Designing Evolutionary-based Interception Strategies to Block the Transition from Precursor Phases to Multiple Myeloma

Many blood cancers are associated with premalignant clonal expansions, the prevalence of which increases with age (1, 2). These entities include monoclonal B lymphocytosis (MBL), a precursor of chronic lymphocytic leukemia; clonal hematopoiesis, a precursor of acute leukemia; and importantly monoclonal gammopathy of uncertain significance (MGUS) and smoldering multiple myeloma (SMM), precursors of multiple myeloma (Fig. 1; refs. 2–6). Clinical management of precancer is becoming increasingly important and should be based on sound biological understandings. A unifying feature of all of these states is their heterogeneity at a mutational level comprising a variable set of gene mutations acquired by apparently stochastic processes that drive changes in biologic behavior. In this model, multiple clonal propagating units compete for survival and clonal dominance resulting in either their suppression or selection and expansion over time leading to clonal diversification and metastatic seeding (7, 8). The model is credible because it can explain mutational differences at different sites (spatial variation), the acquisition or selection of mutations at relapse (temporal variation), and the development of resistance following treatment. Thus, before invasive cancer develops, many premalignant clones have been present for years within apparently normal cell populations. Importantly, clinical intervention in this preexpansion phase before the full repertoire of drivers is acquired, is likely to offer the greatest potential to eradicate the clonal cells and result in long-term disease-free survival and potentially cure.

The life history of multiple myeloma is characterized by premalignant clinical phases (i.e., MGUS and SMM), which are recognized by the presence of a monoclonal immunoglobulin produced by the premalignant clone (6, 7, 9, 10). According to historical models, the premalignant clone is immortalized by a limited number of key initiating genetic events in the germinal center (7, 9) and then homes to the bone marrow, where it evolves, predominantly below the level of clinical detection until identified as either MGUS or SMM (11, 12). Thus, before multiple myeloma presents, many premalignant clones have been present for decades and, within this ecosystem, multiple genetically diverse multiple myeloma–propagating units compete for survival. The clinical features of multiple myeloma reflect the end result of evolutionary processes occurring as a result of the genetic diversification of the clone over time (13–16). While originally it was proposed that evolution proceeded steadily overtime, more recently it has been appreciated that it may also be driven by catastrophic events were multiple genes are deregulated simultaneously (i.e., “punctuated evolution”; refs. 17, 18), which can also result in rapid changes in behavior (5, 19, 20). Intercepting these processes to prevent transition to cancer has clinical relevance in multiple myeloma because 5% of the over 60-year-old population have a detectable monoclonal immunoglobulin that transforms to multiple myeloma at 1%–10% per annum a malignancy (5, 19).

The use of metaphase cytogenetic analysis, Southern blotting, fluorescence in situ hybridization (FISH), and gene expression analysis (7, 9) identified two main initiating events: multiple trisomy's of the odd numbered chromosomes (i.e., hyperdiploid) and recurrent chromosomal translocations involving the IGH locus that overexpressed key oncogenes including CCND1, CCND2, CCND3, MMSET, MAF, and MAFB (Supplementary Fig. S1; refs. 21–23). These events were identified as being etiologic because they are present in all tumor cells in all disease phases (24, 25). The formation of these translocations is a consequence of activation-induced cytidine deaminase (AID) activity in the germinal center, introducing DNA double-strand breaks (DBS) in the IGH locus, which are then joined to another DBS occurring elsewhere in the genome. Distinct IGH translocations and recurrent aneuploidies have been linked to poor outcome [e.g., t(4;14)(MMSET;IGH), amp 1q21, and del17p13 (TP53)] and integrated in clinical/biological prognostic scoring systems (26). The events on 1q21 and 17p were well-know before SNP arrays, but clearly arrays and next-generation sequencing have improved our understanding of their prevalence, evolution, and the underlying mechanisms involved in their acquisition (Supplementary Fig. S1; refs. 26–28). While the first 1q21 gain is often an early event often occurring together with other large trisomies, it is clear that jumping translocations and additional amplifications of 1q21 and del17p13 are often late events (17, 28).

The initial genetic analyses of multiple myeloma have allowed us to make a number of important conclusions about the contribution of genetic changes to disease progression. While superficially, the acquired genetic features of multiple myeloma appear complex, it is only a limited number of these events that are involved in “driving” the malignant process. Exome sequencing focused attention on approximately 80 recurrent “driver genes” mainly impacting RAS/MAPK and NF-κB signaling, with some being prognostically important (13, 14, 16, 17, 29–34). Despite a significant fraction of these mutations in driver genes being selected, they are often subclonal suggesting that, they are late events occurring after other earlier clonal events (32, 35).

To fully understand the development of multiple myeloma, it is critical to realize that most sequencing data have been generated from cases of newly diagnosed multiple myeloma and that the genetic make-up of precursor phases is largely unknown. A number of different datasets including SNP array, and exome and targeted sequencing data revealed that the frequency of recurrent driver mutations, copy-number changes, and structural events typical of multiple myeloma occur at much lower frequencies in myeloma precursor disease (6, 14, 36–38). Furthermore, the prognostic and clinical impact of any given lesion may be different, that is, RAS mutations lack prognostic importance in newly diagnosed multiple myeloma but in SMM are emerging as a strong prognostic factor of transition to multiple myeloma (39, 40).

While exome sequencing has provided most data, it does not fully capture structural variants (SV) and copy-number abnormalities (CNA) that occur in the noncoding sequences, which since the advent of whole-genome sequencing (WGS) have been shown to be pathologically crucial (6, 17, 18, 41). Using CoMMpass, more than a 100 chromosomal regions of recurrent gain or loss involving amplified or deleted genes have been identified, further expanding the compendium of known disease drivers (Supplementary Fig. S1; ref. 18).

The SVs burden in multiple myeloma is greater than chronic lymphocytic leukemia and acute myeloid leukemia, and is in the same order as B-cell non-Hodgkin lymphoma, but is considerably less than in most solid cancers (17, 42, 43). Cataloguing SVs in multiple myeloma, WGS identified three main complex SV events: chromothripsis, chromoplexy, and templated insertions (Fig. 2; refs. 17, 18, 44). Chromothripsis occurs in a single catastrophic event where one or more chromosomes are shattered and randomly rejoined, leading to a pattern of oscillating copy-number changes and localized clustering of breakpoints. This event was initially thought to be rare, but using WGS it can now be detected in 20%–30% of multiple myelomas and is associated with adverse survival (18, 45). Chromoplexy is a complex chained event detectable in 11% of patients on the basis of concatenated rearrangements between different chromosomes causing copy-number losses (17, 18, 44). A type of complex rearrangement termed “templated insertions” is associated with multiple focal gains and is seen in 20% of multiple myelomas, a prevalence that is higher than other cancers (17, 18, 44). Templated insertions involve the immunoglobulin loci in 34% of cases and lead to the focal amplification of key drivers and immune therapy targets such as MYC, CCND1, CCND2, TNFRSF17 (BCMA), SLAMF7, and KLF2 (Fig. 2; ref. 18). Overall, from a structural perspective, the genomic profile of multiple myeloma is more stable than some other cancers, lacking evidence of the recently proposed complex events: rigma, thyphonas, or double minute events (18, 46). These complex SVs are particularly important because they potentially involve multiple drivers simultaneously, suggesting that the full multiple myeloma driver repertoire can be acquired in only few events. This concept is particularly important in the precursor settings and in its monitoring, where it could explain rapid changes in clinical behavior of these entities.

The median number of mutations per multiple myeloma exome and genome is 149 (range, 3–6,186) and 5,885 (range, 1,511–15,884), respectively, and lies at the median of the genetically simpler acute leukemias and more complex tumors such as melanoma (Fig. 3A; refs. 13, 14, 16, 17, 25, 32, 33, 35, 41, 47). The mutational catalogue of each cancer represents the sum of the activity of multiple different mutational processes over time (25, 48). Understanding the processes that drive these mutations active in early-disease stages points the way toward active therapeutic intervention able to selectively influence clonal competition to suppress aggressive subclones. Signatures or “mutographs” of mutational prosses are based on 96-mutational classes defined by the base on 5′ and 3′ side of each mutated nucleotide (i.e., six possible single-nucleotide variants × 16 possible trinucleotide context) being readily generated for each tumor. Mutations in newly diagnosed multiple myeloma are the result of seven main mutational processes (Fig. 3B; ref. 41). APOBEC deaminase signatures [single-base substitution (SBS2 and SBS13)] are seen in a significant fraction of multiple myeloma (14, 25, 29, 33, 35, 49). A high mutational burden with the features of APOBEC is associated with t(14;16) in MAF translocations, chromothripsis, biallelic TP53, and poor outcome (17, 18, 33). Two “clock-like” signatures have been identified, SBS1 and SBS5, that are characterized by constant rate over time in both normal and tumor cells (50). Three additional mutational processes, SBS8, SBS9, and SBS18, have recently been described in newly diagnosed multiple myeloma, but their true relevance is unclear (14, 25, 41, 49). SBS8 has been seen in breast and ovarian where it has been associated with BRCA1 and BRCA2 deficiency (51). However, in multiple myeloma there has been no suggestion of a BRCA-null phenotype, suggesting a different etiology (25). SBS18 is seen in a small fraction of multiple myeloma and is linked to free radical oxygen stress (48). Of key importance to multiple myeloma progression is SBS9, termed noncanonical AID to differentiate it from the canonical AID signature (SBS84), and is directly related to the germinal center, somatic hyper mutation, and AID exposure (25, 48, 52–54) The SBS9 mutational process does not have the typical feature of AID-mediated somatic hypermutation (C to T/G mutation at WRCY motifs, W = A or T, R = purine, and Y = pyrimidine), but is enriched in T>G and T>C consistent with mismatch repair pathway repair of AID-related mutations. In contrast, canonical AID (SBS84) is mostly involved in somatic hypermutation or kataegis on distinct targets (e.g., IGH/IGK/IGL loci). While some of these canonical AID mutations are likely to be passenger events (e.g., intronic BCL6 mutations), others show some evidence of positive selection (25). Both AID mutational activities occurred are early, clonal, and usually precede the gains of odd chromosomes leading to hyperdiploidy.

At relapse, two mutational signatures have been recently reported and involved in increasing the number of nonsynonymous mutations: (i) SBS35 and (ii) a new mutational signature, named SBS-MM1, as a consequence of exposure to the mutagenic activity of platinum and melphalan, respectively (Fig. 3B; refs. 41, 55).

Computational tools can deconvolute these temporally related events allowing the reconstruction of the life history of multiple myeloma and an ability to build an encompassing disease model. Different approaches have been used to estimate the timepoint at which driver events are acquired during cancer life history. The most common uses the cancer cell fraction (CCF; i.e., number of cells carrying a distinct alteration to define early and late timepoints based on their subclonal percentage; refs. 17, 32). Applying this to multiple myeloma shows that the hyperdiploid (HRD) trisomies and 1q21 gain have high CCF and are usually clonal, consistent with them being early events. In contrast, the recurrent deletions (e.g., del1p, del14q, and del16q) and nonsynonymous mutations have a heterogeneous distribution, consistent with them occurring later in the disease process.

Greater insights into the chronologic order of driver acquisition can be gained by reconstructing the phylogenetic tree (17). The trunk of the tree being composed of clonal events acquired before the events identified in the branches (i.e., subclonal; Supplementary Fig. S2). Phylogenetic tree reconstruction is based on few simple rules. If in one sample the fraction of events A and B within the cancer are present at 75% and 40%, respectively, then according to the “pigeonhole principle” (56), there must be at least one tumor cell that contains both mutations, because the existence of two independent subclones with a cumulative CCF > 100% is not possible (Supplementary Fig. S2A and S2B). Conversely, if B were present at a cell fraction of 20% this would be compatible with either the presence of independent cluster or with a clone carrying both mutations (Supplementary Fig. S2C). The resolution provided by the phylogenetic tree can be improved by the inclusion of multiple samples from the same individual collected at different timepoints or at different disease sites (17, 55, 57). Using this information, it is possible to reconstruct the chronologic order of driver events and define patient-specific “evolutionary trajectories” based on pairwise precedence. By taking the relative order of drivers in all patients, it is possible to define the order in which genes are mutated relative to one another (17, 41). This approach has shown that more than 50% of all mutations in multiple myeloma occur as early events.

To refine temporal estimations, further large chromosomal duplications can be used on the basis of the fact that when an allele is duplicated all the genomic events acquired until that point are also duplicated (17, 58). Thus, while their CCF will be always 100% the variant allele frequency (VAF) of the duplicated mutations will change from 50% to 66%, being present on two of the three alleles. Conversely, clonal mutations acquired either before or after the gain on the minor allele and after the gain on the duplicated alleles will have a VAF of 33%. This technique has shown that the final karyotype, in a significant fraction of hyperdiploid patients, reflects the sum of multiple independent events separated by time (17) By combining molecular time and CCF assessment for the timing of molecular events it is possible to define an evolutionary trajectory based on driver events that occur in the early (pregain), intermediate (postgain and clonal), and late (subclonal) evolutionary phases of disease.

Despite their critical importance, SVs have not been routinely used to reconstruct cancer evolutionary trees. To test the impact of SV on subclonal evolution, two main approaches can be used. The first links each SV to its corresponding CNA and uses the clonality of the latter. This approach cannot be used all the time because not all SVs have copy change for accurate CCF estimation. The second approach requires multiple samples collected at different timepoints. In this analysis, conserved clonal events are acquired before unique selected events and a chronologic hierarchy is recreated. This analysis showed, as expected, that etiologic IGH translocations are clonal. Chromothripsis was demonstrated to be an early event, and was detected in the majority of clonal cells as were templated insertions. A fraction of complex events involves MYC as a late-progression event. Chromoplexy is usually a late event and occurs at relapse.

Applying this classification, IGH translocations, copy-number gains, AID-mediated mutations on driver genes, and chromothripsis emerged as early events and detectable at the early precursor stage of disease. Templated insertions are also usually clonal and detectable at the early-disease stage. However, a fraction of these complex events involves MYC, a known driver of progression of SMM. In contrast, the distribution of chromoplexy, focal and large chromosomal deletions, and non-AID mutations on driver genes suggest they are usually acquired during late phases of cancer development and at relapse.

The same workflow has been used to define the timeline of each mutational process in multiple myeloma and two main time windows were observed (Fig. 4; refs. 14, 41, 49). The first is characterized by high AID and low/absent APOBEC, SBS18, and SBS8 mutational activity. During this phase, the premalignant clone is chronically exposed to AID and the germinal center activity leading to the acquisition of key drivers. Different from what has historically been believed, these mutations are not acquired in one single passage through the germinal center, but are more consistent with evidence suggesting a prolonged exposure to AID and the germinal center of the premalignant clone. This chronic and prolonged AID activity is consistent with the recently proposed model wherein antigen stimulation results in clonal plasma cell expansion and M spike increase in mice (59, 60).

The second time window is at a phase when the premalignant cell is independent of the germinal center–mediated AID mutational activity and at this stage, SBS8, SBS18, and APOBEC play major roles in ongoing mutation acquisition. In particular, APOBEC has been shown to play a critical role in the acquisition of nonsynonymous mutations and in the subclonal diversification after the emergence of the most common recent ancestor (35, 41).This model does not apply only in the MAF/MAFB group that has a high APOBEC mutational burden (29, 33). Preliminary investigations suggest that in these patients the APOBEC signature has been active from the earliest disease phases, while AID and the germinal center contribution are barely detectable. In these patients, the APOBEC mutational activity shows the same profile of APOBEC-hypermutated solid cancers, with a major APOBEC3A/3B ratio compared with all other samples (41, 48, 61).

These findings are important for intervention strategies with APOBEC emerging as a strong marker for risk of SMM progression (40), in redefining high-risk (29, 33) multiple myeloma, and in understanding clonal evolution (41). Importantly, APOBEC is known to increase nonsynonymous mutational burden, potentially introducing new drivers and enhancing clonal progression (41); APOBEC, in particular APOBEC3B, has emerged as the most important and dominant mutational process in a subset of multiple myeloma characterized by MAF and MAFB translocations (41).

Functional genomics can provide additional key insights into molecular targets that if manipulated could exert a potential therapeutic impact. If because of subclonal heterogeneity mutational-targeted therapy is flawed, then biological-based targeting aimed at essential cell functions offers a way forward. Furthermore, functional genomics can identify critical biological nodes that can be targeted. A number of functional screens have been performed in multiple myeloma (62), but the approach has potential limitations because of introducing the guide sequences into a plasma cell background and because it is based on myeloma cell lines whose genomic and transcriptomic background does not properly reflect the one observed in primary samples from patients diagnosed with multiple myeloma. Despite this potential drawback, this approach has successfully identified targets that can overcome therapeutic resistance for immunomodulatory agents (IMiD) drugs and proteasome inhibitors (63, 64). In the future, a closer integration of genomic and function study will be critical to identify key drivers and therapeutic targets.

The pathway to multiple myeloma is not direct and involves a long precursor phase where off-target effects of the germinal center and AID are emerging as crucial for the first phase of cancer development. Dissecting the molecular processes active in the early natural history of multiple myeloma can allow us to build credible models of the disease that could build alternate therapeutic strategies appropriate for premalignant disease states.

During normal plasma cell ontology, B cells with high-affinity receptors exit the germinal center as either memory B cells or as plasma cells, which home to the bone marrow where they either produce antibody or become long-lived with appropriate survival signals from the bone marrow niches (65). Thus, B-cell maturation to a plasma cell is characterized by the potential for the acquisition of hundreds of mutations and SVs as a consequence of AID-mediated DNA DSBs, and by chance alone cells carrying these may be immortalized and have the potential to develop into malignancy. The observation of early and prolonged AID activity provides further evidence for the importance of the germinal center and implicates a cell of origin with the ability to reenter a germinal center; potentially one with the features of a memory B cell. The presence of this signature also implies a pre-MGUS phase before the full features of a bone marrow plasma cell are developed and during which time subclonal diversity and selections are enhanced by exposure to the germinal center mutator mechanism (Fig. 5).

After initiation and homing to the marrow, the plasma cell grows in its bone marrow niche where both intrinsic (i.e., cancer genomics) and extrinsic (i.e., microenvironment) factors influence the rate progression. These factors may vary over time depending upon the nature of acquired mutations, but at this point there is little evidence of acquired genetic instability in multiple myeloma. Despite multiple myeloma is characterized by a higher cytogenetic complexity compared with other hematologic malignancies, recent data showed that many of these aneuploidies are acquired in single catastrophic events (i.e., complex SVs), suggesting that that the full driver landscape of multiple myeloma can be acquired in few events (17, 18). The impact of different spontaneous mutational processes can be seen at the SMM–multiple myeloma transition, which is comprised of at least three distinct patterns of transition to multiple myeloma, a stable MGUS type disease, multiple myeloma in the process of transition, or an indolent subtype of multiple myeloma (5, 6). The speed of this transition is also variable, some occurring rapidly, while others occur slowly. Rapid changes may reflect punctuated evolution where a catastrophic event such as chromothripsis may deregulate more than one driver simultaneously (Fig. 2).

Extrinsic factors take account of the natural selection acting on the clonal cells within their microenvironment (66–69). Such factors include the availability of resources, access to the multiple myeloma niche, and the impact of the immune microenvironment.

On the basis of the evolutionary consideration we have outlined, the use of the same treatment strategies for premalignant disease states as for presenting multiple myeloma may be inappropriate (70). While using combinations of drugs with multiple modes of action has been successful at partially overcoming the diversity of newly diagnosed multiple myeloma, the potential for generating resistance, side-effects, and off-target effects on normal tissue makes it difficult to justify the exposure of these patient to such risks in premalignant states. Such considerations are clearly relevant with the known mutagenic activity of melphalan and platinum, which has been demonstrated to increase the mutational burden after treatment (41, 55, 71). Furthermore, the long-term impact of cytotoxic chemotherapy on nontumor cells can result in prolonged immunosuppression, which, in turn can facilitate multiple myeloma genomic accelerated evolution and progression. These emerging data suggest that chemotherapy should be actively discouraged in the precursor settings (41, 55, 71). The evolutionary models of multiple myeloma progression provide an alternate framework for considering the treatment of early cancers by modifying the evolutionary trajectory of the disease. For a successful interception approach, the assessment of disease will need to be changed, to take account of both intrinsic (i.e., tumor cell) and extrinsic (i.e., microenvironment) factors. This will require the development of evolutionary classifications that capture both the genetics of the cancer as well as an assessment of the immune response to it. Using the combination of both these sets of data it will be possible to define the likelihood of progression and to personalize therapeutic decision-making.

The role of genetic assessment is supported by recent work suggesting that mutations in early-disease stages (i.e., precursor disease) may have different clinical meaning to those seen in patients who fulfill the diagnostic criteria for multiple myeloma. A key example of this is RAS mutations, which are predictive of a higher risk of SMM transformation in multiple myeloma, but have no clear prognostic value in newly diagnosed multiple myeloma (40). Other emerging genetic features, such as APOBEC, have a clear prognostic role both in the precursors settings and in the newly diagnosed multiple myeloma (29, 33, 40). The development of new single-cell technologies, able to more effectively interrogate the immune microenvironment, will enhance the potential to fully determine the contribution of this to disease progression and to the likelihood of responding to immune therapies (69).

We believe that interception strategies will become critical features of cancer management in the future as the number of potential interventions increases. Importantly, one of the major goals of interception at a time before cancer symptomatic progression is the potential for cure. Targeted therapies aimed at key biological targets have good characteristics for clinical application in early-disease stages, but mutationally targeted therapy unless directed at initiating mutations such as t(11; 14) (CCND1;IGH) multiple myeloma, which are present in all clonal cells, will require novel application approaches to overcome the subclonal nature of mutation (72).

For many reasons, immune interventions are preferable for early precursor stages of disease and the use of vaccine-based strategies offers a nontoxic and potentially ideal intervention especially if monitored by the stability of subclonal structure and immune response. However, clinical proof for the efficacy of such an approach is currently lacking. In contrast, there is evidence that other immune-based approaches could prove useful in this setting. IMiD drugs either alone or in combination with anti-CD38 therapeutic are potentially relevant and a number of clinical trials have shown them to be relatively safe and potentially beneficial, but their potential to cause resistant disease at future timepoints is currently unknown (6). More active interventions such as the bispecific anti-BCMA antibodies used to direct T and natural killer cells or chimeric antigen receptor T cells offer a clear way forward. These agents are associated with dramatically increased response rates over prior agents and may deliver significant advances for interception if acute adverse reactions such as cytokine release syndrome can be overcome. What is clear for all such interventions is that effective clinical trial evaluation will be required. For this to be practical, new endpoints will have to be developed to give readouts of efficacy in a reasonable time period. Such endpoints are likely to be based on achieving deep responses where disease cannot be detected, but in this setting there will be a strong need to take account of the impact of the intervention on the immune microenvironment.

O. Landgren reports receiving grants from Amgen (research funding for investigator-initiated trial), Janssen (research funding for investigator-initiated trial), Takeda (research funding for investigator-initiated trial), LLS/Rising Tide Foundation (research funding), FDA (project grant), and CDC (project grant), personal fees from Genentech (advisory board) and Cellectis (advisory board), and other from Takeda (IDMC member), Janssen (IDMC member), and Theradex (IDMC member) outside the submitted work. G. J. Morgan reports receiving personal fees from Bristol-Myers Squibb, Karyopharm, Sanofi, GlaxoSmithKline, Genentech, and Roche during the conduct of the study. No disclosures were reported by the other author.

This study was supported by the MMRF and Perelman Foundation. F. Maura and O. Landgren were supported by the Memorial Sloan Kettering Cancer Center NCI Core grant (P30 CA 008748). F. Maura was supported by the American Society of Hematology, the International Myeloma Foundation, and The Society of Memorial Sloan Kettering Cancer Center. G.J. Morgan was supported by Leukemia and Lymphoma Society.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.