 249 subclones based on shared SNV. The result indicated a bifurcation during cancer evolution in 250 pBL_03, but not pBL_01 and 02, with S6 isolated from all other subclones ( Figure S4D), 251 agreeing with our inferCNV-based phylogenetic tree. 252 Interestingly, we observed a BCR-negative compartment (BLBCR-) in pBL_03, a bone marrow 253 sample with the most advanced stage IVB (Table 1), increasing in proportion from S1 to S2, 254 eventually acquired progenitor-like program in S5 (Figure 3G). This suggests that progenitor-255 like status was favored under selection pressure. The BLBCR- compartment was negligible in 256 pBL_02, a nodal sample also staged to IVB, suggesting this evolution ary trend could be 257 localized and dependent on environment. Overall, BLBCR- largely retained BLBCR+ traits, such 258 as MYC expression, across cell clusters capturing most lymphoma cells ( black arrowhead, 259 Figure 3H). However, when compared to BLBCR+, BLBCR- had a distinct transcriptome (Figure 260 3I) and was enriched in Progenitor B cell gene set s (Figure 3J), adding to the evidence that 261 pBL_03 was evolving to wards a progenitor-like status. Furthermore, c omparison between 262 BLBCR+, BLBCR-, and non -BL cells revealed 9 genes specifically expressed by BL cells with 263 higher expression in BLBCR-, among which MSI2 was the top hit (Figure 3K). MSI2 expression 264 is often linked to hematopoietic and leukemic stem cells (34-36). Indeed, MSI2 protein level 265 exhibited extensive heterogeneity in pBL samples compared to pReLy, confirming existance of 266 a high MSI2 compartment in pBL (Figure 3L, S3C). Additionally, pBL were characterized by 267 high expression of MYC, Ki67 and γH2AX remarking their cancerous nature ( Figure 3L ). 268 Combining these results , we discovered pBL transcriptomically diversify itself after 269 transformation into a progenitor-like BCR-negative subclone in advanced disease. 270 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 11 271 Figure 3. pBL obtains a progenitor-like, BCR negative program and upregulates MSI2: 272 A. Integration of pReLy (in color) with three pBL samples (in grey). B. BCR/TCR clonotype of pBL single cells. 273 C. Proportion of Ig light chain usage (kappa, IGKC or lambda, IGLC) across pBL and pReLy samples. D. 274 Phylogenetic analysis of Ig heavy chain CDR3 sequences for hyperexpanded BCR clones from pBL_01 and pBL-275 03, and a representative leading clone from pReLy_07. E. (Left) inferCNV subclustering results of pBLs using the 276 1st or 2nd CNVref as reference. (Right) UMAP distribution of CNV subclones eventually annotated as “Support” 277 Figure 3 A B F G I UMAP1UMAP2 CD45.high.DZGCB CD45.low.DZGCB Cytotoxic.T Double.neg.T LZGCB Mono/mDC Naive.B Naive.CD4.T Naive.CD8.T NK Non.switch.memB pDC Plasma Progenitor.B Progenitor.T Proliferation.T Switched.memB T.helpers T.exhaust Tfh Treg Cell Type Burkitt SamplepReLy pBL T Cell NaiveB+MemB+ProgB GCB+Plasma UMAP1UMAP2 BCR_Hyperexpanded (0.1 < X <= 1) BCR_Large (0.01 < X <= 0.1) BCR_Medium (0.001 < X <= 0.01) BCR_Small (1e-04 < X <= 0.001) no_VDJ TCR_Hyperexpanded (0.1 < X <= 1) TCR_Large (0.01 < X <= 0.1) TCR_Medium (0.001 < X <= 0.01) TCR_Small (1e-04 < X <= 0.001) Clonotype pBL_01pBL_02pBL_03 S1 41.1% S2 34.0% S3 24.8% S1 7.0%\60.3% S2 4.5%\28.2% S1 35.6% S2 24.2% S3 7.4%\14.3% S6 0.1%\4.6% 1st 2nd Support Burkitt Chr1 Chr2 Chr3 Chr4 Chr5 Chr6 Chr7 Chr8 Chr9 Chr10 Chr11 Chr12 Chr13 Chr14 Chr15 Chr16 Chr17 Chr18 Chr19 Chr20Chr21 Chr22 Genomic Region 0 1 2 3 4 5 6 Modified Expression Burkitt Support CNVref Cell TypeSubclone E C D Proportion0.00 0.50 1.00 IGLC IGKC Burkitt Sample Reactive Lymph node pBL_01 pBL_02 pBL_03 pReLy_01 pReLy_02 pReLy_03 pReLy_04 pReLy_05 pReLy_01b pReLy_06 pReLy_07 Light Chain Sample S2 S1 9q amp 11q del 22 amp 6q del 7p amp 9p amp S3 S2 6q del 7q amp S3 S2 1q del S5 S6 4p amp 6q amp 13p amp 18p amp 12p del S1 pBL_01 pBL_02 pBL_03 CNV Accumulation Sx InferCNV Subclone Clonotype Proportion 0.00 0.50 1.00 pBL_01 pBL_02 pBL_03 Subclone 2 4 0 MYC Level S1 S2 S3 S1 S1 S2 S2 S1 S2 S3 S3 S4 S5 S6 S6 Prog.BCNV Score Prog.TEnrich CNVref 0.6 0.0 0.3 CNV Subclone Support Burkitt Sx Sx CNVref 1st 2nd p.adj 0.10.20.30.4 Gene Ratio 0.50 0.75 Enrichment Clonotype BCR_Hyperexpanded (0.1 < X <= 1) BCR_Large (0.01 < X <= 0.1) BCR_Medium (0.001 < X <= 0.01) BCR_Small (0.0001 < X <= 0.001) no_VDJ TCR_Hyperexpanded (0.1 < X <= 1) TCR_Large (0.01 < X <= 0.1) TCR_Medium (0.001 < X <= 0.01) TCR_Small (0.0001 < X <= 0.001) Reactive Support BLBCR+ BLBCR- Others LZGCB like Prolif. BL dominate H 0 1 2 -2 -1 Exp. MALAT1 MSI2 ANKRD11 FOXP1 BACH2 … EEF1A1 SLC25A5 PRDX1 ACTG1 JCHAIN BTF3 … BLBCR- BLBCR+ BL vs non-BL Lfc > 1 BLBCR- vs BLBCR+ Lfc > 1 BLBCR+ vs BLBCR- Lfc > 0.5 46 9 40 0 24838 00 1 2 3 4 5 0 1 2 3 4 5 BL+BL- ReLy Supp PDE4B MSI2 Expression. Top 3 J K Cell Type LZGCB likeProliferatingBL dominate Support Reactive BLBCR+ BLBCR- MYC TCL1B ZNF385B PLAC8 LPP RGS13 CD83 FOS AC023590.1 CTSH RPS10 RPS4X RPS3A RPL6 SNHG29 RPL26 MTRNR2L8 BX2846132.2 MT-ND1 MT-CYB TFDP2 CDKN2ATCL1B MYC CD24 DNAJB1 SNHG5 LPP RGS13 LTB CRIP1 S100A10 AC023590.1 RPS10 RPS4X HSPB1 PHGDH ID3 MTRNR2L8 AC105402.3 MT-CO3 MT-ND1 MT-CYB MT-ND4L EIF4B MT-ND1 RPS18 DNAJB1 RPLP0 RPLP1 RPS7 SNHG29 RPL5 HSPB1 EEF1A1 LPP AC023590.1 ARHGAP24 TOX MAML3 BANK1 MYC BX284613.2 MIF ZNF385B HIST1H4C RPL13A MEF2C PDGFD Avg. Exp. 1.0 0.5 0.0 -0.5 -1.0 25 50 75 100 Pct. Exp. Support Reactive BLBCR+ BLBCR- Reactive BLBCR+ BLBCR- Marker Genes Running ES 0.5 0.0 -0.5 2500 5000 7500 1 0 -1 Rank ScoreBLBCR- BLBCR+ Rank in Dataset Progenitor B NES:3.069 P .adj:1.5e-118 CD45.high.DZGCB NES:-3.241 P .adj:1.3e-40 Mutations from Germline pBL_01 Dominant BCR Clone pBL_03 Dominant BCR Clone pReLy_07 Dominant BCR Clone 30 31 32 33 12 13 14 15 53 63 73 83 0-1% 1-5% 10-20% 80-90% Percentage Inferred Ancestral Subclone L 100 80 60 40 20 0 100 80 60 40 20 0 -103 103 104 1050 -103 103 104 1050 Normalised Density Signal Intensity MSI2 MYC Ki67 γH2AX Unstained pReLy_07 pBL_03 pBL S4 7.0% S5 6.8% ProgT Naive.CD8 Cytotoxic.T Treg Texhaust Tfh Th Naive .CD4 NK Mono/mDC Naive.B MemB LZGCB DZGCB Plasma n=282 n=2109 n=8899 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 12 (in light green, defined as non -lymphoma cells from lymphoma samples) or “Burkitt” (in dark blue). F. 278 Phylogenetic tree of Burkitt subclones based on CNV accumulation and SNV sharing. G. Key characteristics of 279 CNV sublcones from E. (From top) BCR/TCR clon otype proportion, MYC expression, CNV score (dotted line 280 indicates 0.31), over representation analysis using progenitor B/T marker gene sets, CNVref used, and eventual 281 Burkitt/Support subclone annotation. H. (Top) Three manually curated cell clusters (LZGCB like, Proliferating, 282 and BL dominate) and their composition. Dotplot shows top marker genes of each cell type across these clusters. 283 I. Transcriptome difference between Burkitt cells with (BLBCR+) or without (BLBCR-) functional BCR. J. Gene set 284 enrichment analysis of ranked transcriptome comparing BLBCR- to BLBCR+ against DZGCB and Progenitor B gene 285 sets. K. Venn diagram showing shared upregulated genes across overall Burkitt cells (BL), BL BCR+ and BLBCR-. 286 Expression of top genes shared by BL and BL BCR- was shown as violin plot on the left. L. Representative flow 287 cytometry histograms of MSI2, MYC, Ki67 and γH2AX expression for pReLy_07 and pBL_03. 288 289 Progenitor-like pBL sublones initiate evolution in R/R 290 We suspected that the BLBCR- compartment represent a stem-like niche that contributes to R/R 291 and applied our pBL pipeline to external pBL scRNA-seq data set (28, 32 ). We detected 292 progenitor-like pBL subclones in 7 out of 12 samples with or without primary refractory (66.7% 293 R/R, 55.6% non-R/R), all harbored considerable proportion of BLBCR- cells (red boxes, Figure 294 S5A,B). Upon phylogenetic tree reconstruction, we observed progenitor-like BL subclones 295 among non-R/R samples located terminally at their branches (KC_108, KC_112, KC_113 and 296 KC_115, Figure S 5C) while those in R/R-related samples were clearly initiating daughter 297 clones (KC_104a and KC_106a, Figure S5C), suggesting the maturation of a stem-like niche 298 preceded R/R (Figure S5C). Of note, we found KC_101, a non-R/R related sample yet highly 299 expressing the poor prognostic marker TPM2 (28), had a progenitor-like initiating clone 300 (Figure S5B,C), indicating this patient was high risk but somehow did not develop R/R. Finally, 301 we observed that MSI2 had higher expression in the BLBCR- compartment comparing to others, 302 confirming that MSI2 upregulation is related to R/R in pBL (Figure S5D). To sum up, we 303 confirmed MSI2 upregulation in progenitor-like pBL subclones with external data set and 304 revealed them often as initiating clones R/R samples. 305 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 13 306 Figure 4. Identification of progenitor-like, TCR negative pTLBL compartment: 307 A. Integration of pReLy (in color) with pTLBL samples (in grey). B. BCR/TCR clonotype of pTLBL single cells 308 C. (Left) inferCNV subclustering results of pTLBLs using the 1 st or 2nd CNVref as reference. (Right) UMAP 309 distribution of CNV subclones eventually annotated as “Support” (in light green) or “TLBL” (in orange red). D. 310 Key characteristics of CNV sublcones from C. (From top) BCR/TCR clonotype proportion, CNV score (dotted 311 line indicates 0.19), over representation analysis using progenitor B/T marker gene sets, CNVref used, and eventual 312 Figure 4 A B C D F H G I CD45.high.DZGCB CD45.low.DZGCB Cytotoxic.T Double.neg.T LZGCB Mono/mDC Naive.B Naive.CD4.T Naive.CD8.T NK Non.switch.memB pDC Plasma Progenitor.B Progenitor.T Proliferation.T Switched.memB T.helpers T.exhaust Tfh Treg Cell Type TLBL Sample GCB+Plasma NaiveB+MemB+ProgB T cell pReLy pTLBL UMAP1UMAP2 BCR_Hyperexpanded (0.1 < X <= 1) BCR_Large (0.01 < X <= 0.1) BCR_Medium (0.001 < X <= 0.01) BCR_Small (1e-04 < X <= 0.001) no_VDJ TCR_Hyperexpanded (0.1 < X <= 1) TCR_Large (0.01 < X <= 0.1) TCR_Medium (0.001 < X <= 0.01) TCR_Small (1e-04 < X <= 0.001) Clonotype UMAP1UMAP2 Chr1 Chr2 Chr3 Chr4 Chr5 Chr6 Chr7 Chr8 Chr9 Chr10 Chr11 Chr12 Chr13 Chr14 Chr15 Chr16 Chr17 Chr18 Chr19 Chr20 Chr21 Chr22 Genomic Region Support TLBL1st 2nd Support TLBL S1 49.8% S2 27.9% S3 22.4% 0 321 4 5 6 Modified Expression CNV ref Subcluster SingleCell CNVref Subcluster S1 26.0% S2 25.0% S3 16.2% S4 6.3% S5 13.7% S6 12.8% S1 47.7% S2 28.1% S3 4.8%\15.1% S4 4.4% S1 67.8% S2 14.8% S3 6.3% S4 11.2% S1 49.6% S2 32.4% S3 18.0% pTLBL_01pTLBL_02pTLBL_03pTLBL_04pTLBL_05 6q del 10p amp 20p amp 22 amp 4q amp 6q del 8p del 22q amp 3q amp 21 amp 2p del 7q del 19 amp 4p amp 10p amp pTLBL_01 pTLBL_02 pTLBL_03 4q amp 10p amp 19q del 15p amp 9q amp 6p amp 6q del 5q del 21 amp pTLBL_04 10p amp 21 amp 6p amp 8 amp 15p amp 6q del 11q del 1p amp 21 amp pTLBL_05 Sx Inferred Ancestral Subclone InferCNV Subclone CNV Accumulation S1 S3S2 S2 S4 S3 S1 S6 S2S1 S3 S3 S4S1S2 S1 S2 S3 S4 S1 S2 S3 S1 S2 S3 S3 S4 S1 S2 S3 S4 S5 S6 S1 S2 S3 pTLBL_01 pTLBL_02 pTLBL_03 pTLBL_04 pTLBL_05 1.00 0.50 0.00 Clonotype Proportion CNV Score Prog.B Prog.T CNVref 0.4 Enrich Subclone 0.2 0.0 Support TLBL TCR Chain CNV Subclone Beta_only Alpha_Beta Enrichment p.adj 0.1 0.2 0.3 GeneRatio 0.50 0.75 E 1st 2nd CNV ref Sx Sx BCR_Hyperexpanded (0.1 < X <= 1) BCR_Large (0.01 < X <= 0.1) BCR_Medium (0.001 < X <= 0.01) BCR_Small (1e-04 < X <= 0.001) no_VDJ TCR_Hyperexpanded (0.1 < X <= 1) TCR_Large (0.01 < X <= 0.1) TCR_Medium (0.001 < X <= 0.01) TCR_Small (1e-04 < X <= 0.001) Clonotype Proliferating TLBL_niche Prog_like OthersTT PTCRAβ T β α TLBL Regions TLBLTCR+ TLBLpreTCR TLBLTCR- ProliferatingTLBL_nicheProg_like TLBLTCR+ TLBLTCR- TLBLpreTCR Support Reactive Support Reactive Support TCL1A CD74 IGKC CD79A MS4A1 IGHM MEF2B FOS DNAJB1 PRPSAP2 PDE4D CD69 GMDS CD1E CD1B TRBC1 PTCRA CD3D ITM2A CD3E SELL H1FX CTHRC1 RNF220 CSMD1 SUPT3H HSPA1B MTRNR2L8 FXYD2 CHI3L2 RNF168 TUBA1A JUNB TSC22D3 FOS HSPA1A TXNIP CD52 IL7R CD74 DNAJB1 KLF6 CD69 CD1E PTCRA CD1B LST1 CD96 THEMIS XG MEF2C RUNX2 RNF220 SELL CSMD1 CTHRC1 TCF4 MTRNR2L8 FXYD2 RNF168 CALN1 IGFBP5 PTPRD ADAMTS17 HSPH1 HSPA1A FOS NRXN3 JUN CD74 LYST HSPA1B DNAJB1 HSPE1 CD1E ITM2A CD1B PTPRC H1F0 PTCRA ACTB LTB PSAP CD3E CD96 RNF220 RUNX2 SUPT3H RIMS2 MED12L MEF2C CSMD1 TCF4 PROM1 BCL2 MSI2 % Exp. Avg. Exp. 1 0 -1 0 25 50 75 100 Precursor TALLPrecursor BALL Multiple Myeloma Hodgkin Lymphoma DLBCL-GCB Burkitt Lymphoma ALK+ ALCL Mantle Cell Lymphoma n=145 n=19 n=14 n=71 n=9 n=4 MSI2 TPM MSI2 Dependency 8 6 4 0.0 -0.4 -0.8 1.9e-07 0.0023 0.0064 0.03 Malignancy Type Other Lymphoid Precursor BALL Precursor TALL p=0.015 high (n=31) low (n=17) MSI2 Expression 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 p=0.046 high (n=121) low (n=41) MSI2 Expression 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0Overall Survival Prob. Follow Up in Months Cutaneous T-cell Lymphoma Liu et al. 2022 Peripheral T-cell lymphoma Iqbal et al. 2014 Overall Survival Prob. Follow Up in Months 0 10 20 30 40 50 60 70 80 90 100 110 0 20 40 60 80 100 120 140 160 180 200 220 240 J K Enrichment Score0.0 0.2 0.4 0.6 0.8 0 2 -2 Ranked Metric Bone Marrow Progenitor-like Xu et al 2024 NES=1.906 p.adj=3.44e-07 TLBLTCR- TLBLpreTCR/TCR+ 5000 10000 15000 Ranked Gene MSI2 132 TLBLTCR+ TLBLTCR- TLBLpreTCR TLBLTCR- TLBLpreTCR pTLBL DZGCB NK Mono/mDC pDC Plasma memB ProgB ProgB Treg Naive.CD4 Naive.CD8 Th Tfh Texhau. Cytotoxic.T ProgT n=6218 n=679 n=5619 n=7016 n=4835 MSI2 Exp. 0 2 4 6 Reactive Support TLBLpreTCR TLBLTCR- TLBLTCR+ Cell Status (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 14 TLBL/Support subclone annotation. E. Phylogenetic tree of pTLBL subclones based on CNV accumulation and 313 SNV sharing F. Contour plot of TCR status (TCR_pos, Pre_TCR and TCR_neg) distribution of pTLBL single cells 314 and their composition of manually curated cell clusters (Proliferating, TLBL_niche and Prog_like). G. Top marker 315 genes of pTLBL (by TCR status), Support and Reactive cells across manually curated cell clusters in F. H. MSI2 316 expression across all cell compartments . I. Enrichment analysis of bone marrow progenitor -like cells (BMP) 317 related to TLBLTCR- and TLBLTCR+ cells. J. (left) Cancer type composition of all lymphoid origin cell lines on the 318 Dependency Map (DepMap 2025 Q2). Cell lines annotated as precursor BALL and TALL (based on Expasy 319 Cellosaurus) shown in colors. (right) MSI2 expression and dependency score by precursor status. K. Survival 320 curve of closely related T cell lymphoma types by MSI2 expression level. 321 322 Identification of progenitor-like, TCR negative pTLBL compartment 323 To investigate whether the progenitor-like program also participate in pTLBL 324 lymphomagenesis, we integrated 24,367 pTLBL single cells from 5 samples with the annotated 325 pReLy dataset (Figure 4A). CDR3 analysis showed single-dominating TCR clones in 326 pTLBL_02, 03, and 04 but not pTLBL_01 and 05, agreeing with our observation in Figure 2D 327 (Figure S6). Despite sharing many normal cell clusters with pReLy, a cell cluster exclusively 328 made up of pTLBL cells was found bridging the T cell and proliferating cluster (Figure 4A) 329 and consisted mostly of single cells with hyper-expanded TCR (Figure 4B), suggesting this is 330 a lymphoma specific cluster. CNV inference using 1st CNVref revealed progenitor-like 331 subclusters across all pTLBL samples , which we then isolated and processed by 2 nd CNVref 332 (Figure S 7A-E). We annotated inferCNV-derived pTLBL s ubclusters as either TLBL or 333 Support based on TCR clonality, CNV score, and transcriptome (Figure 4C-D). As expected, 334 many lymphoma subclones that required the 2nd CNVref were enriched in progenitor-like 335 program (red boxes, Figure 4D), most of which appeared evolutionary terminal (red boxes, 336 Figure 4E). Jaccard distance on shared SNV were used to validate major branching events 337 (Figure S7F). Of note, S3 in pTLBL_03 was determined to give rise to S4 merely because it 338 had no predicted CNV and, by the assumption of CNV accumulation, is the absolute root of the 339 phylogenetic tree instead of matching S4 CNV pattern (Figure 4C). Thus, it is unclear whether 340 pTLBL_03 had an established initiating clone, despite having an advanced disease stage IV A 341 (Table 1). 342 In pTLBL_03 and pTLBL_04, the hyperexpanded TCR clones consisted of only TCR beta 343 chain (Figure 4D). This could not be explained by preferential beta chain coverage over alpha 344 chain during scVDJ-seq as the preference occurs at a limited extent (37). Given that the TCR 345 beta chain recombination precedes that for the alpha chain (38), it is likely our pTLBL samples 346 originated from different T cell developmental stages . Specifically, pTLBL_01 and 05 were 347 TCR negative upon transformation while pTLBL_02 expressed canonical TCR and pTLBL_03 348 and 04 expressed preTCR. Thus, we compartmentalized lymphoma cells based on their TCR 349 status: TCR positive ( TLBLTCR+), Pre-TCR (TLBLpreTCR) and TCR negative ( TLBLTCR-). We 350 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 15 also manually curated three cellular clusters covering the lymphoma specific region (TLBL 351 Regions, Figure 4F ). Transcriptomically, all three categories shared the region bridging T 352 helper cells and proliferating cells (TLBL_niche, Figure 4F), suggesting that the equilibrium 353 between T cell identity and proliferation is essential to lymphomagenesis. However, in the 354 TLBLpreTCR, a cluster located towards the progenitor-T cell cluster was observed, which further 355 became a separate cluster in TLBLTCR- (Prog_like, Figure 4F ), indicating increased 356 representation of progenitor-like program in samples with undifferentiated origin. To validate 357 these findings, we compared top marker genes of lymphoma cells from all compartments across 358 the three manually curated regions (Figure 4 G). Pre-TCR alpha chain ( PTCRA) was 359 consistently found among top marker genes for lymphoma cells from TLBLpreTCR (black 360 arrowheads, Figure 4G ), confirming usage of pre-TCR at the moment of transformation . 361 Secondly, lymphoma cells from all three categories had distinct marker genes that pointed to 362 different cellular status. For example, THEMIS, a protein critical to T cell development (39), 363 was found in TLBLpreTCR, whereas RUNX2, a marker for poor prognosis in T-ALL (40), was 364 found in TLBLTCR-. For TCR_TLBLTCR+, we found FXYD2, a gene described in intra -thymic 365 progenitor T cells (41) and in CAR-T cells (42). Lastly, we identified MSI2 as one of the top 366 marker genes for TLBLTCR- in the Prog_like cluster (red arrowhead, Figure 4G), indicating a 367 TCR negative, progenitor-like MSI2 high compartment also exist in pTLBL. This is confirmed 368 by comparing MSI2 expression across all compartments directly (Figure 4H). 369 Given that the TLBLTCR- compartment was also present in progenitor-like TLBL subclones 370 among samples with functional TCR or preTCR at transformation (pTLBL_02 and pTLBL_04, 371 Figure 4D), we speculated that TLBLTCR- compartment, like its BLBCR- counterpart in pBL, is 372 responsible of progenitor-like program uptake and is essential to pTLBL progression regardless 373 of TCR status at the moment of transformation. To this end w e compared TLBLTCR- 374 transcriptome to the other two compartments from the TLBL_niche defined in Figure 4F . We 375 detected enrichment of a bone marrow progenitor (BMP)-like gene set (Figure 4I) that has been 376 reported to promote R/R in pediatric T cell acute lymphoblastic leukemia (43), a closely related 377 hematologic malignancy, suggesting that the TLBLTCR- compartment is related to R/R . 378 Moreover, MSI2 was among the BMP -like gene set and ranked 132 nd in the full TLBLTCR- 379 transcriptome, leading us to hypothesize MSI2 plays an important role in the progenitor-like 380 BLBCR- and TLBLTCR- lymphoma cells. To gain functional insight of MSI2, we categorised all 381 hematologic cancer cell lines on the Dependency Map (44) by Expasy Cellosaurus (45) into 382 either precursor B cell (n=19), precursor T cell (n=14), or other lymphoid malignancies (n=145) 383 and found malignancies with precursor status had significantly higher expression and 384 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 16 dependency on MSI2 (Figure 4 J). Lastly, we utilised the survival analyses data base R2 385 (https://r2.amc.nl/) and detected that MSI2 expression significantly affected patient survival in 386 two types of T cell lymphoma (46, 47) (Figure 4K). To summarise, we identified a progenitor-387 like TLBLTCR- compartment in pTLBL that was associated with R/R and shared high MSI2 388 expression with BLBCR-. 389 390 Progenitor-like lymphoma cells are potentially chemoresistant. 391 To further explore how the progenitor -like subclones are potentially related to R/R, we 392 integrated lymphoma single cells from either pBL or pTLBL samples (Figure 5A,C ) and 393 examined expression of (R)-CHOP based therapy related genes, namely, MS4A1 targeted by 394 rituximab (R) (48), CYP3A5 that activates cyclophosphamide (C) (49), TOP2A as the main 395 target of doxorubicin (H) (50), TUBB targeted by vincristine (O) (51), NR3C1 the receptor for 396 prednisone (P) (52), and ABCC1, also known as multiple-drug resistant protein 1 (MDP1), that 397 mediate efflux of doxorubicin, vincristine and other chemotherapy agents (50, 53) (Figure 5B, 398 D). Upon integration, BLBCR- cells were transcriptomically distinct from BLBCR+ (Figure 5A), 399 expressing slightly less R -CHOP targets like MS4A1 and TUBB while expressing higher 400 NR3C1, suggesting the transcriptome change may cause a shift in responsiveness to therapy 401 (Figure 5B). Interestingly, BLBCR- cells were found to express more ABCC1, a protective factor 402 against doxorubicin and vincristine, showing signs of chemoresistance (Figure 5B). Since the 403 TLBLTCR- compartment consists of two clusters in Figure 4F, we separated progenitor-like cells 404 (TLBL_ProgTCR-), defined as TCR negative lymphoma cells from Prog_like cluster in Figure 405 4F, from other TLBLTCR- cells (Figure 5C) and found TLBL_ProgTCR- had very low CHOP 406 target gene expression such as TUBB and NR3C1 while expressing, higher ABCC1, again 407 suggesting chemoresistance ( Figure 5D ). Of note, MS4A1 expression was found 408 complementary to that of TOP2A and TUBB among pBL single cells, covering all areas across 409 BLBCR+ (bottom row, Figure 5B) while no such pattern was found of in pTLBL (bottom row, 410 Figure 5D). This is aligned with current observation that pBL has better prognosis than pTLBL 411 under (R)-CHOP-based regiment (54, 55 ) and prompt ed us to search for genes expressed 412 complementarily to (R)-CHOP targets. We calculated correlations of gene expression between 413 TUBB, the strongest expressed CHOP target, and transcriptome-wide individual genes, 414 revealing top correlated genes such as TUBA1B, a close partner of TUBB, as well as counter-415 correlated genes (left panel, Figure 5E, G). The top 300 genes negatively correlated to TUBB 416 expression were found to enrich different programs across pBL and pTLBL (right panel, Figure 417 5E, G). Specifically, pBL cells with less TUBB expression were enriched in a canonical B cell 418 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 17 program and small GTPase activity while its counterpart in pTLBL were enric hed in cell 419 adhesion and stem cell development, implying different strategies in maintaining 420 chemoresistance. Indeed, expression of the top complementary genes in pBL, LYN and FCRL1, 421 both linked to BCR signalling (56, 57) and in pTLBL, COL23A1 and CDH4 that has been 422 associated with anchorage-independent cell proliferation (58) and ß-catenin pathway (59), 423 respectively, were elevated in the chemoresistant compartments that shared high MSI2 424 expression (Figure 5F, H), reiterating the importance of MSI2 in progenitor-like lymphoma 425 compartments. 426 To validate that MSI2 potentially serves as a target for progenitor-like lymphoma cells, we used 427 an apoptosis assay to characterize how MSI2 inhibitor treatment (60) affected 3 hematological 428 malignancy cell lines, namely Raji, a Burkitt lymphoma cell line (RRID:CVCL_0511), SUP -429 T1, a TLBL cell line (61), and Jurkat which belongs to precursor T cell leukemia 430 (RRID:CVCL_0065). After 48-hours, Jurkat cells exposed to varying concentration of MSI2 431 inhibitor treatment had the lowest viability among all tested cell lines (Figure 5I), highlighting 432 that MSI2 is important for cells with precursor status. While SUP-T1 cells were not affected by 433 MSI2 inhibition in vitro, Raji cells had increased apoptotic cells upon MSI2 inhibitor treatment 434 (Figure 5I). We reasoned that the progenitor-like lymphoma compartment eventually develops 435 in vivo and plays important role in disease progression. To test the efficacy in vivo, we injected 436 Raji lymphoma cell line subcutaneously in NSG mice and administered MSI2 inhibitor 437 intratumorally and monitored tumor growth over time (Figure 5 J). We observed that Raji 438 lymphoma grew significantly slower after MSI2 inhibition (Figure 5K) and was associated with 439 longer survival time (Figure 5L), suggesting benefit of targeting the MSI2-high compartment 440 in BL . Collectively, we discovered progenitor-like compartments in pBL and pTLBL were 441 chemoresistant while shared MSI2 as vulnerability. 442 443 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 18 444 Figure 5. Progenitor-like lymphoma cells are potentially chemoresistance: A. Integration of all cancerous pBL 445 single cells. BCR status shown in color. B. Expression level of genes involved/targeted by R -CHOP regiment 446 (killing) or related to chemoresistance (resist) in cancerous pBL single cells samples. C. Integration of all 447 cancerous pTLBL single cells. TCR status shown in color. D. Expression level of genes involved/targeted by 448 CHOP regiment (killing) or related to chemoresistance (resist) in cancerous pTLBL samples. E. (Left) expression 449 correlation between TUBB and other genes in pBL. Top positive (in blue) and negative (in red) correlated genes 450 were highlighted together with MSI2. (Right) top enriched terms (gene ontology biological process, GOBP; 451 WikiPathway, WikiPath) for top 300 genes negatively correlated with TUBB expression. F. Highlighted gene 452 expression negatively correlated with TUBB. G. Same as E. but for pTLBL. H. Same as in F. but referred to pTLBL 453 cancerous cells. I. Apoptosis assay for Raji, SUP -T1 and Jurkat cell lines upon MSI2 inhibitor treatment. Cells 454 were incubated for 48h at designated concentrations. J. Schematics of NSG xenograft model utilizing Raji cell 455 Figure 5 Burkitt Cells TLBL Cells Expression High Low A B UMAP1UMAP2 C D Correlation to TUBB Exp. abs(T_val) -0.5 0.0 0.5 0 25 50 75 100 Burkitt CellsE 0.00 0.25 0.50-0.25-0.50 0 20 40 60 Correlation to TUBB Exp. abs(T_val) TLBL Cells I LYN FCRL1 MSI2 COL23A1 CDH4 MSI2 G H K NSG D0 Raji 106/mouse S.C. Vehicle/ MSI2i Intra-tumoral. D10 D12 D14 Treatment every 2nd day End point 1CM3 reached L 2.0 1.5 1.0 0.5 0.0 10 12 14 16 18 20 22 24 26 Days Tumor size (CM3) Vehicle MSI2i **** **** ** Effect: ABCC1 ResistEffect: 100 80 60 40 20 0 10 12 14 16 18 20 22 24 26 Days Vehicle MSI2i ** Survival rate (%) BLBCR+ BLBCR- MS4A1 Killing 0 1 2 3 4 BLBCR+BLBCR- CYP3A5 Killing BLBCR+BLBCR- 0 1 2 TUBB Killing BLBCR+BLBCR- 0 1 2 3 4 NR3C1 Killing 0 1 2 3 BLBCR+BLBCR- Expression UMAP1UMAP2 TLBLTCR+ TLBLPreTCR TLBLTCR- TLBL_ProgTCR- Expression CYP3A5 Killing 2 1 0 TLBL TCR+ TLBL PreTCR TLBL TCR- TLBL_Prog TCR- TUBB Killing 2 1 0 3 4 TLBL TCR+ TLBL PreTCR TLBL TCR- TLBL_Prog TCR- NR3C1 Killing 2 1 0 3 4 TLBL TCR+ TLBL PreTCR TLBL TCR- TLBL_Prog TCR- TLBL TCR+ TLBL PreTCR TLBL TCR- TLBL_Prog TCR- 0 1 2 3 4 TLBL TCR+ TLBL PreTCR TLBL TCR- TLBL_Prog TCR- 0 2 4 6 0 1 2 3 BLBCR+BLBCR- 4 0 1 2 3 4 BLBCR+BLBCR- BLBCR+BLBCR- 0 1 2 3 4 5 ExpressionExpression TOP2A Killing BLBCR+BLBCR- 0 1 2 3 4 ABCC1 Resist BLBCR+BLBCR- 0 1 2 3 TLBL TCR+ TLBL PreTCR TLBL TCR- TLBL_Prog TCR- 2 1 0 3 4 TOP2A Killing TLBL TCR+ TLBL PreTCR TLBL TCR- TLBL_Prog TCR- 2 1 0 3 4 Expression High Low Expression High Low Expression High Low MSI2 Inhib. 0 5µM 10µM 20µM Percentage Live Early Late Dead Apoptosis assay (7AAD/Annexin V) 80 60 40 30 0 100 80 60 15 10 0 100 80 60 20 10 0 100 Raji Burkitt lymphoma SUP-T1 T cell lymphoma Jurkat Precursor T ALL 15 5 5 20 10 F p.adj GeneRatio 0.05 0.10 0.001 0.002 0.003 0.004 B cell activation GOBP (279) WikiPath (156) Regulation of small GTPase signal Small GTPase signal BCR pathway BCR signaling EGFR signling Type I IFN signalling B cell activation by SARS CoV2 Negative Corr. 0.04 0.07 GeneRatio 0.0025 0.0050 0.0075 0.0100 p.adj GOBP (249) WikiPath (142) Rhythmic process Cell-matrix adhesion Multicellular organism growth Focal adhesion assembly Adenoid cystic carcinoma HSC regulation by GABP Mesodermal commitment HSC differentiation Negative Corr. J TLBL TCR+ TLBL PreTCR TLBL TCR- TLBL_Prog TCR- 0 1 2 3 4 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 19 line. MSI2 inhibitor was administered intratumorally every 2 days starting from day 10. Experiment end point is 456 represented by tumour reaching 1cm3. K. Tumor volume evaluated at each time point. Statistical analysis 457 performed utilizing 2 way Anova - Šídák's multiple comparisons test (untill D24). L. Percentage of survival with 458 and without MSI2 inhibitor treatment referred to J., curve comparison analyzed with Logrank (Mantel-Cox) test. 459 460 Intratumoral heterogeneity across pHL tumor microenvironment (TME). 461 To investigate whether progenitor-like program plays a role in pHL, we performed spatial 462 transcriptomics on 4 pHL tissues . Dimension reduction and nearest neighbor clustering 463 identified 15 transcriptomically distinct clusters across all 4 samples. We then applied cell type 464 specific module scores, namely, Hodgkin (TNFRSF8, TXN, CCL17) (62-64), histiocyte (C1QA, 465 LYZ, S100A8 ) (65-67), M2 -like macrophage ( CCL18, MMP9 ) (68, 69 ), and B cell ( CD19, 466 MS4A1) (70, 71) to each cluster. This revealed that cluster 3 and 9 were dominated by B cells, 467 whereas several clusters had high histiocyte (macrophage) activity, among which cluster 11 468 seemed to adopt a M2 macrophage-like program (M2-Mq). Hodgkin cell activity was almost 469 exclusively found in cluster 10, pointing to similar pHL niche composition throughout samples 470 (Figure 6A). Regarding heterogeneity of the samples, we observed pHL_02 was voided of 7 471 out of 15 clusters from the integration and without any cluster unique to it, strongly suggesting 472 lack of intra-tumor heterogeneity and low quality. This is confirmed by quality parameter and 473 expert knowledge ( Figure S8A). By investigating spatial arrangement of each integrated 474 cluster, we found cluster 3 (B cell) and 11 (M2 -Mq) consistently occupied opposite side of 475 cluster 10 (Hodgkin), forming a 3-layered structure that can be observed across all high-quality 476 samples (left panel Figure 6B, Figure S8B). This indicated their distinct but universal roles in 477 lymphomagenesis. To validate cluster 10 as the Hodgkin niche, we calculated module score 478 using gene panels curated by patholog ist (A.K.) for HRS cell identification and TME 479 characterization, namely, Hodgkin inclusive: TNFRSF8 (CD30), FUT4 (CD15), IRF4; Hodgkin 480 exclusive: MS4A1 (CD20), CD79A, CD19, SLC22A2, POU2AF1 ; and T cell exhaustion: 481 CTLA4, LAG3. We observed that the Hodgkin inclusive score not only spatially complemented 482 Hodgkin exclusive score, but also co-localized with the T cell exhaustion score, agreeing with 483 current knowledge on exhausted T cell infiltration in the Hodgkin niche (72) but avoiding other 484 parts of the TME (right panel Figure 6B). Representative genes for M2-Mq (MMP9), Hodgkin 485 cells ( TXN), and B cells ( MS4A1) validated the spatial arrangement observed in Figure 6B 486 (Figure 6C). 487 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 20 488 Figure 6. Spatial transcriptomics reveals pHL tumor microenvironment comparments and checkpoint blockade 489 targets: A. Characterisation of Hodgkin samples by spatial transcriptomics. (Left) Integration, dimension reduction 490 and unsupervised clustering of Visium spatial dots. (Middle) module scores distribution across clusters. (violin 491 plot from left: Hodgkin, Histiocyte, M2 macrophage-like, B cell). (Right) cell distribution across each sample for 492 different clusters visualised as heat map B. Representative Visium slide highlighting locations of clusters of interest 493 (left, cluster3, 10 and 11) and contour plot of module scores calculated by pathologist curated gene sets. Hodgkin 494 inclusive (red contour): TNFRSF8 (CD30), FUT4 (CD15), IRF4; Hodgkin exclusive (green contour): MS4A1 495 (CD20), CD79A, CD19, SLC22A2, POU2AF1; Exhaustion (blue contour): CTLA4, LAG3, TIGIT. C. Selected 496 spatial gene expression of Hodgkin sample. D. Differently expressed gene (DEG) focused on extracellular 497 signalling. (Top row) ligand (in red) or receptor (in blue) transcript upregulated in cluster of interest (from left: 498 cluster 3, 10 & 11). (Bottom row) CellChat visualisation of selected top ligand -receptor pair from above. E. 499 Selected gene expression of Visium clusters categorized into surface expression (cluster 3 only) , progenitor B, 500 Hodgkin markers, selected T cell immune check point expression and metalloproteins. 501 502 To unveil the underlying cause of such TME arrangement, we identified differentially expressed 503 genes (DEGs) for each cluster focusing on extracellular signaling ligand -receptor pairs. We 504 found CR2 (CD21) among the top upregulated receptors by cluster 3 (B cell) with predicted 505 autocrine-like signaling event within the cluster, suggesting active usage of C3-CR2 signaling. 506 Similar approach revealed activity of CTLA4 and LILRB4, a previously reported immune 507 checkpoint on myeloid cells (73), signaling present in cluster 10 (Hodgkin) and 11 (M2 -Mq), 508 respectively (Figure 6D). These findings suggest that more than one immune checkpoint is at 509 play in the pHL TME and may be targeted separately. Indeed, we found expression of 510 completely different sets of immune checkpoints by cluster 10 (Hodgkin) and 11 (M2-Mq) with 511 A B C D E 10X VisiumpHL_01_02_03_04 Unsupervised Clustering Visium Cluster Score UMAP1UMAP2 1 2 3 4 5 6 7 8 9 10 0 11 12 13 14 pHL_01 pHL_02 pHL_03 pHL_04 0 200 400 0.0 1.0 Prop. Count Visium Cluster 10 11 Score Contour Hodgkin Inclusive Hodgkin Exclusive Exhausion pHL_04 3 TXN 4 8 12 16 2.5 5.0 7.510.0 MS4A1 12.510 20 30 40 MMP9 Cluster 3 Cluster 10 Cluster 11 -log(p.adj) 100 50 0 0.0 0.3 0.6 0.9 1.2 logFC 200 150 100 50 0 0.0 0.5 1.0 1.5 2.0 logFC 200 150 100 50 0 0.00 0.25 0.50 0.75 logFC LAIR1-LILRB4CD86-CTLA4C3-CR2 Receptor Ligand Visium Cluster 1 2 3 4 5 6 7 8 9 10 0 11 12 13 14 DEG Clst. 3 Surface Hodgkin Mkr Immune Checkpoint Metal Visium Cluster Prog. B LTB CR2 CD22 CD52 BANK1 AFF3 TXN BATF3 PPA1 LMNA CTLA4 PDCD1 CD274 LAG3 TIGIT VSIR LILRB4 MMP9 MT1G MT2A 0 1 2 3 5 4 6 7 8 9 10 11 12 13 14 Pct. Exp. Avg. Exp. 25 50 75 100 2.5 2.0 1.5 1.0 0.5 0.0 -0.5 3 10 11 B_cellHdk.0 150 0 75 0 30 Hist. M20 400 RALGPS2 PRKCB 0.5 Figure 6 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 21 the former specifically expressing CTLA4 and the latter with high expression of LAG3 and 512 LILRB4 (Figure 6E). Immunofluorescence microscopy showed LAG3 expressing cells with or 513 without co-expressing CD3 in Hodgkin TME , while PD1 exclusively found in CD3 positive 514 cells but not LMNA, validating the marker gene distribution ( Figure S 8C-E). We detected 515 several genes related to metal metabolism (MMP9, MT1G, MT2A) highly expressed by cluster 516 11 (M2-Mq), providing insights of its interaction with extracellular matrix. Among top marker 517 genes of cluster 10 (Hodgkin), we found PP A1 (Figure 6E ), a gene encoding inorganic 518 pyrophosphatase that recently emerged as a novel target for several types of cancer (74-76), 519 alongside canonical Hodgkin markers like TXN and BATF3. As for cluster 3 (B cell), LTB was 520 upregulated, indicating high activity in de novo genesis of lymphoid structure in this region . 521 Interestingly, we also found top progenitor-like B cell marker genes specifically expressed by 522 cluster 3 (B cell) but not by other clusters, suggesting the presence of progenitor-like B cells 523 outside Hodgkin niche (Figure 6E). In summary, spatial transcriptomics reveal ed a 524 characteristic sandwich-like pHL TME consisting of progenitor -like B cell, exhausted T cell 525 infiltrated Hodgkin niche, and M2-Mq layers, each with distinct biological activity that could 526 be targeted in different ways opening up for n ext-generation combination approaches for 527 immune checkpoint therapy (77). 528 529 Relapsed Hodgkin cells are myeloid-like and maintain PPA1 expression 530 Current understanding of HL development starts with transformation of pre-apoptotic GC B 531 cells into proliferative, mononuclear Hodgkin cells , which in turn give rise to multinuclear, 532 TME-shaping Hodgkin Reed Sternberg (HRS) cells (78). As droplet-based scRNA-seq is not 533 suitable to capture multinuclear HRS cells (79), we reason ed that scRNA-seq preferentially 534 captures Hodgkin cells and provides a valuable opportunity to characterize early events in 535 Hodgkin development. Thus, we integrated 4 patient-match pHL samples from Figure 6 536 (pHL_01~04, all stage II disease ), with a termina l-staged sample acquired upon relapse 537 (pHL_05, stage IV disease ) with all pReLy samples, revealing a cluster unique to the pHL 538 sample (between two main B cell clusters, Figure 7A), partially overlapping with most stages 539 of B cells and, surprisingly, myeloid cells, indicating certain degree of lineage promiscuity . 540 scVDJ-seq revealed lack of immune receptor expression (Figure 7B), ruling out the possibility 541 of normal B nor T cells . We next subclustered pHL samples with inferCNV and combined 542 immune receptor expression, Hodgkin score and CNV score to annotate each subcluster as 543 either Support or Hodgkin ( Figure 7C-D, S 9A-E). As gene expression of HRS cells can 544 resemble histiocytes around them (66), the Hodgkin score for the lymphoma subcluster 545 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 22 annotation was designed to exclude overlapped transcriptome from macrophage/histiocyte: 546 firstly, we calculated module score of genes documented to be expressed by HRS cells, namely 547 TXN and CCL17 (63, 80, 81), followed by subtraction with another module score based on gene 548 known to be exclusively expressed by macrophages: C1QA, LYZ, S100A8 (65, 82, 83). Of note, 549 TNFRSF8 (CD30) expression, the most used gene to identify HRS cells (84), was low across 550 all samples and was not included in the Hodgkin module score. In this way, w e identified 3 551 Hodgkin subclones made up of the pHL unique cluster seen in Figure 7A, among which 2 were 552 very small in cell number (~0.5%) and harbored CNV classically seen in Hodgkin lymphoma 553 like 2p, 12q gain and 13q del (Figure 7C,E), aligning with current understanding of Hodgkin 554 cells (85). Unexpectedly, pHL_05 was dominated by subclusters with extensive CNV but in 555 general low Hodgkin score, suggesting this sample exhibited a very different transcriptome. 556 Still, most of the common CNV reported in Hodgkin lymphoma such as 9p, 12q, and 17p gain 557 and 13q del was detected in the cluster containing Hodgkin cells ( Figure 7C,E) (85). 558 Interestingly, progenitor-like program was only enriched by non-malignant CNV subclones 559 (Figure 7D), again exhibiting lower proportion of cells expressing immune receptor s. This 560 agrees with our spatial transcriptomics findings on progenitor-like B cell infiltration. 561 Despite that TNFRSF8 (CD30) transcript level was low among our samples, its distribution was 562 found to be confined in the pHL unique cluster and exhaustion T cells for Hodgkin and support 563 cells, respectively, suggesting our annotation strategy correctly separated the two compartments 564 (Figure 7F). We detected CD14 and CD4 that signify monocytes and T helper s in Hodgkin 565 cells in the pHL unique cluster (Figure 7F), suggesting lineage promiscuity that has been 566 described in Hodgkin lymphoma (78, 86). CD19 transcript was not detected among Hodgkin 567 cells (top right Figure 7F). To characterize the HL TME components, we manually curated 568 three clusters in proximity to the pHL unique cluster: Hodgkin_unique (Hdkn_Unq), monocyte-569 like (Mono_like) and proliferating (Prolif.). Hdkn_Unq cluster consisted of almost exclusively 570 single cells from pHL sample (Figure 7G), underlining the unique transcriptome landscape in 571 pHL TME and the vague transcriptomic boundary between Hodgkin and support cells. Zooming 572 into each curated cluster, we observed: in Mono_like cluster, cells from pHL exhibited distinct 573 expression including complement activities mediated by C1QA/B/C, oxidative stress related 574 protein MT2A, SOD2 & LDHA, and interferon-stimulated transcription IFITM2/3 (Figure 7H), 575 all linked to tumor-associated macrophage (TAM) status (87-91); as for Hodgkin cells in the 576 other two regions, TXN was found as the top marker gene consistently ( Figure 7H). 577 Interestingly, proinflammatory cytokine MIF (78) has been found in all three curated clusters 578 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 23 by Hodgkin cells or support TAMs (black arrowheads, Figure 7H), indicating MIF takes the 579 central role shaping pHL TME. 580 581 Figure 7. Relapsed Hodgkin cells are myeloid-like and maintain PPA1 expression: A. Integration of pReLy (in 582 color) with pHL samples (in grey). B. BCR/TCR clonotype of pHL single cells. C. (Left) inferCNV subclustering 583 Figure 7 CD45.high.DZGCB CD45.low.DZGCB Cytotoxic.T Double.neg.T LZGCB Mono/mDC Naive.B Naive.CD4.T Naive.CD8.T NK Non.switch.memB pDC Plasma Progenitor.B Progenitor.T Proliferation.T Switched.memB T.helpers T.exhaust Tfh Treg Cell Type Hodgkin Sample pReLy pHLNaiveB+MemB GCB+Plasma T UMAP1 UMAP2 BCR_Hyperexpanded (0.1 < X <= 1) BCR_Large (0.01 < X <= 0.1) BCR_Medium (0.001 < X <= 0.01) BCR_Small (1e-04 < X <= 0.001) no_VDJ TCR_Hyperexpanded (0.1 < X <= 1) TCR_Large (0.01 < X <= 0.1) TCR_Medium (0.001 < X <= 0.01) TCR_Small (1e-04 < X <= 0.001) Clonotype pHL_01pHL_02pHL_03pHL_04pHL_05 Support Hodgkin S1 50.2% S2 25.5% S3 10.1% S5 2.7% S70.6% S1 46.8% S2 20.4% S3 2.8% S4 17.7% S5 11.8% S1 57.0% S2 24.9% S5 5.4% S1 79.33% S2 12.3% S3 8.4% S7 S6 Chr1 Chr2 Chr3 Chr4 Chr5 Chr6 Chr7 Chr8 Chr9 Chr10 Chr11 Chr12 Chr13 Chr14 Chr15 Chr16 Chr17 Chr18 Chr19 Chr20 Chr21Chr22 Genomic Region 0 1 2 3 4 5 6 Modified Expression 2nd 1st CNVref Single Cell Hodgkin Support Subclone CNVref Subclone S7 pHL_01 pHL_02 pHL_03 pHL_04 pHL_05 2p amp 13q del 21 amp S6 2p amp 12q amp 22 amp NA S1 9p amp 12q amp 13q del NA pHL-02 pHL-03 pHL-04 pHL-05 Clonotype Proportion 0.0 0.5 1.0 Subclone Hodgkin Score 0 2 4 S1 S2 S3 S4 S5 S6 S7 S1 S2 S3 S4 S5 S6 S1 S2 S3 S4 S5 S1 S2 S3 S1 S2 S3S1 CNV Score 1.0 0.5 0.0 Prog.B Prog.TEnrich 1st 2nd CNVrefEnrichment Gene ratio CNVref Subclone 0.50 0.75 p.adj 0.05 0.10 0.15 0.20 Sx Sx Hodgkin Support BCR_Hyperexpanded (0.1 < X <= 1) BCR_Large (0.01 < X <= 0.1) BCR_Medium (0.001 < X <= 0.01) BCR_Small (1e-04 < X <= 0.001) no_VDJ TCR_Hyperexpanded (0.1 < X <= 1) TCR_Large (0.01 < X <= 0.1) TCR_Medium (0.001 < X <= 0.01) TCR_Small (1e-04 < X <= 0.001) Clonotype pHL-01 A B C D E H F SupportHodgkin CD30 CD19 CD14 CD4 G Others Mono_like Prolif. Hdkn_Unq Support ReactiveHodgkin Mono_likeProlif.Hdk_unq GSTP1 RPS4Y1 FOS CSTA CAPG IL18 MALAT1 ATOX1 RASSF2 STK17B CORO1A RPS23 APOC1 ZFAS1 HMGN3 MT2A C1QB C1QA C1QC SOD2 MT-ATP8 MIF IFITM3 IFITM2 LDHA LAP3 GLUL CTSC CTSB STAT1 Support Reactive TXN S100A11 LMNA LGALS1 VIM CCDC50 MIF BATF3 ANXA2 S100A6 BACH2 RGS13 MARCKSL1 LRMP MEF2B MS4A1 CD79A CD79B LPP RPS4Y1 IL32 SRGN RNF213 B2M IFITM1 HSP90B1 PIM2 MT2A JUNB FOS Support Reactive Hodgkin TXN MIF LGALS1 VIM ENO1 CCDC50 ATP5MC3 SNHG29 S100A10 CRIP1 NME2 RPL37A FOS TCL1A BANK1 CD79B AFF3 BACH2 MALAT1 RALGPS2 CD37 CD69 C1QA MT2A FTL B2M FTH1 SH3BGRL3 MTRNR2L12 MS4A1 CD79A Support Reactive Hodgkin 1.0 0.5 0.0 -0.5 -1.0 Avg. Exp. 25 50 75 Pct. Exp. 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 0 20 40 60 80 100 120 140 160 180 200 220 240 260 p=0.178 high (n=24) low (n=5) PPA1 expression Overall Survival ProbabilityFollow-up in Months Hodgkin Lymphoma (n=29) Vogel et al. 2014 Avg. Exp.SharedTop Individual pHL-01 pHL-02 pHL-05 CCL17 CCL22 PARK7 TMEM120A S100A10 MIF MT2A TXN BATF3 LMNA PPA1 Pct. Exp. 25 50 75 100 0 1 2 3 4 I J K L TfhT.exh T.reg Breast Cancer G1 Embryonic Stem Cell Pediatric Cancer OXPHOS Mitochondrial Gene Cervical Cancer MYC Targets IR Response EGFR Inhibition Dn Undiff. Cancer 2.40 2.45 2.50 2.55 2.60 NES MSigDB C2 Count p.adj <4.46e-09 100 150 Hodgkin vs Nonmalignant Hodgkin vs Nonmalignant Log2FC -log(p_adj) 90 60 30 0 -2.5 0.0 2.5 5.0 Treatment Response Responder non-Responder Antioxidant Cytoskeleton Extracellular sigaling Glycolysis Intracellular signaling Ribosome Gene Function Treatment pHL Th Cytotoxic.T Naive.CD4 Naive.CD8 NK pDC ProgT ProgB Naive.B MemB Mono/mDCLZGCB LZGCB Plasma n=20135 n=6236 n=2547 n=18046 n=10473 S4 6.4% S6 4.5% S6 0.5% S3 7.2% S4 5.4% S1 24.1%\26.4% S2 44.3% S3 5.2% (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 24

of pHL using the 1st or 2nd CNVref as reference. (Right) UMAP distribution of CNV subclones eventually 584 annotated as “Support” (in light green) or “Hodgkin” (in royal blue). D. Key characteristics of CNV sublcones 585 from C. (From top) BCR/TCR clonotype proportion, Hodgkin score, CNV score (dotted line indicates 0.35), over 586 representation analysis using progenitor B/T marker gene sets, CNVref used, and eventual TLBL/Support subclone 587 annotation. E. Phylogenetic tree of pHL subclones based on CNV accumulation F. Contour plot of Hodgkin or 588 Support single cells with selected gene expression. G. Three m anually curated cell clusters (Monocyte like, 589 Proliferating & Hodgkin unique) and their composition (pie chart). H. Top marker genes of different cell types 590 across different cell clusters defined in G. I. Gene set enrichment analysis on ranked transcriptome comparing 591 Hodgkin against non-malignant cells (Reactive + Support) using MSigDB C2 database. J. Differently expressed 592 genes (DEG) of Hodgkin comparing to non-malignant cells. Selected genes are highlighted in colors representing 593 different function. K. Survival curve of Hodgkin lymphoma patient stratified by PP A1 expression. L. Top 594 individual (left) and shared (right) marker genes of Hodgkin cells from each patient. 595 596 Since the pReLy-defined progenitor-like program was not utilized by lymphoma subclones in 597 pHL, it is likely Hodgkin cells adopt a different precursor program to aid lymphomagenesis. To 598 test this hypothesis, we performed GSEA on ranked gene s comparing Hodgkin and non -599 malignant compartment in a cell -number balanced fashion . Among top enrich ment were 600 epithelial cancer-related (like breast and cervical cancers), metabolic and embryonic stem cell 601 terms, suggesting Hodgkin lymphoma utilized a general stemness program common to 602 epithelial originated malignancies (Figure 7I). We examined top DEGs from the ranked genes 603 and found PP A1, a metabolic enzyme that catalyzes pyrophosphatase hydroxylation, alongside 604 other well characterized genes in HL (Figure 7J), and PP A1 was clinically associated with poor 605 prognosis (Figure 7K). Furthermore, PP A1 expression was detected in Hodgkin cells from all 606 lymphoma subclones (Figure 7L), meaning PP A1 vulnerability was kept upon R/R . Besides, 607 pHL_05 Hodgkin cells had higher expression of myeloid-like genes like S100A10, MIF and 608 MT2A compared to the non-R/R samples (Figure 7L). Taken together, we identified Hodgkin 609 single cells in pHL that depend on PP A1 expression and were myeloid-like in relapse sample. 610 611 R/R pHL was accompanied by CD74highCCL5+ CD8 T cells 612 Since pHL_05 received rituximab previously (Table 1), no B cell compartment was found in 613 this particular sample (Figure 7D). We therefore focused on the T cell compartment in search 614 of a relapse-specific cell type. We integrated all support T cells from pHL defined by TCR 615 expression in scVDJ-seq and found 12 transcriptomically distinct clusters, among which cluster 616 7 was uniquely abundant in pHL_05 (Figure S10A). Gene expression profile indicated cluster 617 7 as CD8 T cells expressing high cytotoxic factors like NKG7 and GZMA and exhausting 618 markers LAG3 and TIGIT (Figure S10A). As for receptors, we found cluster 7 was void of 619 canonical T cell homing receptors CCR7 and IL7R but exceptionally high in CD74 expression, 620 suggesting that the MIF-CD74 axis was responsible of recruitment of this cluster (Figure 621 S10A). Clonotype analysis revealed cluster 7 harbored several large TCR clones all belonging 622 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 25 to pHL_05 ( Figure S 10B,C), indicating weak selection force during clonal expansion. To 623 further characterize these expanding cells, we compared T cells with large TCR clone against 624 other T cells within pHL_05 and confirmed cluster 7 marker genes like CD8A/B, CCL5 and 625 GZMA were higher in these expanding T cells (Figure S10D), suggesting they were the main 626 effectors of the cluster 7 phenotype . Finally, we examined key cluster 7 gene expression by 627 patient and validated that the phenotype was indeed unique to R/R: we found higher cytotoxic, 628 pro-inflammatory and exhaustion gene expressions in pHL_05, which was rarely found in other 629 samples (NK-like, Granzymes, Cytokine and Exhaustion Figure S10E), suggesting a unique T 630 cell cluster in the R/R TME. We confirmed low CCR7 and high CD74 expression in T cells 631 through sample-wide comparison (Receptor, Figure S10E), pointing to a major shift in T cell 632 recruitment mechanism. In short, we identified a pro-inflammatory T cell subset unique to the 633 R/R TME in pHL and recruited by the MIF-CD74 axis. 634 635

636 637 The main challenge in pediatric lymphoma treatment is balancing cure rates against the risk of 638 R/R and the burden of treatment -related toxicity, where R/R is often due to resistance to 639 treatment (92, 93). In this study, we identified progenitor-like lymphocytes as a unique trait in 640 children, whose progr am was enriched in pNHL subclones and likely contribute d to R/R 641 especially when becoming an initiating clone for malignant transformation . In pHL the 642 progenitor-like program was found in non-cancerous cells in the TME and was unrelated to 643 R/R. However, Hodgkin cells in R/R were myeloid-like and recruits CD74highCCL5+ T cells 644 likely via the MIF-CD74 axis. Finally, we proposed MSI2 and PP A1 as novel vulnerability of 645 pNHL and pHL for further exploitation. 646 647 The pediatric -specific, progenitor -like lymphocytes identified in this study are likely to 648 participate in early lymphopoiesis. Indeed, some of the enrichment gene programs like NOTCH 649 and WNT pathways in progenitor-like lymphocytes have been reported in early lymphopoiesis 650 (94), among which NOTCH2 was documented to promote children specific cell type (95). 651 Moreover, a recent study showed that non-cancerous B cells from rituximab treated lymphoma 652 survivor were incapable of maturation (96), suggesting de novo lymphopoiesis is more 653 complicated than just one step leading to another. Therefore, it is likely that the progenitor-like 654 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 26 lymphocytes represent a physiological yet transient compartment that is essential to early 655 lymphopoiesis. 656 657 Our data revealed that the progenitor-like program was utilized by lymphoma subclones. In 658 pBL, these subclones showed BCR silencing, traits of chemoresistance, and were often found 659 as the initiating clone in R/R sample s. This aligns with findings in diffuse large B cell 660 lymphoma (DLBCL) , a closely related GC-derived B cell lymphoma , which became more 661 aggressive upon BCR silencing or extinction (97, 98 ). Such phenomenon has not yet been 662 described in BL , likely because bulk sequencing techniques (99) is not suitable to detect the 663 BCR negative compartment which consist roughly 10~30% of a sample (this study) as 664 compared to up to 65% in DLBCL (97). In pTLBL progenitor-like subclones, TCR silencing 665 was observed in 4 out of 5 patients, and pHL are known to be immune receptor negative (100), 666 suggesting that immune receptor silencing is a common trait of lymphoma progression. These 667 immune receptor silenced cells, especially in pNHL , likely represent a long-overlooked 668 compartment under the assumption lymphoma cells always express immune receptor. It will be 669 important to better define immune receptor negative subclones to identify markers for 670 pathology evaluation and to identify new treatment modalities for R/R subclones (28, 101, 102). 671 672 In pBL and pTLBL, lymphoma cells that underwent immune receptor silencing had higher 673 expression of MSI2, an RNA binding protein that has been reported to maintain 674 hematopoietic/leukemic stem cell s and drive chemoresistance (34, 103, 104 ). Emerging 675 evidence suggests that MSI2 plays an important role in lymphoma as well, and inhibition of 676 MSI2 reduces growth of T acute lymphoblastic leukemia in a xenograft model (105, 106). When 677 searching for genes complementarily expressed to TUBB, MSI2 was among the top, alongside 678 with enrichment of gene ontology terms that we detected in progenitor-like B and T cells ; 679 heightened BCR pathway and stem-like terms, respectively. These lines of evidence place MSI2 680 in the center of lymphomagenesis and chemoresistance. 681 682 Identification of single Hodgkin cells allowed us to compare transcriptome changes of this early 683 compartment upon relapse , including extensive CNV alteration, a switch to a myeloid like 684 phenotype, and high expression of PP A1, an emerging target against several types of cancer 685 (75), including colorectal (74), lung (107), and gastric cancer (108). Moreover, Hodgkin cells 686 were enriched in g enes related to oxidation phosphorylation (OXPHOS) (Figure 7I ), a 687 metabolic process where PP A1 plays a key role . Given that Hodgkin cells are considered a 688 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 27 precursor of Reed-Sternberg cells, this unique metabolic vulnerability may provide future 689 treatment benefit. Additionally, we found CD74highCCL5+ T cell infiltration in relapsed pHL 690 TME that is well known to respond to MIF, an inflammatory cytokine in autoimmune disease 691 like inflammatory bowel disease (109), providing new insight to disrupt relapsed HL. 692 693 In conclusion, we thoroughly characterized several main types of pediatric lymphoma at single-694 cell level. Our data reveals chemoresistant progenitor-like pNHL subclones undergo immune 695 receptor silencing, express MSI2 and, when becoming the initiating clone, is associated with 696 R/R. As for pHL, Hodgkin cells adopt myeloid -like status for relapse and recruit 697 CD74highCCL5+ T cell while maintaining PP A1 as vulnerability. 698 699

700 701 Study design and Patient recruitment 702 The “Pediatric lymphoma ” study ( EPM: 2021-01381) received ethical approval from the 703 Swedish Etikprövningsmyndigheten in 2021 . The recruitment opened in 2021 and recruited 704 patients from Karolinska University Hospital. All children and adolescents (below 18 years of 705 age) with pediatric lymphoma were eligible to enroll at the time of diagnosis upon giving 706 informed consent. Patients with clinically suspected lymphoma who underwent a biopsy or 707 surgery as part of the standard of care were eligible for enrollment into the study which recruited 708 patients from May 2021 to December 2025. The study was performed in accordance with the 709 World Medical Association Declaration of Helsinki. The Code of Ethics of the World Medical 710 Association (Declaration of Helsinki) for human samples was followed . The guardians and 711 patients have provided written consent after receiving oral and written study information. 712 713 Sample collection 714 All samples were obtained as surgical or core needle biopsies from Karolinska University 715 Hospital, Stockholm, Sweden. All lymphoma specimens were sampled at primary diagnosis 716 except for pHL_05 which was samples at first relapse. Histopathological evaluation was 717 performed by specialist hematopathologists to determine the diagno sis for the different 718 lymphoma subtypes according to the current WHO classification system, as well as for reactive 719 lymph nodes. 8 pediatric reactive lymph node (pReLy) specimens of which one was sampled 720 twice (pReLy _01, 01b), 3 pediatric Burkitt lymphoma (pBL) specimens, 5 pediatric T cell 721 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 28 lymphoblastic lymphoma (pTLBL) specimens, 5 pediatric classical Hodgkin lymphoma (pHL) 722 specimens, 1 pediatric anaplastic large cell lymphoma (pALCL), 1 pediatric primary 723 mediastinal large-B-cell lymphoma (pPMBCL), and 2 adult reactive lymph nodes (aLN) were 724 anonymously included in this study and patien t characteristics are summarized in the 725 Supplementary tables S1 and S2. 726 727 Single-cell preparation 728 Samples were mechanically dissociated straight after tissue collection and strained to obtain a 729 single cell suspension . The freshly isolated single-cell suspension used for scRNA-Seq was 730 washed twice with 1x phosphate-buffered saline (PBS) and Dead Removal kit (Miltenyi Biotec 731 #130-090-101) was used to eliminate dead cells . The final concentration of the single cell 732 suspension was adjusted to 900 cells/µL in 1x PBS with 0,04% bovine serum albumin (BSA) 733 following 10X Genomics recommendations . Leftover cell suspensions were cryopreserved as 734 single-cell suspensions in complete RPMI with 10% fetal bovine serum (FBS) plus 10% 735 dimethyl sulfoxide (DMSO) and stored in liquid nitrogen for further analysis. 736 737 Single-cell RNA and VDJ library construction and sequencing 738 We used the Chromium X instrument, Chromium Next GEM Chip K and the Chromium Next 739 GEM Single Cell 5’ Reagents Kits (V2) to prepare individually barcoded single-cell RNA-Seq 740 libraries and VDJ-Seq libraries following the manufacturer’s protocol (10X Genomics). For 741 quality control and to quantify the library concentration, we used the BioAnalyzer High 742 Sensitivity DNA kit (Agilent Technologies) and Qubit High sensitivity kit (ThermoFischer 743 Scientific). Sequencing using dual indexing was conducted on an Illumina NextSeq machine, 744 using the 150-cycle High Output kit. Sample demultiplexing, barcode processing, and single -745 cell 39 gene counting were performed with the Cell Ranger Single Cell Software Suite CR2.0.1. 746 747 Bioinformatic pipeline for single-cell RNAseq 748 For each sequenced scRNAseq library, gene count matrix was generated with 10X CellRanger 749 pipeline (6.1.2) with 2020-A GRCh38 (https://cf.10xgenomics.com/supp/cell-exp/refdata-gex-750 GRCh38-2020-A.tar.gz ) and CR 7.1 ( https://cf.10xgenomics.com/supp/cell-vdj/refdata-751 cellranger-vdj-GRCh38-alts-ensembl-7.1.0.tar.gz ) as gene expression and VDJ references, 752 respectively. All quality control and analytic steps were carried out using Seurat package 4.0.5 753 unless indicated otherwise. 754 755 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 29 Quality control steps were carried out as follows. First, immune receptor variable genes 756 including IGV, IGD, IGJ, TRV, TRD, TRJ as well as all HLA genes were excluded. 757 Mitochondrial gene percentage was acquired with `PercentageFeatureSet()` function. Single 758 cells from reactive lymph node and lymphoma samples with more than 5% and 10% of 759 mitochondrial gene were excluded, respectively. For reactive lymph node, unique (n.feature) 760 and total (n.count) gene counts were used to exclude single cells with either feature outside the 761 range 300 to mean value plus standard deviation (sd) times 3, and 1000 to mean value plus sd 762 times 3, respectively. The same process was applied to lymphoma samples with different 763 ranges. (BL n.feature = 50 ~ mean +10*sd, n.count = 50 ~ mean+10*sd; TLBL & HL n.feature 764 = 150 ~ mean + 10*sd, n.count = 200 + 10*sd). 765 766 High quality single cells passing the abovementioned steps were subjected to standard Seurat 767 workflow. For each sample, gene counts were normalized with `NormalizeData()` function 768 specifying “LogNormalize” method. Top 2000 variable genes were identified using 769 `FindVariableFeatures()` function with “vst” method. Data scaling was done with `ScaleData()` 770 function. Principle components (PCs) were acquired using `RunPCA()` function with top 771 variable genes identified in previous steps, after which `doubletFinder_v3()` from 772 DoubletFinder R package (2.0.3) was used to predict doublet specifying the first 10 PCs. 773 Predicted doublets were removed immediately after this step. 774 775 Data integration was carried out step -wise with sanity checks in between. First, we generate 776 integration by sample diagnoses (namely reactive lymph node, BL, TLBL, HL, PMBCL, 777 ALCL), sequentially utilizing Seurat functions `SelectIntegrationFeatures()`, 778 `FindIntegrationAnchors()` and `IntegrateData()`. Taking batch effect and biological relevance 779 into account, we compared canonical correlation analysis (CCA), reciprocal principal 780 component analysis (RPCA) and harmony as reduction method and found CCA most suitable 781 to our research scope. After that, we further integrated different sample types to facilitate 782 investigation with different scopes. After each integration, we carried out data scaling and PCA 783 on the “integrated” assay of the object and used the first 30 PCs as inputs for `RunUMAP()` 784 function and `FindNeighbors()` function for uniform manifold approximation and projection 785 (UMAP) dimension reduction and Seurat cluster identification. 786 787 Cell type annotation was done on integrated reactive lymph node object only. Briefly, we 788 trained two cell type classifiers with a support vector machine-based algorithm scPred (1.9.2) 789 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 30 with training reference originating from PBMC s (32) and paediatric tonsil s 790 (immunesinglecell.com). Generally, each Seurat cluster was annotated in majority vote-fashion. 791 However, conflict s between two predictors and expert knowledge were resolved manually 792 based on marker gene expression which also enable additional cell type annotation wherever 793 feasible. Information from scVDJ-seq was attached to integrated object using scRepertoire 794 (1.7.0) for exclusion of wrongly annotated cells (i.e. annotated B cells having TCR). The final 795 cell type annotation was used as reference to train a pediatric lymph node specific cell type 796 classifier to aid investigation on integration with lymphoma samples. 797 798 Infer CNV, lymphoma subclone annotation & phylogenetic analysis 799 To detect possible lymphoma cells that lacked hyperexpanded BCR or TCR, we estimated copy 800 number variation (CNV) of each single cell from BL and TLBL samples with inferCNV (1.8.1) 801 against the 1st CNVref described in Figure 1G , setting analysis mode to “subclusters”. 802 Individual subclusters were then tested with their resemblance with 2nd CNVref signatures. 803 Specifically, resemblance was decided by chromosome gain of the first third of the chromosome 804 10 and loss of the whole chromosome 19 . To reveal the true CNV under the 2nd CNVref 805 signature, subclusters identified resemble d 2nd CNVref then underwent a second round 806 inferCNV using 2nd CNVref as reference. 807 808 CNV subclusters were annotated as either lymphoma or support subclones based on the 809 following steps: for pBL, each subcluster is assigned a CNV score defined as standard deviation 810 of “modified expression” across its full transcriptome. Subclusters with CNV score above cut-811 off (empirically set to 0.31) are annotated as lymphoma subclones . If a lymphoma subclon e 812 defined in this way has low MYC expression, it is overwritten as support subclone. If a support 813 subclone upon this step had more than 50% of cells possessing hyperexpanded BCR, it is 814 overwritten as lymphoma subclone; if hyperexpanded BCR is less than 50%, cells with 815 hyperexpanded BCR are separated into a new lymphoma subclone . Lastly, cells with TCR 816 expression are discarded from lymphoma subclone. For pTLBL , subclones with CNV score 817 greater than cut-off (set to 0.18) are annotated as lymphoma. Cells with hyperexpanded TCR 818 in support subclone s are separated into a new lymphoma subclone. For pHL, subclones with 819 CNV score and Hodgkin score greater than cut -off (set to 0.35 and 0.7 respectively ) are 820 annotated as lymphoma subclones. 821 822 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 31 Phylogenetic trees of lymphoma subclones from each lymphoma sample was inferred based on 823 the assumption of CNV accumulation. As a validation, we detect transcriptome-wide single 824 nucleotide variation (SNV) at single cell level with SComatic ( https://github.com/cortes-825 ciriano-lab/SComatic) and calculated Jaccard distance of SNV sharing between lymphoma 826 subclones. The Jaccard distance was then used to validate major branching events in 827 phylogenetic trees build by inferCNV. 828 829 Phylogenetic analysis on VDJseq 830 VDJ sequencing data obtained from Cell Ranger pipeline was analyzed using the Immcantation 831 (www.immcantation.org) framework. VDJ genes for each sequence were aligned to the IMGT 832 GENE-DB database using IgBlast v1.22.0 (REF: 833 https://pmc.ncbi.nlm.nih.gov/articles/PMC3692102/). Change -O v1.3.4 (REF: 834 http://pmc.ncbi.nlm.nih.gov/articles/PMC4793929/) was used to identify the predicted 835 germline and group sequences into clonal clusters. Nonproductive BCRs and TCRs were 836 removed and only heavy chain (IGH) and beta chain (TCRB) data were selected for further 837 analyses. Clonotype frequency and CDR3 characterization were performed for all samples 838 (BCRs and TCRs). Clonotype abundances were computed by counting unique clone_id 839 occurrences, and the 20 most frequent clonotypes per sample were selected for downstream 840 visualization. For these dominant clonotypes, we quantified complementarity -determining 841 region 3 (CDR3) lengths from amino-acid junction sequences and generated per-sample CDR3 842 length distributions. To characterize sequence diversity within the most abundant clonotype 843 (Clonotype 01), amino -acid sequence logos were created with ggseqlogo v0.2 using 844 probability-based residue frequency estimation. In addition, BCR lineage trees were generated 845 for the top clonotype. Hamming distances were calculated among sequences of equal length 846 using Biostrings v2.76, and neighbor-joining trees were built using ape v5.8, with the tree rooted 847 on the sequence closest to the germline. Mutation burdens relative to germline and clone-level 848 cell counts were mapped onto the tree using ggtree v3.16.3 , generating sample -specific 849 phylograms. 850 851 Tissue processing for spatial transcriptomics 852 Pediatric Hodgkin lymphoma samples (pHL_01-04) were embedded in Optimal Cutting 853 Temperature compound (OCT, Sakura Tissue -TEK) on dry ice and stored at −80 °C. OCT 854 blocks were cut with a pre -cooled cryostat at 8µm thickness, and sections were transferred to 855 fit the 6.5 mm2 oligo-barcoded capture areas on the Visium 10x Genomics slide. Before 856 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 32 performing the complete protocol, sample quality control was performed according to 857 manufacturer’s instructions: RNA extraction to check RNA quality of the samples ( Quiagen 858 RNeasy Mini kit #74104; samples accepted with RIN > 9) and Visium Spatial Tissue 859 Optimization (10x Genomics) to select optimal permeabilization. The experimental slide with 860 Hodgkin Reed-Stenberg cells and Hodgkin cells was fixed and stained with hematoxylin and 861 eosin (H&E) and the sequence libraries were then processed according to manufacturer’s 862 instructions (10x Genomics, Visium Spatial Transcriptomic). 863 864 Immunohistochemistry 865 Pediatric Hodgkin lymphoma samples (pHL_01-04) were embedded in OCT and stored at −80 866 °C for immunofluorescence staining. OCT blocks were cut with a pre -cooled cryostat at 8µm 867 thickness. Sections were blocked with 5% fetal bovine serum for 1 hour . The sections were 868 stained with CD30 (#BAF1028, R&D systems), CD3 (#300415, Biolegend), PD1 (#AB201825, 869 Abcam), LAG3 (#AB270908, Abcam), LMNA (#AB193903, Abcam), DAPI and mounted with 870 ProLong glass antifade Mountant (#P36984, Invitrogen). Slides were acquired by Zeiss 871 LSM800-Airy using 20X objective. Images were processed with ImageJ software. 872 873 Cell culture 874 Raji, Jurkat and SUP-T1 cell lines were cultured with complete RPMI1640 with 5, 10, 20 μM 875 of MSI2 inhibitor (Ro 08-2750, # HY-108466, MedChemExpress) or DMSO (vehicle) for 48h. 876 Cells were harvested and labelled for an apoptosis assay with 7AAD (#420403, Biolegend) and 877 Annexin V (#A13201, Invitrogen) and analyzed by flow cytometry. Data was obtained by BD 878 LSRFortessa X-20 flow cytometry machine and analyzed with FlowJo software. 879 880 Flow cytometry 881 Pathology unit at Karolinska University Hospital provided flow cytometry data for the markers 882 shown in Supplemental Figure 2: CD19, CD3, CD4, CD8, κ and λ Ig chains. From these flow 883 cytometry data it is possible to extrapolate κ/λ and CD4/CD8 ratios as shown in the 884 quantifications. 885 For flow cytometry data of Figure 3L, pReLy_07 and pBL_03 cells were thawed and dead 886 removal kit (#130-090-101, Miltenyi Biotec) was performed. T hen, cells were labelled with 887 LIVE/DEAD Fixable Aqua Dead Cell Stain Kit (# L34966, Invitrogen), CD11b (#101224, 888 Biolegend), CD3 (#317318, Biolegend). PAX5 (#649710, Biolegend), MSI2 (#MA5-57490, 889 Invitrogen), MYC (#MA1-980-AF555, Invitrogen), Ki67 (#350521, Biolegend) and γH2AX 890 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 33 (#613420, Biolegend) staining were performed after permeabilization with FOXP3 kit 891 (#421403, Biolegend). 892 Data were obtained by BD LSRFortessa X -20 flow cytometry machine and analyzed with 893 FlowJo software. 894 895 NSG mice and xenograft 896 Age- and sex-matched mice were bred and maintained under specific-pathogen-free conditions 897 at the animal facility of the Department of Microbiology, Tumor and Cell Biology, Karolinska 898 Insitutet. All mice experiments performed were approved by the Stockholm North Animal Etics 899 Committee (permit #05294-2023). 900 1x106 Raji cells were mixed with Matrigel (# 356231, Corning) in 1:1 ratio and injected 901 subcutaneously into NSG mouse flank. Starting from day 10, 10 mg/Kg mouse of MSI2 902 inhibitor (Ro 08 -2750; #HY-108466, M edChemExpress) or DMSO control were injected 903 intratumorally every two days. Tumor growth was assessed by digital caliper every 2 days until 904 they reached 1 cm3. Statistics were performed using GraphPad Prism version 10. Values were 905 considered statistically significant whether the probability (P) values were equal or below 0,05 906 (*), 0,01 (**), 0,001 (***) or 0,0001 (****). 907 908 Data and code availability 909 Codes to reproduce this study are stored on Github (https://github.com/Westerberg-Lab). 910 Annotated single cell objects are stored on European Genome-Phenome Archive (https://ega-911 archive.org/) as Seurat object with pending accession number. 912 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 34

913 We are grateful to all patients and their families at the Pediatric Oncology Unit at Karolinska 914 University Hospital for supporting our study. We thank Peri Noori, head of the C enter for 915 Molecular Medicine single cell platform at Karolinska Institutet. We thank Gabriela Prochazka, 916 Elisa Basmaci, and Johanna Sandgren at the Childhood Tumor Bank, Karolinska University 917 Hospital and Karolinska Institutet. The data handling was enabled by resources provided by the 918 National Academic Infrastructure for Supercomputing in Sweden (NAISS), partially funded by 919 the Swedish Research Council through grant agreement no. 2022-06725. 920 This work was supported by Karolinska Institutet PhD (KID) fellowships to T.Y ., R.D., a Cancer 921 Society postdoctoral fellowship to J.R., the Childhood Cancer fund and Stockholm region ALF 922 funding to F.B. and L.S.W., the Swedish Research Council, Cancer Society, Worldwide Cancer 923 Research, Radiumhemmet Research Funds, and Karolinska Institutet to L.S.W. L.S.W. is a 924 Ragnar Söderberg fellow in Medicine and holds a senior research position awarded by the 925 Childhood Cancer fund. 926 927 Author Contribution 928 F.B., L.S.W. conceptualized the study, T.Y ., R.D., F.B., L.S.W. designed the research, T.Y ., R.D., 929 G.M.M., M.R.L.B., M.H., C.O., J.R. performed the experiments , R.D., T.Y ., G.M.M., 930 M.R.L.B., M.H., C.O. and L.S.W. analysed the data, A.E., F.B. recruited and maintained clinical 931 records of the patients, A.K. performed pathology diagnostics and evaluation , T.Y . R.D., 932 G.M.M. and L.S.W. wrote the manuscript, and all authors edited the manuscript. 933 934 Conflict-of-interest disclosure: The authors declare no competing financial interests. 935 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 35

936 937 1. J. M. Bowen et al., Lymphoma diagnosis at an academic centre: rate of revision and 938 impact on patient care. Br J Haematol 166, 202-208 (2014). 939 2. I. N. Sheikh et al. , Management of Aggressive Non -Hodgkin Lymphomas in the 940 Pediatric, Adolescent, and Young Adult Population: An Adult vs. Pediatric Perspective. 941 Cancers 14, (2022). 942 3. R. Alaggio et al., The 5th edition of the World Health Organization Classification of 943 Haematolymphoid Tumours: Lymphoid Neoplasms. Leukemia 36, 1720-1748 (2022). 944 4. Y . T. Dai et al. , Transcriptome -wide subtyping of pediatric and adult T cell acute 945 lymphoblastic leukemia in an international study of 707 cases. Proc Natl Acad Sci U S 946 A 119, e2120787119 (2022). 947 5. W. Cheng et al. , Epidemiologic trends for pediatric Hodgkin and non -Hodgkin 948 lymphomas: a U.S. population-based study. Ann Hematol 104, 3299-3307 (2025). 949 6. L. Chen, M. Wang, H. Fan, F. Hu, T. Liu, Comparison of pediatric and adult lymphomas 950 involving the mediastinum characterized by distinctive clinicopathological and 951 radiological features. Scientific reports 7, 2577 (2017). 952 7. V . Camus et al., Outcomes after first-line immunochemotherapy for primary mediastinal 953 B-cell lymphoma: a LYSA study. Blood Adv 5, 3862-3872 (2021). 954 8. W. H. Chang et al. , Late effects of cancer in children, teenagers and young adults: 955 Population-based study on the burden of 183 conditions, in -patient and critical care 956 admissions and years of life lost. Lancet Reg Health Eur 12, 100248 (2022). 957 9. J. Stankiewicz, A. Jablonska, P. Treichel, E. Demidowicz, J. Styczynski, Pediatric B -958 cell Non -Hodgkin Lymphoma: The Impact of Therapy Response and Relapse on 959 Outcome. A Single-center Analysis. In Vivo 38, 2812-2819 (2024). 960 10. B. L. Asselin et al., Effectiveness of high -dose methotrexate in T -cell lymphoblastic 961 leukemia and advanced -stage lymphoblastic lymphoma: a randomized study by the 962 Children's Oncology Group (POG 9404). Blood 118, 874-883 (2011). 963 11. A. W. Lambert, D. R. Pattabiraman, R. A. Weinberg, Emerging Biological Principles of 964 Metastasis. Cell 168, 670-691 (2017). 965 12. S. Gkountela et al. , Circulating Tumor Cell Clustering Shapes DNA Methylation to 966 Enable Metastasis Seeding. Cell 176, 98-112 e114 (2019). 967 13. N. J. Birkbak, N. McGranahan, Cancer Genome Evolutionary Trajectories in 968 Metastasis. Cancer Cell 37, 8-19 (2020). 969 14. T. Reya, S. J. Morrison, M. F. Clarke, I. L. Weissman, Stem cells, cancer, and cancer 970 stem cells. Nature 414, 105-111 (2001). 971 15. K. Biserova, A. Jakovlevs, R. Uljanovs, I. Strumfa, Cancer Stem Cells: Significance in 972 Origin, Pathogenesis and Treatment of Glioblastoma. Cells 10, (2021). 973 16. K. Basso, R. Dalla -Favera, Germinal centres and B cell lymphomagenesis. Nat Rev 974 Immunol 15, 172-184 (2015). 975 17. B. Scheijen et al., Next-generation sequencing of immunoglobulin gene rearrangements 976 for clonality assessment: a technical feasibility study by EuroClonality-NGS. Leukemia 977 33, 2227-2240 (2019). 978 18. A. W. Langerak et al. , EuroClonality/BIOMED -2 guidelines for interpretation and 979 reporting of Ig/TCR clonality testing in suspected lymphoproliferations. Leukemia 26, 980 2159-2171 (2012). 981 19. C. A. C. Peretz et al., Multiomic single cell sequencing identifies stemlike nature of 982 mixed phenotype acute leukemia. Nature communications 15, 8191 (2024). 983 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 36 20. F. Liu et al. , Unraveling the enigma of B cells in diffuse large B -cell lymphoma: 984 unveiling cancer stem cell-like B cell subpopulation at single-cell resolution. Frontiers 985 in immunology 14, 1310292 (2023). 986 21. C. I. E. Smith, R. Zain, Reduced clone size upon BTK inhibitor resistance mutations 987 relates to toxicity caused by inherited PLCG2 gain -of-function variations. Eur J 988 Haematol 113, 130-131 (2024). 989 22. F. Bunz, Passengers, drivers, and "goners". International journal of cancer. Journal 990 international du cancer 155, 1696-1698 (2024). 991 23. C. Dominguez Conde et al., Cross-tissue immune cell analysis reveals tissue -specific 992 features in humans. Science 376, eabl5197 (2022). 993 24. Y . Hao et al., Integrated analysis of multimodal single -cell data. Cell 184, 3573-3587 994 e3529 (2021). 995 25. M. Andreatta, J. Garnica, S. J. Carmona, Identification of malignant cells in single-cell 996 transcriptomics data. Commun Biol 8, 1264 (2025). 997 26. N. Andor et al., Single-cell RNA-Seq of follicular lymphoma reveals malignant B -cell 998 types and coexpression of T-cell immune checkpoints. Blood 133, 1119-1129 (2019). 999 27. M. M. Chen et al., Integrated single -cell and spatial transcriptomics uncover distinct 1000 cellular subtypes involved in neural invasion in pancreatic cancer. Cancer Cell 43, 1001 1656-1676 e1610 (2025). 1002 28. C. Corinaldesi et al., Single-cell transcriptomics of pediatric Burkitt lymphoma reveals 1003 intra-tumor heterogeneity and markers of therapy resistance. Leukemia 39, 189 -198 1004 (2025). 1005 29. K. Okonechnikov et al., Oncogene aberrations drive medulloblastoma progression, not 1006 initiation. Nature 642, 1062-1072 (2025). 1007 30. H. J. Kim et al., Precancerous Cells Initiate Glioblastoma Evolution and Contribute to 1008 Intratumoral Heterogeneity. Cancer Discov 15, 1377-1391 (2025). 1009 31. F. Muyas et al., De novo detection of somatic mutations in high-throughput single-cell 1010 profiling data sets. Nat Biotechnol 42, 758-767 (2024). 1011 32. G. Monaco et al. , RNA -Seq Signatures Normalized by mRNA Abundance Allow 1012 Absolute Deconvolution of Human Immune Cell Types. Cell reports 26, 1627 -1640 1013 e1627 (2019). 1014 33. T. Sundell et al., Single-cell RNA sequencing analyses: interference by the genes that 1015 encode the B-cell and T-cell receptors. Brief Funct Genomics 22, 263-273 (2022). 1016 34. M. G. Kharas et al., Musashi-2 regulates normal hematopoiesis and promotes aggressive 1017 myeloid leukemia. Nat Med 16, 903-908 (2010). 1018 35. S. M. Park et al. , Musashi2 sustains the mixed -lineage leukemia -driven stem cell 1019 regulatory program. J Clin Invest 125, 1286-1298 (2015). 1020 36. D. W. Yeh et al., MSI2 promotes translation of multiple IRES-containing oncogenes and 1021 virus to induce self -renewal of tumor initiating stem -like cells. Cell Death Discov 9, 1022 141 (2023). 1023 37. D. Redmond, A. Poran, O. Elemento, Single-cell TCRseq: paired recovery of entire T-1024 cell alpha and beta chain transcripts in T -cell receptors from single -cell RNAseq. 1025 Genome medicine 8, 80 (2016). 1026 38. M. S. Krangel, Mechanics of T cell receptor gene rearrangement. Curr Opin Immunol 1027 21, 133-139 (2009). 1028 39. G. Fu et al., Themis controls thymocyte selection through regulation of T cell antigen 1029 receptor-mediated signaling. Nat Immunol 10, 848-856 (2009). 1030 40. F. Matthijssens et al., RUNX2 regulates leukemic cell metabolism and chemotaxis in 1031 high-risk T cell acute lymphoblastic leukemia. J Clin Invest 131, (2021). 1032 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 37 41. L. F. Lee et al. , Anti -IL-7 receptor -alpha reverses established type 1 diabetes in 1033 nonobese diabetic mice by modulating effector T-cell function. Proc Natl Acad Sci U S 1034 A 109, 12674-12679 (2012). 1035 42. N. D. Anderson et al., Transcriptional signatures associated with persisting CD19 CAR-1036 T cells in children with leukemia. Nat Med 29, 1700-1709 (2023). 1037 43. J. Xu et al., A multiomic atlas identifies a treatment-resistant, bone marrow progenitor-1038 like cell population in T cell acute lymphoblastic leukemia. Nat Cancer 6, 102 -122 1039 (2025). 1040 44. A. Tsherniak et al. , Defining a Cancer Dependency Map. Cell 170, 564 -576 e516 1041 (2017). 1042 45. A. Bairoch, The Cellosaurus, a Cell-Line Knowledge Resource. J Biomol Tech 29, 25-1043 38 (2018). 1044 46. J. Iqbal et al., Gene expression signatures delineate biological and prognostic subgroups 1045 in peripheral T-cell lymphoma. Blood 123, 2915-2923 (2014). 1046 47. F. Liu et al. , PEG10 amplification at 7q21.3 potentiates large -cell transformation in 1047 cutaneous T-cell lymphoma. Blood 139, 554-571 (2022). 1048 48. D. G. Maloney et al., IDEC-C2B8: results of a phase I multiple -dose trial in patients 1049 with relapsed non-Hodgkin's lymphoma. J Clin Oncol 15, 3266-3274 (1997). 1050 49. W. Shu et al., Cytochrome P450 Genetic Variations Can Predict mRNA Expression, 1051 Cyclophosphamide 4 -Hydroxylation, and Treatment Outcomes in Chinese Patients 1052 With Non-Hodgkin's Lymphoma. J Clin Pharmacol 57, 886-898 (2017). 1053 50. C. F. Thorn et al. , Doxorubicin pathways: pharmacodynamics and adverse effects. 1054 Pharmacogenet Genomics 21, 440-446 (2011). 1055 51. J. Skubnik, V . S. Pavlickova, T. Ruml, S. Rimpelova, Vincristine in Combination 1056 Therapy of Cancer: Emerging Trends in Clinics. Biology (Basel) 10, (2021). 1057 52. C. Yan et al., Prediction and prognostic potential of NR3C1 gene expression level in 1058 DLBCL patients. Hematology 28, 2251199 (2023). 1059 53. P. Sun et al., Substrate recognition diversity and transport dynamics of ABCC1. Nature 1060 communications 16, 10499 (2025). 1061 54. H. Kim et al., Clinical outcome of relapsed or refractory burkitt lymphoma and mature 1062 B-cell lymphoblastic leukemia in children and adolescents. Cancer Res Treat 46, 358-1063 365 (2014). 1064 55. A. D. Hughes, P. Polonen, D. T. Teachey, Relapsed childhood T-cell acute lymphoblastic 1065 leukemia and lymphoblastic lymphoma. Haematologica, (2025). 1066 56. V . Kohlhas, M. Hallek, P. H. Nguyen, Constitutive activation of Lyn kinase enhances 1067 BCR responsiveness, but not the development of CLL in Emicro -TCL1 mice. Blood 1068 Adv 4, 6106-6116 (2020). 1069 57. X. Zhao et al., Fc receptor-like 1 intrinsically recruits c-Abl to enhance B cell activation 1070 and function. Sci Adv 5, eaaw0315 (2019). 1071 58. K. A. Spivey et al. , A role for collagen XXIII in cancer cell adhesion, anchorage -1072 independence and metastasis. Oncogene 31, 2362-2372 (2012). 1073 59. B. Der, H. Bugacov, B. M. Briantseva, A. P. McMahon, Cadherin adhesion complexes 1074 direct cell aggregation in the epithelial transition of Wnt -induced nephron progenitor 1075 cells. Development 151, (2024). 1076 60. G. Minuesa et al. , Small -molecule targeting of MUSASHI RNA -binding activity in 1077 acute myeloid leukemia. Nature communications 10, 2691 (2019). 1078 61. R. Burger, T. E. Hansen-Hagge, H. G. Drexler, M. Gramatzki, Heterogeneity of T-acute 1079 lymphoblastic leukemia (T -ALL) cell lines: suggestion for classification by 1080 immunophenotype and T-cell receptor studies. Leuk Res 23, 19-27 (1999). 1081 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 38 62. C. A. Smith et al., CD30 antigen, a marker for Hodgkin's lymphoma, is a receptor whose 1082 ligand defines an emerging family of cytokines with homology to TNF. Cell 73, 1349-1083 1360 (1993). 1084 63. P. V . T et al., Thioredoxin-1, chemokine (C-X-C motif) ligand-9 and interferon-gamma 1085 expression in the neoplastic cells and macrophages of Hodgkin lymphoma: 1086 clinicopathologic correlations and potential prognostic implications. Leuk Lymphoma 1087 58, 1-13 (2017). 1088 64. M. Niens et al. , Serum chemokine levels in Hodgkin lymphoma patients: highly 1089 increased levels of CCL17 and CCL22. Br J Haematol 140, 527-536 (2008). 1090 65. A. Birgersdotter et al., Inflammation and tissue repair markers distinguish the nodular 1091 sclerosis and mixed cellularity subtypes of classical Hodgkin's lymphoma. British 1092 journal of cancer 101, 1393-1401 (2009). 1093 66. S. Hartmann et al. , Macrophages in T cell/histiocyte rich large B cell lymphoma 1094 strongly express metal -binding proteins and show a bi -activated phenotype. 1095 International journal of cancer. Journal international du cancer 133, 2609-2618 (2013). 1096 67. S. Bordin, E. T. Young, Tumor -associated macrophages as the primary source of 1097 lysozyme in the urine of mice bearing GPC-11, a transplantable reticulum cell sarcoma. 1098 J Natl Cancer Inst 57, 827-835 (1976). 1099 68. A. P. Cardoso et al., The immunosuppressive and pro-tumor functions of CCL18 at the 1100 tumor microenvironment. Cytokine Growth Factor Rev 60, 107-119 (2021). 1101 69. E. Zajac et al. , Angiogenic capacity of M1 - and M2 -polarized macrophages is 1102 determined by the levels of TIMP -1 complexed with their secreted proMMP -9. Blood 1103 122, 4054-4067 (2013). 1104 70. K. Wang, G. Wei, D. Liu, CD19: a biomarker for B cell development, lymphoma 1105 diagnosis and therapy. Exp Hematol Oncol 1, 36 (2012). 1106 71. G. Pavlasova, M. Mraz, The regulation and function of CD20: an "enigma" of B -cell 1107 biology and targeted therapy. Haematologica 105, 1494-1506 (2020). 1108 72. M. Pourmaleki et al., Multiplexed Spatial Profiling of Hodgkin Reed -Sternberg Cell 1109 Neighborhoods in Classic Hodgkin Lymphoma. Clin Cancer Res 30, 3881-3893 (2024). 1110 73. T. Yang et al., LILRB4, an immune checkpoint on myeloid cells. Blood Sci 4, 49-56 1111 (2022). 1112 74. Y . Chen et al., PPA1 promotes oxidative phosphorylation and malignant progression of 1113 colorectal cancer under glucose restriction via AMPK/ULK1/FUNDC1 -mediated 1114 mitophagy. Cell Death Discov 11, 549 (2025). 1115 75. S. Wang et al., PPA1, an energy metabolism initiator, plays an important role in the 1116 progression of malignant tumors. Front Oncol 12, 1012090 (2022). 1117 76. M. Mentes, C. Yandim, Identification of PPA1 inhibitor candidates for potential 1118 repurposing in cancer medicine. J Cell Biochem 124, 1646-1663 (2023). 1119 77. S. Goswami, K. E. Pauken, L. Wang, P. Sharma, Next -generation combination 1120 approaches for immune checkpoint therapy. Nat Immunol 25, 2186-2199 (2024). 1121 78. M. A. Weniger, R. Kuppers, Molecular biology of Hodgkin lymphoma. Leukemia 35, 1122 968-981 (2021). 1123 79. T. Aoki et al. , Single -Cell Transcriptome Analysis Reveals Disease -Defining T-cell 1124 Subsets in the Tumor Microenvironment of Classic Hodgkin Lymphoma. Cancer 1125 Discov 10, 406-421 (2020). 1126 80. R. Kuppers et al., Identification of Hodgkin and Reed-Sternberg cell-specific genes by 1127 gene expression profiling. J Clin Invest 111, 529-537 (2003). 1128 81. S. C. Peh, L. H. Kim, S. Poppema, TARC, a CC chemokine, is frequently expressed in 1129 classic Hodgkin's lymphoma but not in NLP Hodgkin's lymphoma, T -cell-rich B-cell 1130 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 39 lymphoma, and most cases of anaplastic large cell lymphoma. Am J Surg Pathol 25, 1131 925-929 (2001). 1132 82. M. H. Barros, P. Segges, G. Vera -Lozada, R. Hassan, G. Niedobitek, Macrophage 1133 polarization reflects T cell composition of tumor microenvironment in pediatric 1134 classical Hodgkin lymphoma and has impact on survival. PLoS One 10, e0124531 1135 (2015). 1136 83. G. Srikrishna, S100A8 and S100A9: new insights into their roles in malignancy. J Innate 1137 Immun 4, 31-40 (2012). 1138 84. D. P. O'Malley et al. , American Registry of Pathology Expert Opinions: 1139 Immunohistochemical evaluation of classic Hodgkin lymphoma. Ann Diagn Pathol 39, 1140 105-110 (2019). 1141 85. S. Joos et al., Classical Hodgkin lymphoma is characterized by recurrent copy number 1142 gains of the short arm of chromosome 2. Blood 99, 1381-1387 (2002). 1143 86. A. Tzankov et al., Rare expression of T-cell markers in classical Hodgkin's lymphoma. 1144 Mod Pathol 18, 1542-1549 (2005). 1145 87. M. Revel, C. Sautes -Fridman, W. H. Fridman, L. T. Roumenina, C1q+ macrophages: 1146 passengers or drivers of cancer progression. Trends Cancer 8, 517-526 (2022). 1147 88. X. X. Hou, X. Q. Wang, W. J. Zhou, D. J. Li, Regulatory T cells induce polarization of 1148 pro-repair macrophages by secreting sFGL2 into the endometriotic milieu. Commun 1149 Biol 4, 499 (2021). 1150 89. A. T. Al Haq, H. Y . Tseng, L. M. Chen, C. C. Wang, H. L. Hsu, Targeting prooxidant 1151 MnSOD effect inhibits triple -negative breast cancer (TNBC) progression and M2 1152 macrophage functions under the oncogenic stress. Cell Death Dis 13, 49 (2022). 1153 90. F. Khan et al. , Lactate dehydrogenase A regulates tumor -macrophage symbiosis to 1154 promote glioblastoma progression. Nature communications 15, 1987 (2024). 1155 91. M. H. Do et al. , Reprogramming tumor -associated macrophages to outcompete 1156 endovascular endothelial progenitor cells and suppress tumor neoangiogenesis. 1157 Immunity 56, 2555-2569 e2555 (2023). 1158 92. B. Burkhardt et al., Treatment and Outcome Analysis of 639 Relapsed Non -Hodgkin 1159 Lymphomas in Children and Adolescents and Resulting Treatment Recommendations. 1160 Cancers 13, (2021). 1161 93. L. Giulino -Roth et al. , Outcome of children and adolescents with relapsed Hodgkin 1162 lymphoma treated with high-dose therapy and autologous stem cell transplantation: the 1163 Memorial Sloan Kettering Cancer Center experience. Leuk Lymphoma 59, 1861-1870 1164 (2018). 1165 94. M. van de Wetering, W. de Lau, H. Clevers, WNT signaling and lymphocyte 1166 development. Cell 109 Suppl, S13-19 (2002). 1167 95. M. Descatoire et al. , Critical role of WASp in germinal center tolerance through 1168 regulation of B cell apoptosis and diversification. Cell reports 38, 110474 (2022). 1169 96. Z. Zhong et al., Human immune organoids to decode B cell response in healthy donors 1170 and patients with lymphoma. Nat Mater 24, 297-311 (2025). 1171 97. G. Varano et al., B-cell Receptor Silencing Reveals the Origin and Dependencies of 1172 High-Grade B-cell Lymphomas with MYC and BCL2 Rearrangements. Blood Cancer 1173 Discov 6, 364-393 (2025). 1174 98. R. M. Southern et al., IgM and IgG1 B cell receptors differentially affect B cell fates 1175 and dictate the pathogenesis of mature B cell lymphomas. Sci Adv 11, eadp9391 (2025). 1176 99. J. Corso et al., Elucidation of tonic and activated B -cell receptor signaling in Burkitt's 1177 lymphoma provides insights into regulation of cell survival. Proc Natl Acad Sci U S A 1178 113, 5688-5693 (2016). 1179 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint 40 100. H. Kanzler, R. Kuppers, M. L. Hansmann, K. Rajewsky, Hodgkin and Reed -Sternberg 1180 cells in Hodgkin's disease represent the outgrowth of a dominant tumor clone derived 1181 from (crippled) germinal center B cells. J Exp Med 184, 1495-1505 (1996). 1182 101. K. A. Blum, B -cell receptor pathway modulators in NHL. Hematology. American 1183 Society of Hematology. Education Program 2015, 82-91 (2015). 1184 102. J. Ge et al. , TRBC2 -targeting antibody -drug conjugates for the treatment of T cell 1185 cancers. Nat Cancer 6, 2011-2024 (2025). 1186 103. L. P. Vu et al., Functional screen of MSI2 interactors identifies an essential role for 1187 SYNCRIP in myeloid leukemia stem cells. Nat Genet 49, 866-875 (2017). 1188 104. J. Zhang et al., Clonal evolution dissection reveals that a high MSI2 level promotes 1189 chemoresistance in T-cell acute lymphoblastic leukemia. Blood 143, 320-335 (2024). 1190 105. T. Erazo et al., TP53 mutations and RNA-binding protein MUSASHI-2 drive resistance 1191 to PRMT5 -targeted therapy in B -cell lymphoma. Nature communications 13, 5676 1192 (2022). 1193 106. M. Sureda -Gomez et al. , Tumorigenic role of Musashi -2 in aggressive mantle cell 1194 lymphoma. Leukemia 37, 408-421 (2023). 1195 107. D. Luo et al., PPA1 promotes NSCLC progression via a JNK - and TP53-dependent 1196 manner. Oncogenesis 8, 53 (2019). 1197 108. S. H. Jeong et al., Pyrophosphatase overexpression is associated with cell migration, 1198 invasion, and poor prognosis in gastric cancer. Tumour Biol 33, 1889-1898 (2012). 1199 109. S. Wu et al., CD74(+)CCL5(+) effector CD8(+) T cells drive mucosal inflammation and 1200 predict biologics response in inflammatory bowel disease. Journal of translational 1201 medicine 24, 22 (2025). 1202 1203 (which was not certified by peer review) is the author/funder. All rights reserved. No reuse allowed without permission. The copyright holder for this preprintthis version posted April 23, 2026. ; https://doi.org/10.64898/2026.04.21.719850doi: bioRxiv preprint