TO THE EDITOR: Alterations of the tumor suppressor gene TP53 are frequent in Diffuse Large B-cell Lymphoma (DLBCL), the most commonly diagnosed blood cancer at over 30,000 diagnoses per year, and are associated with poor prognosis [1]. DLBCL has traditionally been divided into three cell of origin (COO) subcategories based on RNA expression profiles or IHC: Activated B-cell (ABC), Germinal Center B-cell (GCB), and Unclassified cases [2]. Patients with ABC tumors are characterized by a more aggressive profile alongside active NF-κB and BCR signaling pathways, while GCB cases are associated with alterations that drive aberrant chromatin-modification, PI3K signaling, and the upregulation MYC and BCL2 through structural variants [3]. Both subtypes are associated with worse outcomes if impaired TP53 is present, although these alterations do not enrich in one subtype vs. the other [1]. Recently, new classification models have utilized previously-identified DLBCL driver DNA alterations to categorize de-novo tumors into subsets [4-7]. All 3 models, supervised or unsupervised, produce a GCB-like classification: EZB (Schmitz/Wright), BCL2 (Lacy), and C3 (Chapuy). These tumors are defined by alterations significantly enriched in aggressive cases of GCB DLBCL: CREBBP, BCL2 (mutation or translocation), EZH2, KMT2D, and TNFRSF14.

Although EZB tumors have an overall favorable prognosis when treated with frontline RCHOP regimens, DLBCL patients that refract therapy or later relapse face a dismal outlook, with only 20% predicted to survive after 5 years, and few studies have examined the molecular landscape of relapsed or refractory DLBCL (rrDLBCL) [8]. The 2019 analysis by Rushton and colleagues analyzed the largest single rrDLBCL cohort, identifying several emergent genetic differences between rrDLBCL and de-novo tumor populations [9]. Importantly, the landmark FDA approval of Chimeric Antigen T-cell (CAR-T) therapy makes characterizing the genomic landscape of rrDLBCL a priority. Herein, we present data supporting that increased TP53 alterations in rrDLBCL are enriched towards GCB-like subclassified cases, such as EZB, and co-associate with related DNA alterations significantly more than they do in de-novo populations.

Data collection and NMF clustering

DNA-sequencing data from the Rushton rrDLBCL analysis (N=127) and 3 de-novo analyses were assembled (N=2,128) [4, 5, 7, 9]. These alterations designated COO via standard gene expression panels, Nanostring, the Hans algorithm, or a combination. Cases within each population were DNA-alteration classified by their respective methods or relied on the LymphGen algorithm. DNA Alterations present in less than 5 cases and patients with less than 2 alterations were filtered out. De-novo patients were filtered out of survival analyses if they were not treated with RCHOP or arose from transformation. DNA alterations included missense, deletion, frameshift, high-impact splice, and truncations. A GCB rrDLBCL validation cohort combining results from four sequencing studies was also assembled (N=54) [10-13]. The rrDLBCL population was analyzed using the Broad Institute’s GenePattern Non-negative Matrix Factorization (NMF) module with assigned cluster values up to 8 [14]. Two groups emerged from this analysis – designated RR1 and RR2. FDR-corrected Marker Selection was applied to uncover differential presence of DNA alterations, with values P<0.05 considered significant. Pearson distance and similarity analyses to TP53 alterations were performed using the Morpheus tool.

Statistical analyses

GraphPad Prism software and GenePattern tools were used to plot and format figures, and analyze data. Analyses were performed by Welch’s t-tests, Holm-Šídák's multiple comparisons-corrected One-way ANOVA tests, and Fisher’s exact tests. TP53 alteration Pearson distances were converted to Z-scores for cross-study comparison. All reported P-values were two-sided and considered statistically significant below 0.05.

TP53 alterations enrich within the GCB-like subset of rrDLBCL patients and associate with EZB alterations

We integrated patient and targeted sequencing panel data from the Rushton analysis into 127 profiles (Supplementary Fig. 1). Unsupervised NMF clustering was applied to the cases, producing the highest cophenetic values when tumors were grouped as 2 clusters (0.9215) (Fig. 1A, Supplementary Fig. 2) (Supplementary Table 1). The 2 clusters were categorized as RR1 (N=58) and RR2 (N=69). Differential Marker Selection revealed that 13 of the 91 gene alterations were significantly enriched within either group (FDR <0.05) (Fig. 1B, Supplementary Tables 2, 3). Patients within the RR1 family were associated with MYD88, PIM1, IGLL5, HIST1HC, SOCS1, HIST1H1E, and CD79B alterations. Patients within the RR2 family were associated with CREBBP, EZH2, STAT6, BCL2, TNFRSF14, and TP53 alterations. Double/Triple-hit structural variations were also associated with RR2 family tumors (P=0.0391) (Supplementary Fig. 3). RR1 was heterogenous in its composition, harboring MCD/ABC or BN2/Unclassified tumors, while RR2 was more homogenous, composed primarily of EZB/GCB classified tumors (Supplementary Fig. 4).

Figure 1. TP53 alterations enrich within GCB-like DNA subclassification alterations and rrDLBCL patients compared to de-novo cohorts. (A) K2 (2-cluster) NMF clustering. rrDLBCL patients (N=127) were analyzed for the best fit when measuring the association patterns of DNA alterations (N=91). Patient similarity is designated by color, with red representing the most co-association and blue the least. The RR1 and RR2 subsets that emerged from clustering are designated by light grey and dark grey coloring, respectively. (B) A volcano plot displays differentially enriched DNA alterations between RR1 and RR2. Comparative marker selection between the groups resulted in 2-sided T and FDR values. Greater T values were associated with RR1 and lesser values with RR2. The dotted line represents the 0.05 FDR threshold to be met for significant association with one family over the other. Significant alterations are color coded for their corresponding LymphGen cluster, if designated. (C) TP53 alterations are significantly enriched towards EZB tumors in rrDLBCL but not in de-novo DLBCL. Stacked bar graphs denote the presence of TP53 alterations within EZB and non-EZB tumors. The pre-treatment Schmitz et al. 2018 [3] cohort is compared to the Rushton et al. [9] rrDLBCL cohort. Significance was determined with a Fisher’s Exact test within both groups. (D) TP53 Alterations significantly co-occur with EZB alterations and significantly occlude MCD alterations. A volcano plot displays TP53 Pearson distance for measured rrDLBCL genes. FDR-corrected correlation similarity values are plotted on the Y-axis, with FDR <0.05 noted with a dotted line. Genes are labelled and noted for LymphGen subclassification. (E) RR2 driver genes increased association with TP53 alterations in comparison to pre-treatment association measurements. Z-score normalized Pearson Distance values are plotted on the Y-axis against RR1 and RR2 genes on the X-axis. rrDLBCL associations (Rushton) are compared to 3 separate de-novo cohort values (Lacy, Reddy, and Wright) based on TP53 co-association. Two-way ANOVA analysis was used to measure significance between pre-treatment and rrDLBCL values after Bonferroni multiple comparison correction.

TP53 alterations were significantly enriched within EZB-designated rrDLBCL cases compared to de-novo EZB cases (P=0.0018) (Fig. 1C). TNFRSF14, EZH2, CREBBP, BCL2, and KMT2D alterations significantly co-occurred in tumors bearing altered TP53 (Fig. 1D) after Pearson similarity matrix analysis (Supplementary Fig. 5, 6, Supplementary Table 4). In contrast, CD79B, PIM1, MYD88, GRHPR, and SOCS1 alterations were significantly exclusionary of TP53. Z-scored Pearson distance values from TP53 also displayed collective shifts away from RR1 genes (ABC-like) and trends towards RR2 genes (GCB-like) when compared individually (Fig. 1E) and collectively to de-novo levels (Supplementary Fig. 7, Supplementary Table 5). Specifically, rrDLBCL TP53 associations significantly associated with EZH2 (P=0.0156) and TNFRSRF14 (P=0.0073).

TP53 rrDLBCL enriches towards EZB/GCB-like alterations and cases in multiple rrDLBCL cohorts and is associated with inferior RCHOP response within EZB and BCL2 subsets

We next isolated EZB-associated genes to compare TP53-impairment differences between de-novo analyses and rrDLBCL cases. Collective EZB Z-score association with TP53 alterations was significantly greater in the rrDLBCL cohort compared to de-novo cohorts (P=0.0166) (Fig. 2A). TP53 co-associations were next compared across all LymphGen-subsets. EZB genes harbored a significantly greater TP53 co-association than genes associated with the MCD (P=0.0026), BN2, (P=0.0237) or unclassified (P=0.0017) classifications in rrDLBCL (Fig. 2B). This significant trend was observed once more within a second population of rrDLBCL patients (N=44) (Fig. 2C) [10]. We added cases to this rrDLBCL validation population, integrating Greenawalt, Morin (N=25), Juskevicius (N=21), and Jain (N=24) (Supplementary Table 6) [10-13]. Significant enrichment of TP53 alterations within GCB cases were noted for both rrDLBCL populations (Fig. 2D, Supplementary Fig. 8). As a final measure of TP53-impariment’s role driving refractory cases of GCB-like subclassified tumors, Kaplan-Meier survival analyses revealed significantly inferior patient survival within both when TP53 alterations are present before RCHOP treatment (Fig. 2E). These data inform that the detrimental role of TP53 impairment at diagnosis rises within GCB-related subclassifications of rrDLBCL.

Figure 2. TP53 rrDLBCL enriches towards GCB-like subclassification alterations and cases in multiple rrDLBCL cohorts and is associated with inferior RCHOP response within de-novo EZB and BCL2 Subset patients. (A) A dot plot summarizes the Z-score normalized TP53 alteration Pearson Distance values of EZB classifier genes in the rrDLBCL population vs. combined de-novo DLBCL cohort values. A Welch’s t-test was used to measure significance. Mean±SD is designated by error bars. (B) TP53 alterations are significantly associated with EZB alterations in rrDLBCL. TP53 Pearson distance values are displayed for 90 rrDLBCL cohort genes (Z-score normalized). One-way ANOVA analysis was used to measure significance and was corrected by a Holm-Šídák's multiple comparisons test. Gene color is matched to LymphGen subclassification. Mean±SD is designated by error bars. (C) Gene Pearson Distance from TP53 Alterations within the Greenawalt rrDLBCL cohort (N=44). A dot plot summarizes the Z-score normalized TP53 alteration Pearson Distance values of LymphGen classifier genes of each subset. One-way ANOVA analysis was used to measure significance and was corrected by a Holm-Šídák's multiple comparisons test. Gene color is matched to LymphGen subclassification. Mean±SD is designated by error bars. Alterations present in the cohort but not included in LymphGen classification are displayed in white. (D) Percentage of EZB (or corresponding) cases with non-altered (WT) vs. altered TP53 in DLBCL vs. rrDLBCL populations. A gradient denotes percent composition of each population. Significance was determined through a Fisher’s Exact test between TP53 alteration enrichment within EZB or GCB cases vs. Non-GCB cases. The number of GCB/corresponding cases examined is listed on the right. (E) The presence of TP53 alterations at diagnosis in the GCB-like subset EZB (Wright et al. [5]) and BCL2 (Lacy et al. [7]) patients is associated with inferior overall survival. Kaplan-Meier survival curves assess the overall survival impact of TP53 alteration within EZB or BCL2 populations when treated with RCHOP. Significance was measured with Logrank analysis. The 95% confidence interval for each alteration is designated by dotted lines.

Impairment of the TP53 tumor suppressor is one of most negative prognostic indicators of DLBCL, strongly associating with inferior survival and aggressive disease biology. However, DLBCL is a heterogenous disease, often the result of cumulative epistatic genetic alterations instead of a single driver. Further, recent studies have identified associated sets of genetic alterations that co-segregate into distinctive subsets, supplementing traditional disease gene expression models [4-7]. Importantly, TP53 alterations have not found a consensus position among these subclassifications.

When focusing rrDLBCL tumors, we identified that alterations of TP53 enrich towards GCB-like subclassifications (EZB/C3/BCL2) compared to frequencies observed in de-novo populations, significantly co-occurring with EZB-classified genes as well. Our analysis relied on established, unbiased methods of clustering: the Broad Institute’s NMF tool, notably responsible for the Chapuy 2018 analysis that led to the C1-C5 subsets [6]. A disadvantage is that certain cases will not find a strong consensus when few alterations are present, but we preferred an unbiased approach in favor of seed-based models. We hypothesize that TP53 impairment may be selected for after RCHOP treatment in a portion of EZB/corresponding cases that otherwise would not be able to reject therapy, joining rrDLBCL cases that originally presented with impaired TP53 at diagnosis. Similarly, EZH2 was among the EZB genes strongly associated with TP53 impairment in our analysis, and the Reddy analysis designated GCB+EZH2 cases as one of their low-risk classifications, indicating that a robust loss of tumor suppression may be necessary to observe progression [4]. Similarly, strong co-associations and enrichment between TP53 and CREBBP alterations may hint at cooperation between the resulting pathways during transition.

Future studies should address specific caveats encountered in our report. To begin, validation among additional rrDLBCL patient cohorts remains an important follow up for supporting these results. Equally, the lack of CNA data may preclude these cases forming their own A53/C2 cluster, but the steep overall rise in rrDLBCL TP53 alterations should be noted as a critical factor to consider. Our secondary rrDLBCL cohort indicated that increased TP53 alterations remain favored to enrich in GCB-designated cases, but larger populations and sequencing studies are warranted to identify de-novo EZB/corresponding patients most likely to develop TP53-induced relapse. Lastly, although KMT2D alterations were enriched and clonally stable after the rrDLBCL transition (and typically enrich in de-novo EZB/C3/BCL2 cases), the Rushton analysis did not highlight if this rise was restricted to the GCB-like subsets. Our analysis did not observe significant association with GCB-like alterations in rrDLBCL, although they did trend towards them (Supplementary Table 2). Heavy overall enrichment (64/127 patients) may have proven difficult for KMT2D alterations to find a consensus association with other genes. This could also indicate that their role in rrDLBCL biology is unique.

In conclusion, we report the enrichment of TP53 alterations towards EZB/corresponding rrDLBCL tumors (and their designate alterations) during disease transition. The strong EZB/corresponding profile of our RR2 group and consequent enrichment of TP53 alterations highlights the importance of DNA classification models and adds clarity to patterns of rrDLBCL evolution. The rising importance of liquid biopsy monitoring in rrDLBCL patients is made all the greater by the growing indication that TP53 impairment may predict poor response to CAR-T therapy due to the introduction of immunosuppressive pathways, specifically discussed by Shouval and colleagues at the 2021 American Society of Hematology conference [15]. TP53 mutation status, in combination with other prognostic factors, may therefore be used to identify high-risk patients prior to CAR-T treatment.

This study was supported by the Minnesota State University, Mankato College of Science Engineering and Technology through startup funds, facilities, and resources. Dr. Javeed Iqbal and Dr. Alyssa Bouska from the University of Nebraska Medical Center kindly offered their support and counsel. Author Contributions are as follows: S.L., V.N., K.W., J.G., A.R., and K.T.H. performed investigations and analyses. K.T.H. and S.L. were responsible for revision. K.T.H. was responsible for conceptualization, data curation, formal analysis, funding acquisition, writing, and supervision.

No potential conflicts of interest relevant to this article were reported.