Generation and External Validation of a Histologic Transformation Risk Model for Patients with Follicular Lymphoma

Journal homepage: https://modernpathology.org/ Research Article Generation and External Validation of a Histologic Transformation Risk Model for Patients with Follicular Lymphoma Ismael Fern andez-Miranda a , Lucía Pedrosa a , Julia Gonz alez-Rinc on a , b , Blanca Espinet c , d , F atima de la Cruz Vicente e , Fina Climent f , Sagrario G omez a , Ana Royuela g , Francisca I. Camacho h , Paloma Martín-Acosta i , Natalia Yanguas-Cas as a , j , Marina Domínguez a , Miriam M endez a , k , Luis Colomo c , Antonio Salar l , Beatriz Horcajo a , Marta Navarro a ,M  onica García-Cosío m , Miguel Piris-Villaespesa n , Marta Llanos o , Juan F. García p , Silvia Sequero q , Santiago Mercadal r , Sonia García-Hern andez s , Bel en Navarro t , Manuela Mollejo u , Mariano Provencio a , v , Margarita S anchez-Beato a , * a Department of Medical Oncology, Lymphoma Research Group, Hospital Universitario Puerta de Hierro-Majadahonda, IDIPHISA, Madrid, Spain; b CoE Data Intelligence, Fujitsu Technology Solutions S.A., Pozuelo de Alarc on, Madrid, Spain; c Translational Research on Hematological Neoplasms Group, Cancer Research Program, Institut Hospital del Mar d’Investigacions M ediques (IMIM), Barcelona, Spain; d Department of Pathology, Hospital del Mar, Barcelona, Spain; e Department of Hematology, Hospital Universitario Virgen del Rocío, Instituto de Biomedicina de Sevilla (IBIS)/CSIC/Universidad de Sevilla, Seville, Spain; f Department of Pathology, Hospital Universitari de Bellvitge-IDIBELL, Barcelona, Spain; g Biostatistics Unit, Hospital Universitario Puerta de Hierro-Majadahonda, IDIPHISA. CIBERESP, ISCIII. Madrid, Spain; h Department of Pathology, Hospital Universitario de Getafe, Madrid, Spain; i Department of Pathology, Cancer Molecular Pathology Group, Hospital Universitario Puerta de Hierro-Majadahonda, IDIPHISA, Madrid, Spain; j Centro de Investigaci on Biom edica en Red Fragilidad y Envejecimiento Saludable (CIBERFES), Madrid, Spain; k Department of Medical Oncology, Hospital Universitario Puerta de HierroMajadahonda, IDIPHISA, Madrid, Spain; l Department of Hematology, Hospital del Mar, Barcelona, Spain; m Department of Pathology, Hospital Universitario Ram on y Cajal, Madrid, Spain; n Department of Hematology, Hospital Universitario Ram on y Cajal, Madrid, Spain; o Department of Oncology, Hospital Universitario de Canarias, Tenerife, Spain; p Department of Pathology, Hospital MD Anderson Cancer Center, Madrid, Spain; q Department of Oncology, Hospital Universitario San Cecilio, Granada, Spain; r Department of Hematology, ICO-Hospital Duran I Reynals, Barcelona, Spain; s Department of Pathology, Hospital Universitario de Canarias, Tenerife, Spain; t Department of Hematology, Hospital Universitario Puerta de Hierro, Majadahonda, Madrid, Spain; u Department of Pathology, Complejo Hospitalario de Toledo, Spain; v Department of Medical Oncology, Hospital Universitario Puerta de Hierro-Majadahonda, Facultad de Medicina, Universidad Aut onoma de Madrid, IDIPHISA, Madrid, Spain ARTICLE INFO Article history: Received 23 November 2023 Revised 23 April 2024 Accepted 4 May 2024 Available online 17 May 2024 Keywords: follicular lymphoma genomics histologic transformation predictive model ABSTRACT Follicular lymphoma (FL) is the most frequent indolent lymphoma. Some patients (10%-15%) experience histologic transformation (HT) to a more aggressive lymphoma, usually diffuse large Bcell lymphoma (DLBCL). This study aimed to validate and improve a genetic risk model to predict HT at diagnosis.We collected mutational data from diagnosis biopsies of 64 FL patients. We combined them with the data from a previously published cohort (total n ¼104; 62 from nontransformed and 42 from patients who did transform to DLBCL). This combined cohort was used to develop a nomogram to estimate the risk of HT. Prognostic mutated genes and clinical variables were assessed using Cox regression analysis to generate a risk model. The model was internally validated by bootstrapping and externally validated in an independent cohort. Its performance was evaluated using a concordance index and a calibration curve. The clinicogenetic nomogram included the mutational status of 3 genes (HIST1HE1,KMT2D, and TNFSR14) and high-risk Follicular Lymphoma International Prognostic Index and predicted HT with a concordance index of 0.746. Patients were classified as being at low or high risk of transformation. The probability HT function at 24 months was 0.90 in the low-risk group vs 0.51 in the high-risk *Corresponding author. E-mail address: [email protected] (M. S anchez-Beato). 0893-3952/©2024 THE AUTHORS. Published by Elsevier Inc. on behalf of the United States &Canadian Academy of Pathology. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/). https://doi.org/10.1016/j.modpat.2024.100516 Mod Pathol 37 (2024) 100516 group and, at 60 months, 0.71 vs 0.15, respectively. In the external validation cohort, the probability HT function in the low-risk group was 0.86 vs 0.54 in the high-risk group at 24 months, and 0.71 vs 0.32 at 60 months. The concordance index in the external cohort was 0.552. In conclusion, we propose a clinicogenetic risk model to predict FL HT to DLBLC, combining genetic alterations in HIST1H1E,KMT2D, and TNFRSF14 genes and clinical features (Follicular Lymphoma International Prognostic Index) at diagnosis. This model could improve the management of FL patients and allow treatment strategies that would prevent or delay transformation. ©2024 THE AUTHORS. Published by Elsevier Inc. on behalf of the United States &Canadian Academy of Pathology. This is an open access article under the CC BY-NC-ND license (http://creativecommons. org/licenses/by-nc-nd/4.0/). Introduction Follicular lymphoma (FL) is the second most common type of B-cell non-Hodgkin lymphoma, with an annual incidence of 3 to 5 per 100,000 habitants in the United States and Europe. 1 FL, a heterogenous but generally indolent disease, is characterized by slow progression, high response rates to therapy, and long overall survival (OS). However, it remains an incurable disease. 2 Approximately 20% of patients experience disease progression within 24 months of treatment (POD24) and face poor outcomes. 2,3 Additionally, 10% to 30% of patients experience transformation to high-grade lymphoma, most frequently to a diffuse large B-cell lymphoma (DLBCL). 4-6 The histologic transformation (HT) event, which occurs in 2% to 3% of patients annually, has been associated with treatment resistance and worse prognosis and is a critical process in FL patients. 4,7-10 It is still challenging to predict transformation at diagnosis. Several studies have associated clinical risk factors, such as the Follicular Lymphoma International Prognostic Index (FLIPI), advanced stage, Eastern Cooperative Oncology Group (ECOG) performance status, and histologic grade, with a higher risk of transformation, but none of these is specifictoHT. 6,11,12 An association between early progression and risk of transformation has also been described, 13 both events associated with shorter OS. However, HT may occur in patients without evidence of progression and vice versa: some patients progress without further transformation. From a molecular point of view, the transformed DLBLC sample (tFL) is more complex genomically than the precedent FL sample (pre-tFL). Pre-tFL samples are more complex than nontransformed (ntFL) ones. 14-16 HT has previously been associated with alterations in TP53,CDKN2A,MYC,CCND3,GNA13, and SOCS1, among other genes. 14,15,17,18 A recent study by Dreval et al 19 shed some additional light on the FL transformation process and, using whole-genome sequencing, described 2 genetically distinct subgroups of FL, DLBCL-like (dFL) and constrained FL (cFL), whose time to transformation differed significantly. Each subtype had different mutational landscapes and aberrant somatic hypermutation rates. The cFL subtype was associated with a lower mutational burden, in line with our previous results. 15 The dFL subtype showed a similar mutational pattern to that in DLBCL, including, for example, a higher incidence of GNA13 mutations. Our results and others revealed greater genomic complexity in pre-tFL than in ntFL samples, with higher rates of subclonal mutations. 15,17,18 The transformation process is the consequence of genetic alterations in tumor B cells and the result of interactions between lymphoma cells and the immune system. The type and distribution of immune cells that infiltrate the tumor microenvironment have been proposed as predictors of transformation. 20 In 2019, Gonz alez-Rinc on et al 15 published a genetic model to predict HT at diagnosis in patients with FL based on mutations in the NOTCH2,DTX1,UBE2A, and HIST1H1E genes. The multivariate Cox regression coefficients classified patients into low-, intermediate-, and high-risk groups. In the present study, we have generated and validated an improved clinicogenetic model to predict the HT of FL into DLBCL. Materials and Methods Patients and Samples Formalin-fixed, paraffin-embedded tissue diagnostic samples from 64 patients with FL (cohort 2022 or C'22) were collected from 16 Spanish hospitals between 2006 and 2018. Twenty-three of the 64 patients underwent HT to DLBCL, and 41 remained nontransformed (Supplementary Fig. S1A, Supplementary Tables S1 and S2). HT was diagnosed when new symptoms appeared, and a new biopsy was performed to confirm the transformation to DLBLC. Transformation into a DLBCL was defined as the histologic progression into a high-grade lymphoma with a diffuse pattern. Patients diagnosed with FL were selected for this study according to the following criteria: grade 1 to 3A, excluding grade 3B cases, and a minimum follow-up of 5 years for patients without HT. Mutational data from FL diagnostic samples from 19 of 64 patients (P2022_51 to P2022_74, Supplementary Tables S1 and S2) have been previously published. 21 Median follow-up was 7 years (8 years for patients still alive) (range, 3.9-10.0 years), minimum follow-up for alive ntFL patients was 4.8 years. The mean time to HT for the whole series was 5.2 years (range, 0.16-10.0 years) and 2.9 years for transformed cases (0.16-5.8 years). Further details are provided in Supplementary Table S1. Additionally, results from the previously published cohort (2018 cohort or C'18) 15 were used in this study. The C'18 cohort includes 19 cases diagnosed with FL who underwent HT to DLBCL and 21 cases without transformation (Table 1 and Supplementary Table S1). Patients included in this study were selected following the same criteria as the C'22 cohort. The median follow-up of the C 0 18 series was 6.7 years (8.0 years for live patients). The study was approved by the Ethics Committee of the Hospital Universitario Puerta de Hierro-Majadahonda (PI-67/14). It was conducted in compliance with the principles of the Declaration of Helsinki. All participants gave their signed informed consent for inclusion. Samples were collected, and clinical data were managed following standardized protocols to guarantee the quality of the samples and the confidentiality of donor data. After approval by the corresponding ethics committees, material and data from other centers were anonymously Ismael Fern andez-Miranda et al. / Mod Pathol 37 (2024) 100516 2 transferred to our laboratory in compliance with the current Spanish legislation (Ley 14/2007 de Investigaci on Biom edica and Real Decreto 1716/2011). DNA Extraction and Targeted DNA Sequencing All samples were reviewed upon arrival to confirm the diagnosis and select tumor celleenriched areas in the cases that required it. Further details are in the Supplementary Methods in the Supplementary Information. We used 2 targeted sequencing panels (human reference genome GRCh38/hg38), including coding regions, splice sites and untranslated regions of genes recurrently altered in germinal center-derived B-cell lymphomas (Supplementary Table S3) 15,21,22 : SureSelect XT HS, Agilent Technologies, and Twist Bioscience. We analyzed the 44 genes that were common to the 2 panels. The probes designed for the common genes covered the same regions with the 2 systems (both based on hybridization and capture). They delivered similar results in terms of coverage and depth. Fifty DNA libraries were prepared with the SureSelect XT HS kit and 14 with the Twist Target Enrichment kit, as previously described. 21 Genomic data have been deposited in the Sequence Read Archive (BioProject ID PRJNA904556). Bioinformatics Pipeline A bioinformatic analysis was performed, and the genetic alterations were identified as previously described. 15,21 See Supplementary Information for further details. Statistical Analyses Time to transformation was defined as the duration of the period from the date of FL diagnosis to that of HT or last follow-up. We used the Cox proportional hazards (PH) model for the univariate and multivariable analyses. 23 We did not consider the competing risks approach due to the small number of deaths in the nontransformed group; the external validation process in the competing risk setting is more complex to apply with the most used statistical approaches. We developed the following 3 models: (1) we selected those genes with more than 4 mutations and performed a Cox univariate regression for each gene. Genes with a Pvalue of <.10 were selected for inclusion in a multivariable Cox regression model derived by backward elimination and genes with a Pvalue of <.05 were retained in the final “genetic”model; (2) in parallel, we developed the “clinical”model, which included FLIPI, by adopting the Cox PH method; and (3) we merged the genetic and clinical variables to obtain the “clinicogenetic”Cox PH model. We noted the corresponding Harrell’s C-index, the Akaike information criterion (AIC), and the Bayesian information criterion (BIC) for each of the 3 models. AIC is a stastistical model based on in-sample fit to estimate the likelihood of a model to predict/estimate future values. BIC is another criterion for model selection that measures the trade-off between model fit and complexity of the model. A lower AIC or BIC value indicates a better fit. 24 We compared the 3 features in the “clinical”and “clinicogenetic”models and selected the better model of the 2. The PH assumption was assessed by inspecting the Schoenfeld residuals. We also estimated the Brier score for the derivation and validation cohorts as an overall measure of fit. 25 The Brier score measures the accuracy of probabilistic predictions and may take a value between 0 and 1, with a smaller value indicating a better outcome. The probability of HT at 2 or 5 years for an individual patient was calculated using the following equation, derived from the Cox PH model: b PðHTÞ¼1S0ðtÞexp ðPrognostic IndexÞ where S 0 (t) is the baseline probability HT function at time t(ie, at 2 years), and the Prognostic Index is the sum of the products of the predictors and their coefficients. For ease of interpretation, we developed a nomogram 26,27 that shows the risk of transformation at 2 and 5 years according to the variables retained in the final model. The internal validity of the model was assessed using a bootstrapping approach with 500 resamples. Two risk groups (low and high risk) were established based on the median value of the linear predictor 28 and illustrated graphically with Kaplan-Meier curves. The log-rank test was used to compare the 2 groups. Calibration was evaluated through 2 calibration plots at 2 and 5 years. We also developed a Kaplan-Meier graph with each risk group's observed vs expected curves. Discrimination was assessed with Harrell’s C-index. 29 The AIC and BIC were also calculated. External validation was performed in the British Columbia Cancer (BCC) independent cohort, 14 which included mutational data from 246 FL diagnostic samples. We used the mutational and Table 1 Summary of clinical features of follicular lymphoma whole cohort Variables C 0 18 and C 0 22 P Total ntFL pre-tFL N % N% N% Patients 104 62 42 Female 64 61.5 35 55.0 29 69.0 .195 Age >60 y 45 43.3 28 45.2 17 40.5 .636 FLIPI 104 62 42 .001 High grade 29 27.9 10 16.1 19 42.2 Low/intermediate grade 75 72.1 52 83.9 23 54.8 Exitus 31 29.8 7 11.3 24 57.1 .001 ECOG 93 52 41 .045 0 52 55.9 35 67.3 17 41.4 1 36 38.7 15 28.8 21 51.2 2 5 5.4 2 3.8 3 7.3 Stage 103 61 42 .086 I 7 6.8 7 11.5 0 0.0 II 16 15.5 8 13.1 8 19.0 III 30 29.1 15 24.5 15 35.7 IV 50 48.5 31 50.8 19 45.2 Firstline treatment 103 61 42 .161 R-CHOP 44 42.7 26 42.6 18 42.8 R-Bendamustine 22 21.4 14 22.9 8 19.0 Other IC (with R) 22 21.4 10 16.4 12 28.6 Chemo 10 9.7 6 9.8 4 9.5 W&W/radio 5 4.9 5 8.2 0 0.0 Grade 83 50 33 .075 I 21 25.3 17 34.0 4 12.1 II 40 48.2 22 44.0 18 54.5 IIIA 22 26.5 11 22.0 11 33.3 BCL2 rearrangement 57 40 17 .222 t(14;18) positive 40 70.2 30 75.0 10 58.8 t(14;18) negative 17 29.8 10 25.0 7 41.2 C 0 18, cohort 2018 15 ; ECOG, Eastern Cooperative Oncology Group performance status; C 0 22, cohort 2022; FL, follicular lymphoma; FLIPI, Follicular Lymphoma International Prognostic Index; IC, immunochemotherapy; ntFL, nontransformed FL; pre-tFL, pretransformed FL; R-CHOP, rituximab, cyclophosphamide, doxorubicin, vincristine, prednisone. Ismael Fern andez-Miranda et al. / Mod Pathol 37 (2024) 100516 3 clinical data from 146 cases. We excluded data from patients whose FLIPI or follow-up time information was unavailable (Supplementary Fig. S1B). The original publication did not provide individual follow-up times for ntFL patients. Still, according to their results, all these patients had a minimum OS (follow-up) of 7 years. The cohort included patients diagnosed with FL grade 1 to 3A, excluding grade 3B cases. Calibration was evaluated through a Kaplan-Meier curve showing the observed vs the expected curves for each risk group in the validation cohort. We also created a calibration plot for every follow-up year. Discrimination was assessed with Harrell’s C-index. Pearson’s c 2 test was used to compare categoric data between 2 groups. Statistical analyses were performed using STATA v18 (StataCorp, 2023, Stata Statistical Software: Release 18: StataCorp LLC.). Results The C'22 cohort comprised 64 patients diagnosed with FL (see Materials and Methods section). There were 25 males (39.1%) and 39 females; 32 patients (50%) were older than 60 years at the time of diagnosis, and 15 (23.4%) had a high-risk FLIPI. Twenty-three of the 64 patients underwent HT to DLBCL, and 41 remained nontransformed (Table 1,Supplementary Fig. S1A, Supplementary Tables S1 and S2). We performed a targeted deep-sequencing analysis in the C'22 cohort diagnostic FL samples as previously described. 15,21,22 A total of 520 alterations were identified, considering nonsynonymous mutations in the exonic regions and splice sites. The most frequently mutated genes were KMT2D (67.2%), CREBBP (65.6%), BCL2 (46.9%), TNFRSF14 (32.8%), IGLL5 (26.6%), and EZH2 (21.9%) (Fig. 1,Supplementary Table S4), with frequencies similar to those previously described in other series. 18,30,31 Some genes were more frequently mutated in pre-tFL samples, such as EP300 (in 26% of the pre-tFL cases vs 12% of ntFL cases), HIST1H1E (17% vs 5%), TP53 (13% vs 5%), and CSMD3 (13% vs 2.5%). Other genes were more frequently mutated in ntFL samples, such as TNFRS14 (in 17.4 % of the pre-tFL cases vs 41.5% of ntFL cases), CREBBP (56% vs 71%), KMT2D (61% vs 71%), and EZH2 (13% vs 27%) (Fig. 1). The study's initial aim was to validate the HT-predictive model previously generated by Gonz alez-Rinc on et al. 15 However, when we applied it to the C 0 22 cohort, no significant differences were found in the HT probabilities between the 3 risk groups (log-rank test, P¼.21, based on the Kaplan-Meier analysis), probably due to the underpowered analysis performed before (Supplementary Fig. S2). Therefore, we combined the C'22 cohort with the previously published C'18 cohort 15 to generate a new HT risk model. Thus, the new cohort included 104 cases, of which 62 were ntFL and 42 were pre-tFL (Supplementary Fig. S1A). The patient characteristics of the 2 cohorts are summarized in Table 1 and Supplementary Table S1. We performed a Cox univariate regressiontoexamine the genes mutated in at least 5% (5 cases) in the whole combined cohort (CREBBP,TP53,PIM1,BCL2,CSMD3,CARD11,EP300,POU2AF1, NOTCH2,UBE2A,DTX1,KMT2D,ATP6V1B2,KMT2C,TNFRSF14,IRF8, IGLL5,HIST1H1C,DSP,SOCS1,MEF2B,PIK3CD,GNA13,ARID1A,STAT6, EZH2,B2M,EBF1,HIST1H1E,BCL7A,FOXO1,POU2F2,PAX5,ATM, and TNFAIP3). The univariate analysis found that nonsynonymous mutations in the CSMD3,NOTCH2,UBE2A,B2M, and HIST1H1E genes were associated with a higher risk of HT, and mutations in TNFRSF14 and KMT2D were associated with a lower risk of HT (P< .10) (Fig. 2A). A multivariate Cox regression model initially including all these genes was derived by backward elimination. The results showed that mutations in HIST1H1E and B2M were significantly associated with a higher risk of HT, and mutations in TNFRSF14 and KMT2D were linked to a lower risk of HT (P<.05) (Fig. 2B). These 4 genes comprise the final genetic model. We used univariate PH Cox analysis to estimate the associations of clinical characteristics with HT. High-risk FLIPI 32 (FLIPI score 3) was found to be associated with HT (HR, 3.046; 95% CI,1.652-5.615; P<.001), but no significant association was found with gender, stage, age >60 years, or histologic grade. ECOG was not analyzed, because there were so few patients with an ECOG >1(n¼5). Therefore, the FLIPI was the only variable in the clinical model. For the final clinicogenetic histologic transformation risk (HTR) model, we combined the FLIPI and genes in the genetic model and finally included FLIPI and HIST1H1E,KMT2D, and TNFSR14 genes (Supplementary Table S5). The clinicogenetic HTR model was slightly, but not significantly, better than the genetic model (Cindices: 0.746 vs 0.704; P¼.297) but significantly better than the FLIPI (C-indices: 0.746 vs 0.618; P<.001). The AIC and BIC values also indicated that the clinicogenetic HTR model was better than the others (Table 2). Brier scores were also calculated. At 2 years, these were 0.098 for the clinicogenetic HTR model and 0.121 for the FLIPI. The 5-year corresponding values were 0.175 and 0.217, respectively (Table 2). A nomogram was then constructed based on the HTR model to predict the risk of transformation at 2 and 5 years (Fig. 3A). The calibration plots showed good agreement between the 2and Figure 1. Genetic alterations identified in the tumor DNA samples of the C022 cohort. The bar diagram represents the most frequently altered genes in patients who did (red) or did not (green) undergo histologic transformation. Ismael Fern andez-Miranda et al. / Mod Pathol 37 (2024) 100516 4 5-year predicted and observed risk of HT (Fig. 3B, C, Supplementary Fig. S4A). The probabilities of HT at 2 and 5 years for an individual patient were calculated using the following equations, derived from the Cox PH model: b PðHT at 2 yearsÞ¼10:9954exp ðPrognostic IndexÞ b PðHT at 5 yearsÞ¼10:9858exp ðPrognostic IndexÞ where Prognostic Index ¼1.639193 *FLIPI (high-risk) þ1.383305 *KMT2D (not mutated) þ1.615207 *TNFRSF14 (not mutated) þ 2.102908 *HIST1H1E (mutated). A Web-based calculator is available at https://github.com/Lymphoma-IDIPHISA. We internally validated the HTR model by bootstrapping with 500 resamples. Discrimination was evaluated using Harrell's C and Somers' D indices, 29 which indicated that the model had a good discriminating ability (Harrell's C ¼0.746; Somers' D ¼0.492). Patients were classified into lowand high-risk HT groups based on the median value of the linear predictor. 28 The KaplanMeier time to HT analysis revealed that patients in the HT highrisk group had a significantly shorter time to transformation than those in the low-risk group (Fig. 4A, Supplementary Fig. S5A). The HT time-to-event probability at 24 months in the low-risk group was 0.90 (95% CI, 0.81-0.95) vs 0.51 (95% CI, 0.24-0.73) in the high-risk group; and at 60 months it was 0.71 (95% CI, 0.61-0.80) and 0.15 (95% CI, 0.02-0.37), respectively. Finally, we validated the model in an external cohort. We used the BCC cohort 8 described in Materials and Methods (Supplementary Fig. S1B, Supplementary Table S7). We applied the HTR model to 146 cases (65 pre-tFL and 81 ntFL) after discarding cases for which the FLIPI or follow-up time information was unavailable. The calibration plots in the validation cohort also showed good agreement between the predicted and observed 1to 7-year risk of HT (Supplementary Fig. S3B). When we applied the nomogram and divided the series into lowand high-risk groups as described before, the Kaplan-Meier analysis revealed a significantly shorter time to transformation among the high-risk group of patients than the low-risk group (Fig. 4B, Supplementary Fig. S4B). The HT time-to-event function at 24 months in the low-risk group was 0.86 (95% CI, 0.79-0.91) vs 0.54 (95% CI, 0.32-0.72) in the high-risk group. At 60 months, the values were 0.71 (95% CI, 0.62-0.78) and 0.32 (95% CI, 0.14-0.51), respectively. The values of Harrell's C-index (0.552) and Somers' D (0.105) pointed to a fair level of discrimination in the external validation cohort. The Brier score for the validation cohort was 0.138 (95% CI, 0.099-0.178) at 2 years and 0.208 (95% CI, 0.178-0.238) at 5 years. Figure 2. Forest plots of (A) univariate and (B) multivariate Cox regression analysis of the effect of mutated genes of histologic transformation in follicular lymphoma patients. Table 2 Harrell’s C-index, the Akaike information criterion, the Bayesian information criterion, and the Brier scores for the 3 histologic transformation risk models Models C-index AIC BIC Brier score 2 y (95% CI) Brier score 5 y (95% CI) Genetic 0.704 345.834 356.411 Clinic (FLIPI) 0.618 355.836 358.481 0.121 (0.076-0.166) 0.217 (0.179-0.256) Clinicogenetic 0.746 331.635 342.213 0.098 (0.056-0.140) 0.175 (0.132-0.218) AIC, Akaike information criterion; BIC Bayesian information criterion; C-index, Harrell’s C-index; FLIPI, Follicular Lymphoma International Prognostic Index. Ismael Fern andez-Miranda et al. / Mod Pathol 37 (2024) 100516 5 Discussion Transformation to an aggressive lymphoma is a critical event in the clinical course of FL patients. It has been associated with a worse prognosis than that of patients who do not experience HT. 6 Additionally, transformation <2 years after diagnosis is associated with shorter survival than that in patients who undergo later transformations. 33 Therefore, predicting HT at the time of diagnosis is a valuable clinical task. Several clinicopathologic factors have been associated with a higher risk of HT, such as a high FLIPI, advanced stage, ECOG 2, and grade 3A. 7,34 Some individual genetic alterations have also been associated with HT, such as TP53 mutations, MYC alterations, CDKN2A loss by deletion or promoter hypermethylation, and other factors. 14,15,17,18 Several attempts have been made to predict clinical outcomes of FL patients by combining clinical factors and mutations in multiple genes, such as m7-FLIPI, to predict failure-free survival, 35 and POD24-PI, to predict early progression after firstline immunochemotherapy. 36 However, their predictive value differs between series, and none of them can predict HT. 6,35,37 Our previous analysis led us to propose a mutational signature associated with a shorter time to transformation. 15 However, we still lack an accurate tool that can predict transformation at diagnosis. In this study, we propose a new clinicogenetic model to predict FL risk of HT to DLBCL based on the FLIPI and the mutational status of 3 genes. We used a nomogram to develop an applicable method Figure 3. (A) Nomogram estimating 2-year and 5-year probability of follicular histologic transformation to diffuse large B-cell lymphoma. (B) Calibration curve of the nomogram estimating histologic transformation probability for follicular lymphoma patients at 2 years and (C) 5 years. Error bars ¼95% CIs. Blue dashed diagonal lines represent the perfect calibration models with identical predicted and observed probabilities. Ismael Fern andez-Miranda et al. / Mod Pathol 37 (2024) 100516 6 that could predict individual risk of HT to DLBLC at 2 and 5 years, taking into consideration the FLIPI and the mutational data of HIST1H1E,KMT2D, and TNFSR14. HIST1H1E is a chromatin-remodeling gene whose alterations have been identified as clonal in the FL samples and have been previously associated with HT. 14,15,17 A recent study by Yusufova et al 38 analyzed the effect of H1 lymphoma-associated alterations in a mouse model that includes HIST1H1E. They showed that loss of HIST1H1E and HIST1H1C bestowed enhanced self-renewal capacity on germinal center B cells, which prompted the development of aggressive lymphomas. KMT2D (also known as MLL2) encodes an H3K4 trimethyltransferase. Loss-of-function mutations, which frequently occur in this gene, are clonal, ancestral, driver mutations in FL that give rise to B-cell lymphomagenesis. 39,40 KMT2D is the most commonly mutated gene in our cohort, having been noted as occurring in up to 80% of cases in 2 earlier studies. 18,30 This figure is higher than that recorded in DLBCL. 22,41,42 The relationship of KMT2D with B lymphomagenesis indicates a possible dependence on the loss of function of KMT2D in FL tumors but less so in DLBCL, supporting the hypothesis that KMT2D gene alteration limits the ability of FL to acquire additional genetic alterations for HT. However, although we found an inverse association with HT, the mechanism responsible has not been investigated. However, the role of TNFRSF14 alterations is controversial in lymphomas. TNFRS14 encodes herpes virus entry mediator, which is known to activate the NFk B pathway, leading to the induction of proinflammatory and cell survivalepromoting genes, but which can function as a “molecular switch”with activating or inhibitory functions. 43 Loss of TNFRSF14 in mouse models leads to cellautonomous activation of B-cell proliferation, which drives the development of GC lymphomas. 44 Some studies found an association of TNFRSF14 mutations with shorter OS and worse prognosis, 45,46 whereas others found TNFRSF14 mutations and deletions (1p36.23-1p36.32) to be associated with better prognosis 47,48 or to be more frequently mutated in ntFL than in pre-tFL cases. 15 Mutations in TNFRSF14 are common in a subgroup of t(14;18)-negative FLs with inguinal localization, which are linked to a localized presentation and a more indolent disease. 49 But TNFRSF14 is also frequently mutated in classic t(14;18)-positive FL cases. 46,50 We have found TNFRSF14 mutations in 32 of 104 cases. Our series included (14;18) positive and t(14;18) negative FL (Table 1), and TNFSRF14 mutations were detected in both subgroups (12 and 4 FL, respectively). A recent study by Dreval et al 19 has shown that mutations in TNFRS14 and KMT2D genes, among others, were more frequently detected in non-HT cases, which is consistent with our study. This new clinicogenetic model could pave the way to HT prediction by combining genetic alterations with clinical features. HT stratification could be improved by identifying 2 risk groups, leading to a higher proportion of HT cases being assigned to the high-risk and lower-risk groups, more accurately than solely with FLIPI. We have internally cross-validated the model and validated it in an external cohort. Nevertheless, our study has limitations because it is based on retrospective data of patients, some of whom were diagnosed as early as 2006. We recognize that more recent series might have behaved differently. Therefore, further clinical validation of the model in prospective series is needed to be eventually used as an HT risk predictor. Additionally, the immune microenvironment of FL is being extensively investigated, and the different tumor microenvironment immune compositions in different samples may influence the path of tumor evolution and the clinical course. Integrating genetic alterations and microenvironment composition should give us a better understanding of the HT process and allow us to identify integrative signatures that can predict HT risk and could eventually be used as biomarkers for clinical decisions regarding treatment and followup. In conclusion, we have built and externally validated a clinicogenetic HT risk model, combining mutation information from only 3 genes (HIST1H1E,KMT2D, and TNFRSF14) with clinical features at diagnosis (FLIPI). Predicting the risk of HT is relevant for patients with FL, improving their clinical management and allowing treatment strategies that could prevent or delay transformation. Acknowledgments The authors are indebted to the patients who contributed to this study and the members of Grupo Oncol ogico para el Tratamiento y el Estudio de los Linfomas (GOTEL). The authors especially thank E. Ramil (Sequencing Unit) and R. Mu~ noz-Viana Figure 4. KaplaneMeier histologic transformation time-to-event estimates. (A) Time to transformation for the combined C022 and C018 cohort, and (B) for the British Columbia Cancer validation cohort, stratified into low-risk and high-risk groups by the median value of the linear predictor. Ismael Fern andez-Miranda et al. / Mod Pathol 37 (2024) 100516 7 (Bioinformatics Unit) of the Instituto de Investigaci on Sanitaria Puerta de Hierro-Segovia de Arana (IDIPHISA) for their help. The authors acknowledge the Biobanks of the Hospital Universitario Puerta de Hierro-Majadahonda, Parc de Salut MAR (MARBiobanc), Barcelona (supported by grants from ISCIII) (PT20/ 00023), and the Xarxa de Bancs de Tumors de Catalunya (XBTC) (sponsored by Pla Director d'Oncologia de Catalunya), the Hospital Universitari de Bellvitge-IDIBELL (PT20/00171), and the Sistema Sanitario Público de Andalucía. Author Contributions M.S.-B., I.F.-M., and M.P. contributed to conception and design. M.S.-B. and M.P. made financial support. M.S.-B. was involved in administrative support. B.E., F.dl.C., F.C., F.I.C., P.M.-A., M.M.L.C., A.S., M.G.-C., M.P.-V., M.L., J.F.G., S.S., S.M., S.G.-H., B.N., M.M., and M.P. were involved in provision of study materials and patients. I.F.-M., L.P., J.G.-R., S.G., N.Y.-C., M.D., B.H., and M.N. contributed to collection and assembly of data. I.F.-M., A.R., M.P., and M.S.-B. were involved in data analysis and interpretation. I.F.-M. and M.S.-B. contributed to manuscript writing. All the authors critically reviewed and approved the final manuscript, and are accountable for all aspects of the work. Data Availability Genomic data have been deposited in the Sequence Read Archive (BioProject ID PRJNA904556). Funding This work was supported by the Spanish Ministry of Economy and Competitiveness (MINECO) and Instituto de Salud Carlos III (ISCIII), ISCIII-MINECO AES-FEDER (DTS17/00039, PI17/00272, PI20/00591, and PI23/01587); Direcci on General de Universidades e Investigaci on de la Consejería de Educaci on e Investigaci on de la Comunidad de Madrid (CAM) (B2017/BMD-3778); and Fundaci on de Investigaci on Biom edica Puerta de Hierro. L.P. and J.G.R. were recipients of iPFIS predoctoral fellowships (IFI18/0004, and IFI14/0003 respectively, ISCIII-MINECO AESFEDER, Plan Estatal IþDþI 2014-2020). I.F.M. is supported by B2017/BMD-3778 and the Fundaci on de Investigaci on Biom edica HU Puerta de Hierro-Majadahonda. N.Y.C. is supported by the Asociaci on Espa~ nola Contra el C ancer. M.N. and B.H. are supported by the Plan de Empleo Juvenil de la CAM (PEJ-2020-AI/BMD-19527 and PEJ-2020-TL/BMD-19530, respectively). The funders had no role in the study design, data collection, and analysis, the decision to publish, or the preparation of the manuscript. Declaration of Competing Interest None declared. Ethics Approval and Consent to Participate The study was approved by the Ethics Committee of the Hospital Universitario Puerta de Hierro-Majadahonda (PI-67/14). It was conducted in compliance with the principles of the Declaration of Helsinki. All participants gave their signed informed consent for inclusion. Samples were collected, and clinical data were managed following standardized protocols to guarantee the quality of the samples and the confidentiality of donor data. After approval by the corresponding ethics committees, material and data from other centers were anonymously transferred to our laboratory in compliance with the current Spanish legislation (Ley 14/2007 de Investigaci on Biom edica, Real Decreto 1716/2011 and Ley Org anica 3/2018). Supplementary Material The online version contains supplementary material available at https://doi.org/10.1016/j.modpat.2024.100516. References 1. Kanas G, Ge W, Quek RGW, Keeven K, Nersesyan K, Jon E. Arnason. Epidemiology of diffuse large B-cell lymphoma (DLBCL) and follicular lymphoma (FL) in the United States and Western Europe: population-level projections for 2020e2025. Leuk Lymphoma. 2022;63(1):54e63. https://doi.org/10.1016/ j.cell.2016.08.032 2. Casulo C, Byrtek M, Dawson KL, et al. Early relapse of follicular lymphoma after rituximab plus cyclophosphamide, doxorubicin, vincristine, and prednisone defines patients at high risk for death: an analysis from the National LymphoCare Study. J Clin Oncol. 2015;33(23):2516e2522. https://doi.org/ 10.1200/JCO.2014.59.7534 3. Casulo C, Dixon JG, Le-Rademacher J, et al. Validation of POD24 as a robust early clinical end point of poor survival in FL from 5225 patients on 13 clinical trials. Blood. 2022;139(11):1684e1693. https://doi.org/10.1182/blood.2020010263 4. Wagner-Johnston ND, Link BK, Byrtek M, et al. Outcomes of transformed follicular lymphoma in the modern era: a report from the National LymphoCare Study (NLCS). Blood. 2015;126(7):851e857. https://doi.org/10.1182/ blood-2015-01-621375 5. Lossos IS, Gascoyne RD. Transformation of follicular lymphoma. Best Pract Res Clin Haematol. 2011;24(2):147e163. https://doi.org/10.1016/j.beha. 2011.02.006 6. Kridel R, Sehn LH, Gascoyne RD. Can histologic transformation of follicular lymphoma be predicted and prevented? Blood. 2017;130(3):258e266. https://doi.org/10.1182/blood-2017-03-691345 7. Sarkozy C, Trneny M, Xerri L, et al. Risk factors and outcomes for patients with follicular lymphoma who had histologic transformation after response to first-line immunochemotherapy in the PRIMA trial. J Clin Oncol. 2016;34(22):2575e2582. https://doi.org/10.1200/JCO.2015.65.7163 8. Federico M, Caballero Barrig on MD, Marcheselli L, et al. Rituximab and the risk of transformation of follicular lymphoma: a retrospective pooled analysis. Lancet Haematol. 2018;5(8):e359ee367. https://doi.org/10.1016/S23523026(18)30090-5 9. Casulo C, Burack WR, Friedberg JW. Transformed follicular non-Hodgkin lymphoma. Blood. 2015;125(1):40e47. https://doi.org/10.1182/blood-201404-516815 10. Casulo C. Treatment of histologic transformation. Hematol Oncol Clin North Am. 2020;34(4):785e794. https://doi.org/10.1016/j.hoc.2020.03.001 11. Huet S, Sujobert P, Salles G. From genetics to the clinic: a translational perspective on follicular lymphoma. Nat Rev Cancer. 2018;18(4):224e239. https://doi.org/10.1038/nrc.2017.127 12. Casulo C. Follicular lymphoma: is there an optimal way to define risk? Hematology Am Soc Hematol Educ Program. 2021;2021(1):313e319. https:// doi.org/10.1182/blood-2016-05-717355 13. Freeman CL, Kridel R, Moccia AA, et al. Early progression after bendamustinerituximab is associated with high risk of transformation in advanced stage follicular lymphoma. Blood. 2019;134(9):761e764. https://doi.org/10.1182/ blood.2019000258 14. Kridel R, Chan FC, Mottok A, et al. Histological transformation and progression in follicular lymphoma: a clonal evolution study. PLoS Med. 2016;13(12), e1002197. https://doi.org/10.1371/journal.pmed.1002197 15. Gonz alez-Rinc on J, M endez M, G omez S, et al. Unraveling transformation of follicular lymphoma to diffuse large B-cell lymphoma. PLoS ONE. 2019;14(2), e0212813. https://doi.org/10.1371/journal.pone.0212813 16. Eide MB, Liestøl K, Lingjaerde OC, et al. Genomic alterations reveal potential for higher grade transformation in follicular lymphoma and confirm parallel evolution of tumor cell clones. Blood. 2010;116(9):1489e1497. https:// doi.org/10.1182/blood-2010-03-272278 17. Pasqualucci L, Khiabanian H, Fangazio M, et al. Genetics of follicular lymphoma transformation. Cell Rep. 2014;6(1):130e140. https://doi.org/ 10.1016/j.celrep.2013.12.027 18. Okosun J, B€ od€ or C, Wang J, et al. Integrated genomic analysis identifies recurrent mutations and evolution patterns driving the initiation and progression of follicular lymphoma. Nat Genet. 2014;46(2):176e181. https:// doi.org/10.1038/ng.2856 19. Dreval K, Hilton LK, Cruz M, et al. Genetic subdivisions of follicular lymphoma defined by distinct coding and noncoding mutation patterns. Blood. 2023;142(6):561e573. https://doi.org/10.1182/blood.2022018719 20. Glas AM, Knoops L, Delahaye L, et al. Gene-expression and immunohistochemical study of specific T-cell subsets and accessory cell types in the Ismael Fern andez-Miranda et al. / Mod Pathol 37 (2024) 100516 8 transformation and prognosis of follicular lymphoma. J Clin Oncol. 2007;25(4):390e398. https://doi.org/10.1200/JCO.2006.06.1648 21. Fern andez-Miranda I, Pedrosa L, Llanos M, et al. Monitoring of circulating tumor DNA predicts response to treatment and early progression in follicular lymphoma: results of a prospective pilot study. Clin Cancer Res. 2023;29(1): 209e220. https://doi.org/10.1158/1078-0432.CCR-22-1654 22. Pedrosa L, Fern andez-Miranda I, P erez-Callejo D, et al. Proposal and validation of a method to classify genetic subtypes of diffuse large B cell lymphoma. Sci Rep. 2021;11(1):1886. https://doi.org/10.1038/s41598-020-80376-0 23. Delgado J, Pereira A, Villamor N, L opez-Guillermo A, Rozman C. Survival analysis in hematologic malignancies: recommendations for clinicians. Haematologica. 2014;99(9):1410e1420. https://doi.org/10.3324/haematol.2013. 100784 24. Mohammed EA, Naugler C, Far BH. Emerging Business Intelligence Framework for a Clinical Laboratory Through Big Data Analytics. In: Tran QN, Arabnia H, eds. Emerging trends in computational biology, bioinformatics, and systems biology:. systems biology: algorithms and software tools. New York: Elsevier/Morgan Kaufmann; 2015:577e602. 25. Riley RD, Archer L, Snell KIE, et al. Evaluation of clinical prediction models (part 2): how to undertake an external validation study. BMJ. 2024;384, e074820. https://doi.org/10.1136/bmj-2023-074820 26. Balachandran VP, Gonen M, Smith JJ, DeMatteo RP. Nomograms in oncology: more than meets the eye. Lancet Oncol. 2015;16(4):e173ee180. https:// doi.org/10.1016/S1470-2045(14)71116-7 27. Iasonos A, Schrag D, Raj GV, Panageas KS. How to build and interpret a nomogram for cancer prognosis. J Clin Oncol. 2008;26(8):1364e1370. https:// doi.org/10.1200/JCO.2007.12.9791 28. Royston P. Tools for checking calibration of a Cox model in external validation: prediction of population-averaged survival curves based on risk groups. Stata J. 2015;15(1):275e291. https://doi.org/10.1177/1536867X1501500116 29. White IR, Rapsomaniki E, Emerging Risk Factors Collaboration. Covariateadjusted measures of discrimination for survival data. Biom J. 2015;57(4): 592e613. https://doi.org/10.1002/bimj.201400061 30. Morin RD, Mendez-Lago M, Mungall AJ, et al. Frequent mutation of histonemodifying genes in non-Hodgkin lymphoma. Nature. 2011;476(7360): 298e303. https://doi.org/10.1038/nature10351 31. Mozas P, L opez C, Grau M, et al. Genomic landscape of follicular lymphoma across a wide spectrum of clinical behaviors. Hematol Oncol. 2023;41: 631e643. https://doi.org/10.1002/hon.313232 32. van de Schans SAM, Steyerberg EW, Nijziel MR, Creemers GJ, JanssenHeijnen ML, van Spronsen DJ. Validation, revision and extension of the Follicular Lymphoma International Prognostic Index (FLIPI) in a populationbased setting. Ann Oncol. 2009;20(10):1697e1702. https://doi.org/10.1093/ annonc/mdp053 33. Link BK, Maurer MJ, Nowakowski GS, et al. Rates and outcomes of follicular lymphoma transformation in the immunochemotherapy era: a report from the University of Iowa/MayoClinic Specialized Program of Research Excellence Molecular Epidemiology Resource. J Clin Oncol. 2013;31(26): 3272e3278. https://doi.org/10.1200/JCO.2012.48.3990 34. Gin e E, Montoto S, Bosch F, et al. The Follicular Lymphoma International Prognostic Index (FLIPI) and the histological subtype are the most important factors to predict histological transformation in follicular lymphoma. Ann Oncol. 2006;17(10):1539e1545. https://doi.org/10.1093/annonc/mdp053 35. Pastore A, Jurinovic V, Kridel R, et al. Integration of gene mutations in risk prognostication for patients receiving first-line immunochemotherapy for follicular lymphoma: a retrospective analysis of a prospective clinical trial and validation in a population-based registry. Lancet Oncol. 2015;16(9): 1111e1122. https://doi.org/10.1016/S1470-2045(15)00169-2 36. Jurinovic V, Kridel R, Staiger AM, et al. Clinicogenetic risk models predict early progression of follicular lymphoma after first-line immunochemotherapy. Blood. 2016;128(8):1112e1120. https://doi.org/10.1182/blood2016-05-717355 37. Lockmer S, Ren W, Brodtkorb M, et al. M7-FLIPI is not prognostic in follicular lymphoma patients with first-line rituximab chemo-free therapy. Br J Haematol. 2020;188(2):259e267. https://doi.org/10.1111/bjh.16159 38. Yusufova N, Kloetgen A, Teater M, et al. Histone H1 loss drives lymphoma by disrupting 3D chromatin architecture. Nature. 2021;589(7841):299e305. https://doi.org/10.1038/s41586-020-3017-y 39. Zhang J, Dominguez-Sola D, Hussein S, et al. Disruption of KMT2D perturbs germinal center B cell development and promotes lymphomagenesis. Nat Med. 2015;21(10):1190e1198. https://doi.org/10.1038/nm.3940 40. Vlasevska S, Garcia-Ibanez L, Duval R, et al. KMT2D acetylation by CREBBP reveals a cooperative functional interaction at enhancers in normal and malignant germinal center B cells. Proc Natl Acad Sci U S A. 2023;120(11): e2218330120. https://doi.org/10.1073/pnas.2218330120 41. Chapuy B, Stewart C, Dunford AJ, et al. Molecular subtypes of diffuse large B cell lymphoma are associated with distinct pathogenic mechanisms and outcomes. Nat Med. 2018;24(5):679e690. https://doi.org/10.1038/s41591018-0016-8 42. Schmitz R, Wright GW, Huang DW, et al. Genetics and pathogenesis of diffuse large B-cell lymphoma. N Engl J Med. 2018;378(15):1396e1407. https:// doi.org/10.1056/NEJMoa1801445 43. Shui JW, Steinberg MW, Kronenberg M. Regulation of inflammation, autoimmunity, and infection immunity by HVEM-BTLA signaling. J Leukoc Biol. 2011;89(4):517e523. https://doi.org/10.1189/jlb.0910528 44. Boice M, Salloum D, Mourcin F, et al. Loss of the HVEM tumor suppressor in lymphoma and restoration by modified CAR-T cells. Cell. 2016;167(2): 405e418.e13. https://doi.org/10.1016/j.cell.2016.08.032 45. Schmidt J, Gong S, Marafioti T, et al. Genome-wide analysis of pediatric-type follicular lymphoma reveals low genetic complexity and recurrent alterations of TNFRSF14 gene. Blood. 2016;128(8):1101e1111. https://doi.org/10.1182/ blood-2016-03-703819 46. Cheung KJJ, Johnson NA, Affleck JG, et al. Acquired TNFRSF14 mutations in follicular lymphoma are associated with worse prognosis. Cancer Res. 2010;70(22):9166e9174. https://doi.org/10.1158/0008-5472.CAN-10-2460 47. Launay E, Pangault C, Bertrand P, et al. High rate of TNFRSF14 gene alterations related to 1p36 region in de novo follicular lymphoma and impact on prognosis. Leukemia. 2012;26(3):559e562. https://doi.org/10.1038/leu. 2011.266 48. Martin-Guerrero I, Salaverria I, Burkhardt B, et al. Recurrent loss of heterozygosity in 1p36 associated with TNFRSF14 mutations in IRF4 translocation negative pediatric follicular lymphomas. Haematologica. 2013;98(8): 1237e1241. https://doi.org/10.3324/haematol.2012.073916 49. Nann D, Ramis-Zaldivar JE, Müller I, et al. Follicular lymphoma t(14;18)- negative is genetically a heterogeneous disease. Blood Adv. 2020;4(22): 5652e5665. https://doi.org/10.1182/bloodadvances.2020002944 50. Kurz KS, Kalmbach S, Ott M, Staiger AM, Ott G, Horn H. Follicular Lymphoma in the 5th Edition of the WHO eClassification of Haematolymphoid Neoplasms eUpdated Classification and New Biological Data. Cancers (Basel). 2023;15(3):785. https://doi.org/10.3390/cancers15030785 Ismael Fern andez-Miranda et al. / Mod Pathol 37 (2024) 100516 9