Dynamics of minimal residual disease and its clinical implications in multiple myeloma: A retrospective real-life analysis

Abstract

Background

Minimal residual disease (MRD) testing is a promising approach to tailor the treatment of multiple myeloma (MM). However, several major concerns remain to be addressed before moving it into daily practice, most of which stem from the dynamic nature of the MRD status. Thus, it is crucial to understand the MRD dynamics and propose its clinical implications.

Methods

We retrospectively analysed the data of patients with newly diagnosed MM (NDMM) who had flow cytometry-based MRD tests at multiple time points after initiation of therapy. The impact of undetectable MRD (including attainment, duration and loss) on clinical outcomes was analysed.

Results

In a cohort of 220 patients with NDMM, attainment of MRD⁻ offered favourable outcomes (P < 0.0001 for both progression-free survival (PFS) and overall survival (OS)), regardless of baseline risk factors. Notably, MRD⁻ duration ≥12 months was associated with an 83 % (95 % confidence interval (CI), 0.09−0.34; P < 0.0001) or 69 % (95 % CI, 0.13−0.76; P = 0.0098) reduction in risk of progression/death or death, while the longer MRD⁻ was sustained, the better the outcome was. Loss of MRD⁻ led to poor PFS (hazard ratio (HR) 0.01, 95 % CI 0–0.06, P < 0.0001) and OS (HR 0.03, 95 % CI 0–0.24, P = 0.0008). Most patients (70 %) who lost MRD⁻ status carried high-risk cytogenetic abnormalities (HRCAs). While MRD⁻ was temporally inconsistent with conventional therapeutic responses (eg ≥ complete remission or very good partial response), it predicted disease progression or recurrence more robustly than the latter. Last, the predictive value of the MRD status was independent of baseline risk factors (eg high-risk cytogenetic abnormality, International Staging System (ISS) or Revised (R-)ISS staging).

Conclusions

Longitudinal assessment of MRD during the treatment course and follow-up is required for monitoring disease progression or relapse, to guide treatment decisions. Accordingly, a prospective study is currently ongoing to investigate the feasibility and benefit of the MRD-tailored therapy according to the longitudinal changes of the MRD status.

Introduction

Multiple myeloma (MM) is the second most common haematological malignancy, characterised by heterogeneous clinical characteristics, therapeutic responses and outcomes.¹ Introduction of novel agents leads to remarkable improvement in therapeutic response and survival.² Currently, more than 80 % of patients with newly diagnosed MM (NDMM) can achieve complete remission (CR) or better (eg sCR) after first-line treatment with various regimens.³ However, the majority of patients who have CR or better eventually experience relapse,⁴ suggesting the inadequacy of traditional response assessment, and highlighting the need for an additional approach to more precisely evaluate the depth of therapeutic response and predict the outcome of patients, particularly those who have achieved a response ≥CR.^2,3 This is even more important when considering various baseline risk factors (eg age, International Staging System (ISS) or Revised (R-)ISS, and cytogenetic abnormality (CA)), which are known to significantly influence both response and outcome.⁵

Minimal residual disease (MRD) has been emerging as a reliable surrogate endpoint for progression-free survival (PFS) for evaluating the efficacy of developing agents in clinical trials.2, 3, 4 In this context, a large amount of data from clinical trials and meta-analyses has well demonstrated that undetectable MRD strongly correlates with favourable outcomes of MM patients, including both PFS and overall survival (OS).6, 7, 8, 9 Currently, over 50 ongoing phase III trials have included MRD as an endpoint to help estimate the depth of therapeutic response and clinical benefit of MRD-directed treatment assignment.² The MRD status, assessed by next-generation flow cytometry (NGF) or next-generation sequencing (NGS) with a minimum sensitivity of 10^–5, has thus been incorporated into the revised International Myeloma Working Group (IMWG) criteria for therapeutic response in MM,⁵ as well as the US Food and Drug Administration (FDA) and European Medicines Agency (EMA) recommendations for clinical trials.^10,11 The prognostic value of MRD has been found to be independent of other adverse factors such as advanced disease stage (eg ISS or R-ISS III) and high-risk CA (HRCA).12, 13, 14 However, there are several key questions that need to be answered before incorporating measurement of the MRD status as a useful tool into the standard of care for guiding treatment decisions and informing the timing of therapeutic interventions.^4,15,16 Of note, most of these concerns stem from the dynamic nature of MRD during disease trajectory and treatment course. In this context, longitudinal changes of the MRD status during maintenance are associated with outcomes of transplant-eligible and -ineligible patients with MM in the TOURMALINE-MM3 and -MM4 trials.¹⁷ However, the prognostic impact of MRD dynamics remains not fully understood in real-world clinical practice.

In this study, we evaluated the dynamics of MRD and its clinical implications via monitoring the MRD status at multiple time points in a real-life setting.

Methods

Patients and treatment

This retrospective study was approved by the Institutional Review Board of the First Hospital of Jilin University (approval code 2018-087). All patients had given written informed consent to the use of clinical data according to the Declaration of Helsinki and the Good Clinical Practice guidelines. The study included 220 patients with NDMM diagnosed at the First Hospital of Jilin University in China between 1 March 2015, and 30 June 2019, who had NGF-based MRD testing at least once after initiation of therapy, with the first test conducted at partial response or better. Conventional therapeutic responses were evaluated according to the IMWG consensus.¹⁸ All patients were treated in the real-life setting, including proteasome inhibitors (PIs; mainly bortezomib), immunomodulatory drugs (IMiDs; mostly lenalidomide) or both, while only a few patients (<5 %) received daratumumab. 15 % and 25 % of patients received autologous stem cell transplantation (ASCT) and maintenance (mainly IMiD-based regimens such as lenalidomide plus dexamethasone). Patients with lytic bone lesions (∼90 % of all patients) received bisphosphonates.

MRD assessment

MRD was evaluated in bone marrow (BM) samples via the routine procedure for the evaluation of therapeutic responses at our centre, using a modified two-tube eight-colour assay (see Supplemental methods) at the sensitivity of 10^–5,¹⁹ with the median limit of detection (LOD) of 3.86 × 10^–5.²⁰ Undetectable MRD (MRD⁻) was confirmed by at least two separate tests. Duration of undetectable MRD was defined as the time from the achievement of undetectable MRD to the reappearance of detectable MRD or the last follow-up.⁵

Interphase fluorescence in situ hybridisation (iFISH)

CAs were assessed using iFISH in CD138⁺ cells isolated from BM samples via routine diagnostic procedures (see Supplemental methods). Among them, del(17p), t(4;14), t(14;16), and 1q+ (1q21 gain/amplification) were defined as HRCAs by the IMWG.²¹

Outcomes

PFS was defined as the time from the first day of treatment until the date when disease progression, relapse or death was first reported. Patients who did not progress or relapse were censored on the last date that they were known to be alive without progression. OS was defined as the time from the first day of treatment until the date of death due to any cause. Patients without documented death at the time of analysis were censored at the date of the last follow-up.

Statistical analysis

Baseline characteristics were compared using the chi-square test or Fisher's exact probability test. PFS and OS probabilities were estimated using the Kaplan–Meier method. Differences were tested for statistical significance using the (two-sided) log-rank test. Univariate and multivariate analyses were conducted using the Cox regression model. The Cox proportional hazards model was used to analyse the effect of continuous time-dependent variables. All statistical analyses were conducted using SPSS software (version 22.0) and R packages survival and survminer in R/Bioconductor (version 3.6.1). P < 0.05 was considered statistically significant.

Results

Attainment of undetectable MRD offers a highly favourable outcome, irrespective of baseline risk factors

In 220 patients, the MRD status was assessed in a total of 700 BM aspirates, with a median test number of three times (range 1−11 times) per patient. Among them, 35 patients received MRD testing only once, all of whom had detectable MRD (MRD⁺). The first MRD testing was conducted during or after induction (91.8 %), consolidation (7.3 %), and maintenance (0.9 %). Baseline demographics and clinical characteristics were summarised according to the MRD status in Table 1.

Table 1.

Comparison of baseline characteristics and treatments between NDMM patients with undetectable and persistent MRD (n = 220).

All	MRD status, n (%)		P value
	Undetectable MRD	Persistent MRD
	n = 114	n = 106
Age, yrs			0.563
≥65	33 (28.9)	27 (25.5)
<65	81 (71.1)	79 (74.5)
Sex			0.297
Male	65 (57.0)	53 (50.0)
Female	49 (43.0)	53 (50.0)
Subtype
IgG	46 (40.4)	48 (45.3)	0.460
IgA	30 (26.3)	28 (26.4)	0.987
IgD	4 (3.5)	8 (7.5)	0.188
Light chain	27 (23.7)	20 (18.9)	0.384
Non/oligosecretory	7 (6.1)	2 (1.9)	0.173
ISS
I	15 (13.2)	12 (11.3)	0.678
II	43 (37.7)	36 (34.0)	0.562
III	56 (49.1)	58 (54.7)	0.407
R-ISS	(n = 91)	(n = 88)
I	6 (6.6)	5 (5.7)	0.800
II	64 (70.3)	52 (59.1)	0.115
III	21 (23.1)	31 (35.2)	0.073
β2-MG, mg/L			0.738
≥3.5	86 (75.4)	82 (77.4)
< 3.5	28 (24.6)	24 (22.6)
LDH, U/L	(n = 112)	(n = 103)	0.502
≥ ULN	26 (23.2)	28 (27.2)
< ULN	86 (76.8)	75 (72.8)
FISH			0.056
SRCA	35 (38.9)	21 (25.3)
HRCA	55 (61.1)	62 (74.7)
del(17p)	8/104 (7.7)	8/94 (8.5)	0.833
del(13q)	43/103 (41.7)	47/94 (50.0)	0.245
del(1p)	9/99 (9.1)	9/87 (10.3)	0.773
1q+	44/103 (42.7)	59/94 (62.8)	0.005
t(11;14)	10/72 (13.9)	8/62 (12.9)	0.868
t(4;14)	9/72 (12.5)	14/62 (22.6)	0.123
t(14;16)	1/71 (1.4)	6/62 (9.7)	0.050
Induction
PI	80 (70.2)	78 (73.6)	0.574
IMiD	12 (10.5)	20 (18.9)	0.080
PI plus IMiD	22 (19.3)	8 (7.5)	0.011
ASCT	28 (24.6)	5 (4.7)	< 0.001
Maintenance	46 (40.4)	10 (9.4)	< 0.001

β2-MG, serum β2-microglobulin; LDH, lactate dehydrogenase; ULN: upper limit of normal; SRCA: standard-risk cytogenetic abnormality; HRCA: high-risk cytogenetic abnormality; PI, proteasome inhibitor; IMiD, immunomodulatory drug; ASCT, autologous stem cell transplantation.

The median time to event for achieving undetectable MRD (MRD⁻) was four treatment cycles or 4.1 months after initiation of therapy (Fig. 1A). Among them, 57.0 %, 28.9 % and 14.0 % of patients achieved MRD⁻ during induction, after induction and post-ASCT/consolidation, respectively (Fig. 1A, inset). The majority (70 %) of them achieved MRD⁻ within the first 6 months after initiation of therapy (Fig. 1B). A Cox regression model was then used to estimate the effect of time to achieving MRD⁻, treated as a continuous variable, on patient outcome. The speed of achieving MRD⁻ was significantly associated with longer PFS (P = 0.032) but not OS (P = 1.105). There were significant differences in PFS (P = 0.0387), but not OS (P = 0.2326), among patients who achieved MRD⁻ during induction, after induction, and post-ASCT/consolidation (Supplemental Figure 1A and 1B).

Fig. 1.

Time to undetectable MRD and patient outcomes according to the MRD status. (A, B) Cycle number of treatment (A) and time (B) from the start of treatment to undetectable MRD (MRD⁻) attainment (n = 220). Distribution of MRD⁻ patients at different treatment stages (A, inset). (C, D) Kaplan–Meier estimates of PFS (C) and OS (D) for MRD⁻ versus persistent MRD (MRD⁺) in the entire cohort of patients. (E, F) Kaplan–Meier estimates of PFS (E) and OS (F) for MRD⁻ versus MRD⁺ in patients with age ≥65 years.

At 5-year follow-up (median, 33.3 months), Kaplan–Meier analysis using the MRD status as a fixed covariable revealed that the attainment of MRD⁻ substantially prolonged PFS (median, 50.8 versus 12.8 months; Fig. 1C) and OS (median, not reached versus 32.8 months; Fig. 1D). Hazard ratios (HRs) for PFS and OS according to the MRD status were 0.19 (95 % confidence interval (CI), 0.12−0.28; P < 0.0001) and 0.25 (95 % CI, 0.15−0.42; P < 0.0001), respectively. Moreover, the 3-year PFS rate was 52.0 % versus 14.1 % for patients with MRD⁻ and detectable MRD (MRD⁺), with virtually all MRD⁺ patients experiencing progressive disease (PD) or relapse within 4 years. Consistently, the 5-year OS rate was 60.5 % versus 18.4 % for MRD⁻ and MRD⁺ patients. The favourable impact of MRD⁻ attainment was also observed in older patients (age ≥65 years) compared with their MRD⁺ counterparts, including PFS (HR 0.17, 95 % CI 0.08−0.37, P < 0.0001; Fig. 1E) and to a lesser extent, OS (HR 0.27, 95 % CI 0.11−0.67, P = 0.0049; Fig. 1F). Subgroup analysis of OS according to the MRD status revealed that the favourable outcome of MRD⁻ patients was consistent across all patients with different baseline risk factors (eg age ≥65 years, ISS III, R-ISS III, LDH > upper limit of normal/ULN and HRCA; Supplemental Figure S1C). Furthermore, following univariate analysis (Supplemental Table S1), multivariate Cox models adjusted for ISS III, LDH >ULN and HRCA (Supplemental Table S2), or R-ISS III (Supplemental Table S3) revealed that both MRD⁻ attainment and MRD⁻ duration ≥12 months were the strongest independent predictors for PFS and OS (P < 0.001 for both). Together, these observations support a notion that the MRD status is a powerful prognostic predictor for the outcome of NDMM patients, independent of virtually all baseline risk factors,²² while the speed to achieve undetectable MRD might not significantly affect its prognostic value.²³

Durable MRD negativity is essential for its favourable prognostic effect

According to the IMWG guideline, sustained MRD negativity is defined as MRD negativity confirmed for a minimum of 1 year.⁵ However, there is little direct evidence supporting this somewhat arbitrary interval thus far.³ In 114 patients who achieved undetectable MRD (MRD⁻), the median duration of MRD⁻ was 17.3 months (range, 1.2−60.3 months). There was no significant impact of baseline characteristics on the MRD⁻ duration, treated as a continuous variable (Fig. 2A). Notably, the MRD⁻ duration ≥12 months was significantly associated with ASCT (82.1 % versus 60.6 % for those without ASCT), although this finding needs to be further validated due to the relatively small number of patients who received ASCT in this cohort.

Fig. 2.

Impact of durable undetectable MRD on patient outcomes. (A) Impact of baseline risk factors on the duration of undetectable MRD (MRD⁻), treated as a continuous variable (n = 114). (B, C) Kaplan–Meier estimates of PFS (B) and OS (C) for patients with the duration of MRD⁻ ≥12 months versus <12 months. (D, E) Kaplan–Meier estimates of PFS (D) and OS (E) for patients with the duration of MRD⁻ ≥6 months versus <6 months.

MRD⁻ duration ≥12 months considerably improved outcome, with median PFS and OS both not reached versus 26.5 and 47.2 months respectively between MRD⁻ duration ≥12 and <12 months. Notably, MRD⁻ duration ≥12 months led to an 83 % (95 % CI, 0.09−0.34; P < 0.0001; Fig. 2B) or 69 % reduction (95 % CI, 0.13−0.76; P = 0.0098; Fig. 2C) in the risk of progression/death or death. 3-year PFS and 5-year OS rates were 68.7 % versus 7.9 % and 75.1 % versus 22.0 % respectively between MRD⁻ duration ≥12 and <12 months. Moreover, MRD⁻ duration ≥6 months also provided a clear benefit, including PFS (HR 0.16, 95 % CI 0.06–0.42, P = 0.0002; Fig. 2D) and OS (HR 0.12, 95 % CI 0.03–0.47, P = 0.0021; Fig. 2E). When treated as a continuous variable, a 1-month increase in the MRD⁻ duration resulted in a reduction of 12.7 % (95 % CI 0.806−0.945; P = 0.001) or 9.5 % (95 % CI 0.833−0.983, P = 0.018) in the risk of progression/death or death, respectively. Multivariate analysis revealed MRD⁻ duration ≥12 months as an independent factor for PFS and OS (P < 0.001 for both; Supplemental Table S2 and S3). Together, these findings argue strongly that the durability of undetectable MRD is essential for improving the outcome of patients with NDMM, irrespective of their baseline risk factors.

Loss of MRD negativity severely impairs its favourable effect

We then analysed the impact of undetectable MRD (MRD⁻) loss on the outcome of patients with NDMM. In 114 patients who achieved MRD⁻, 51 patients (44.7 %) experienced a conversion from MRD⁻ to MRD⁺ and/or PD, 68.6 % of whom occurred after treatment cessation (Fig. 3A). Among them, 37 patients had available baseline CA data, 70.3 % of whom carried at least one HRCA (Fig. 3A, bottom). Notably, patients with HRCAs lost their MRD⁻ status and/or had PD clearly faster than those with SRCAs, although the difference was not statistically significant (P = 0.1542; Supplemental Figure S2A).

Fig. 3.

Impact of undetectable MRD loss on patient outcomes. (A) Time to loss of undetectable MRD (MRD⁻) and/or progressive disease (PD) (n = 51). Each column represents an individual patient. Bottom, baseline cytogenetic abnormality (CA) information. SRCA, standard-risk CA; HRCA, high-risk CA; Tx, treatment. (B, C) Kaplan–Meier estimates of PFS (B) and OS (C) for patients who had durable MRD⁻ versus lost MRD⁻ within 12 months. (D, E) Kaplan–Meier estimates of PFS (D) and OS (E) for patients who lost MRD⁻ versus persistent MRD (MRD⁺).

In 114 patients who achieved MRD⁻, 100 patients were available for analysing sustained MRD⁻ versus loss of MRD⁻. Compared with sustained MRD⁻, loss of MRD⁻ resulted in a sharp reduction in PFS (HR 0.01, 95 % CI 0–0.06, P < 0.0001; Fig. 3B) and OS (HR 0.03, 95 % CI 0–0.24, P = 0.0008; Fig. 3C). Although patients who lost the MRD⁻ status had relatively longer PFS than those who had persistent MRD (MRD⁺; P = 0.0181; Fig. 3D), there was no significant difference in OS between them (P = 0.0545; Fig. 3E). Early loss of MRD⁻ (within 12 months after achieving MRD⁻) was associated with shorter PFS (P = 0.0191; Supplemental Figure S2B), while no significant difference in OS was observed between patients who lost the MRD⁻ status earlier and later (P = 0.9107; Supplemental Figure S2C). Maintenance treatment did not significantly affect the time to loss of MRD⁻ (P = 0.3873; Supplemental Figure S2D), in association with no marked improvement in PFS (P = 0.1896; Supplemental Figure S2E) and OS (P = 0.2566; Supplemental Figure S2F) compared with patients without maintenance. These observations suggest that loss of MRD⁻, no matter when it occurs, could substantially impair its favourable impact on prognosis.

The MRD status is superior in predicting disease progression/recurrence

Emerging evidence supports a notion that undetectable MRD (MRD⁻) may be a better prognostic predictor than CR,⁷ therefore raising an issue of when would be the optimal time to start MRD assessment during the treatment course. We thus examined the temporal relationship between MRD⁻ attainment and conventional responses. 19 % of patients with CR or sCR as the best response attained MRD⁻ before achieving ≥CR, while 38.0 % and 43 % of patients did so at or after ≥CR, respectively (Fig. 4A). 12 % of patients with very good partial response (VGPR) as the best response attained MRD⁻ prior to VGPR, while the remaining patients (88 %) achieved MRD⁻ at or after VGPR (Fig. 4B), suggesting a lack of temporal consistency between MRD⁻ attainment and conventional objective responses.

Fig. 4.

Temporal relationship between undetectable MRD and conventional objective responses. (A, B) Time from CR or sCR (A) and VGPR (B) to achieving MRD⁻. The value for each column indicates the percentage of cases in MRD⁻ patients. (C, D) Kaplan−Meier estimates of PFS according to the MRD status in patients who achieved CR or sCR (C) and VGPR (D) as the best response.

In patients who achieved ≥CR as the best response, attainment of MRD⁻ significantly prolonged PFS (HR 0.21, 95 % CI 0.13−0.36, P < 0.0001; Fig. 4C), reflecting a delay of disease progression or recurrence, compared with their counterparts who had persistent MRD (MRD⁺). A similar phenomenon was observed in patients who achieved VGPR as the best response (HR 0.08, 95 % CI 0.01−0.64, P = 0.0167; Fig. 4D). Together, these findings argue that the MRD status could more precisely predict disease progression or recurrence (relapse) than conventional objective responses, even ≥CR.²⁴

The MRD status robustly predicts disease progression/recurrence, independently of baseline risk stratification

Then, we analysed the value of the MRD status in predicting disease progression or recurrence in patients with different baseline factors (eg CAs, ISS and R-ISS staging) used for risk stratification at diagnosis.^21,25,26 In patients carrying HRCAs, attainment of undetectable MRD (MRD⁻) significantly prolonged PFS, when compared with their counterparts with persistent MRD (MRD⁺) (HR 0.39, 95 % CI 0.30−0.52, P < 0.0001; Fig. 5A). A similar phenomenon was observed in patients with ISS III (HR 0.38, 95 % CI 0.29−0.51, P < 0.0001; Fig. 5B) or R-ISS III (HR 0.43, 95 % CI 0.29−0.66, P < 0.0001; Fig. 5C). Notably, MRD⁻ patients carrying HRCAs or with ISS III or R-ISS III displayed longer PFS than MRD⁺ patients carrying SRCAs (HR 0.31, 95 % CI 0.16−0.63, P = 0.001; Fig. 5A) or with ISS I/II (HR 0.30, 95 % CI, 0.18−0.52, P < 0.0001; Fig. 5B) and R-ISS I/II (HR 0.35, 95 % CI, 0.17−0.71, P = 0.0035; Fig. 5C). In MRD⁻ patients, patients carrying HRCA (versus SRCA; P = 0.0986) or with ISS III (versus ISS I/II; P = 0.1576) or R-ISS III (versus R-ISS I/II; P = 0.2027) had relatively shorter PFS, while the differences were not statistically significant. Moreover, subgroup analysis showed that baseline risk factors (eg age ≥65 years, HRCA, ISS III, R-ISS III and LDH >ULN) did not significantly affect the value of the MRD status in predicting PFS (Fig. 5D). Consistently, multivariate analysis adjusted for ISS III, LDH >ULN and HRCA revealed MRD⁻ as a robust favourable predictor for PFS (Supplemental Table S2 and S3). Together, these observations suggest that the MRD status might remodulate the prognostic property of baseline risk factors such as HRCA and advanced disease (eg ISS III and R-ISS III), suggesting that the MRD status could predict disease progression or relapse, independently of risk stratification at diagnosis.

Fig. 5.

Prediction of PFS by the MRD status in patients with different baseline risk factors. (A-C) Kaplan–Meier estimates of PFS according to the MRD status for patients carrying standard-risk (SRCA) versus high-risk cytogenetic abnormalities CA (HRCA) (A), with ISS III versus I/II (B) and R-ISS III versus I/II (C). (D) Subgroup analysis of PFS for patients with undetectable (MRD⁻) or persistent MRD (MRD⁺) according to age (≥65 versus <65 years), ISS (III versus I/II), HRCA (yes versus no), and R-ISS (III versus I/II).

Discussion

In the era of novel agents, the accuracy and precision of the conventional therapeutic response criteria have been challenged due to their intrinsic limitations in reflecting the depth of disease remission after treatment. For example, some patients with CR have outcomes similar to or even worse than those with VGPR or PR,^27,28 probably in association with clonal evolution due to the impact of treatment and depth of response on the genetics and sub-clonal structure.²⁹ The inadequacy of conventional response assessment may provide a false-negative result, which could mislead subsequent treatment decisions in clinical care.^2,30,31 In this context, an increasing number of trials have adapted MRD as a secondary or even co-primary endpoint.^32,33 However, there are several key issues to be addressed before incorporating MRD assessment into the standard of care.^4,34 Most of these issues stem from the longitudinal changes of the MRD status during treatment course and follow-up, as well as their relationship with the risk stratification at diagnosis and therapeutic response evaluation currently applied in daily practice.^35,36 In a letter to the editor, we have recommended that the MRD status should be longitudinally monitored for precise estimation of OS in patients with NDMM.³⁷ In the present study, we demonstrated that the durability of undetectable MRD is essential for truly benefiting long-term outcomes (PFS and OS), while loss of undetectable MRD status during or after treatment severely impaired the favourable property of undetectable MRD, highlighting the importance for monitoring the MRD kinetics at multiple time points throughout the entire disease and treatment course.^35,38, 39, 40 As the MRD status robustly predicted disease progression or recurrence, independently of virtually all baseline risk factors as well as conventional objective responses, it could thus serve as a biomarker for relapse to guide early re-treatment of relapsed disease.

Due to longitudinal changes of the MRD status, durable undetectable MRD has been considered crucial for its favourable prognostic value.^3,35,40 In a prospective study, sustenance of the MRD-negative status for a longer period is likely to be more favourable.³⁶ In another study, patients with sustained MRD negativity are less likely to progress than those who lost their MRD-negative response (by NGF at 10^–5) at the 2-year landmark.⁴⁰ In the IMWG consensus, sustained MRD negativity was defined as MRD negative in the BM determined by NGF, NGS or both and confirmed by imaging within a minimum of 1 year.⁵ However, little evidence is available to support this arbitrary interval, probably due to the limited length of study, which has made serial MRD evaluations difficult in most pre-designed clinical trials.^40,41 In the present study, we observed that patients with sustained undetectable MRD longer than 12 months (or even 6 months) had substantially prolonged PFS and OS, consistent with the recent findings (MRD negativity by NGS at 10^–5) in clinical trials.^32,42 Notably, every 1-month increase in the duration of undetectable MRD led to an approximately 13 % or 10 % reduction in the risk of disease progression/death or death, arguing that the longer undetectable MRD is sustained, the better the outcome.⁴³ However, further investigation is required to define the optimal length of undetectable MRD duration, considering toxicity and financial issues due to excess long-term maintenance.³ Conversion from undetectable MRD to detectable MRD is associated with poor outcome, probably equivalent to or even worse than those with persistent MRD.^17,40,44 Consistently, we observed that loss of undetectable MRD status substantially shortened both PFS and OS, leading to a dismal outcome similar to persistent MRD. In the GEM2012MENOS65 and GEM2014MAIN trials, MRD resurgence and/or PD occurs in 42 % of MRD⁻ patients, while it could be predicted by ISS III and circulating tumour cells (CTCs) ≥0.01 %.⁴⁵ In line with a recent finding that maintenance does not further improve PFS in MRD⁻ patients,¹⁷ we observed that maintenance (mainly with IMiDs) did not significantly prolong the time to loss of undetectable MRD and there were no significant differences in either PFS and OS between MRD-negative patients with and without maintenance . In the FORTE trial, MRD was assessed before maintenance and at every 6 months thereafter, showing that 39 % of MRD⁻ patients lost their MRD⁻ status, in association with 1q+, ≥2 HRCAs, high CTCs and time-to-reach MRD negativity post-consolidation.⁴⁶ It also suggests that 1 year between two MRD-negative evaluations may be not enough to identify patients without MRD resurgence and/or relapse. Together, these findings highlight a potential requirement for monitoring the MRD status even during maintenance and follow-up,^34,40 which might help capture the early signs (eg loss of MRD negativity) of disease relapse and thus start re-treatment as early as possible to improve the outcome of patients with relapsed MM.⁴¹ In contrast, maintenance may be discontinued in patients with persistent MRD negativity (eg for 2 years).^47,48 While the existing data supporting the use of MRD to guide therapy remain controversial in either induction/consolidation or maintenance setting, many trials are ongoing to adopt MRD into multiple stages of MM treatment, including maintenance (eg de-escalation or cessation due to sustained MRD negativity), consolidation/ASCT (eg deferral or even omission when achieving MRD negativity and, in contrast, intensification or alternative therapy due to persistent MRD), early intervention for relapsed disease (eg re-treatment due to MRD resurgence).⁴⁹ Therefore, it is expected that, in addition to the use of MRD to evaluate therapeutic response as recommended by the IMWG consensus and other guidelines,^5,10,11 MRD would also be recommended to guide therapeutic decision-making to individualise therapy soon by these guidelines.⁵⁰ Nonetheless, understanding the dynamics of MRD is necessary for MRD-tailored or -adapted therapy.

Consistent with the earlier findings,⁵¹ we observed considerable inconsistency between conventional therapeutic responses (eg CR, defined by negative immunofixation on serum and urine, disappearance of any soft tissue plasmacytomas, and <5 % bone marrow plasma cells)⁵ and MRD results. One-third of patients who achieved CR or sCR were MRD-positive, while most of the MRD-negative patients (∼90 %) had at least CR. Many studies have demonstrated that undetectable MRD can more precisely reflect the depth of disease remission than CR.^28,52 For example, MRD-negative CR was significantly associated with a delay in disease progression compared with no MRD-negative CR.³⁶ Similarly, we observed that patients with MRD-negative CR had substantially longer PFS than those with MRD-positive CR, supporting the notion that the MRD status could further stratify patients who achieved CR or greater regarding the risk of disease progression or recurrence. Moreover, while one-third of patients who achieved VGPR were MRD-negative, their PFS was significantly prolonged compared with those with MRD-positive VGPR, suggesting that monitoring the MRD status might also benefit the patients who achieved VGPR as the best response.^24,53 However, although MRD negativity was confirmed at least twice in virtually all cases, a possibility that some patients had undetectable MRD before CR or VGPR due to false-negative MRD resulting from non-representative BM aspirates could not be excluded.⁵¹ Nevertheless, MRD assessment might need to be conducted at the time of suspected objective response (eg VGPR, CR or sCR) after treatment.

MRD-positive patients with HRCA display the worst outcome,^7,9 while it is controversial whether achieving undetectable MRD could override the inferior prognostic effect of HRCAs.⁴² It has been reported that achievement of undetectable MRD overcomes the poor prognosis of patients with R-ISS III disease, while those who remain MRD-positive have progressively worsening outcomes regardless of R-ISS stage.¹² In this study, the MRD status robustly predicted PFS, reflecting disease progression or recurrence (relapse), regardless of HRCA or ISS/R-ISS stage determined at diagnosis. Therefore, these findings reinforce the notion that the MRD status can modulate patients' risk defined at diagnosis.^3,9,12 They also suggest that the baseline risk status could change over the treatment course,^12,54 largely due to the MRD dynamics.

This study has some major limitations, including its retrospective nature, relatively low sensitivity (10^–5) of the MRD test (although a threshold of 10^–5 or 10^–6 is currently considered standard),^45,46 because other approaches for measuring MRD (eg NGS and imaging) were not routinely used until recently in our centre, and relatively worse outcome of patients (most likely attributing to a large proportion of patients with HRCA and advanced disease at baseline). Since the data cutoff date of this study was 30 June 2019, only a few patients received the anti-CD38 antibody daratumumab that was approved for transplant-ineligible and -eligible patients with NDMM on 7 May 2018 and 26 September 2019, respectively, which prevented the analyses for its effect on MRD. In this context, the possibility that the binding of the therapeutic antibody to the antigenic site on the cell surface could interfere with the binding of antibodies used for flow cytometry, therefore resulting in signal loss in the MRD test, should be considered.⁵⁵ The small number of patients who received ASCT that was not powered for the subgroup analysis on its impact on MRD. Thus, while our observations in this study need to be further confirmed, particularly in appropriately designed prospective studies, caution needs to be taken to explain our observations in the current practice.

In summary, this study provides real-life evidence to address current major concerns stemming from the dynamic nature of MRD, in the application of MRD testing in clinical practice, including appropriate time and time points for MRD testing, impact of undetectable MRD on high-risk or advanced diseases, meaningful duration of MRD negativity, and loss of MRD negativity as a marker of early relapse. Our findings may evoke insightful considerations how to translate the powerful prognostic value of the MRD dynamics into the standard of care and to design future clinical studies involving MRD, ultimately leading to personalised MRD-tailored therapy. Base on these findings, we designed a prospective study to investigate the feasibility and benefit of the MRD-tailored therapy according to the longitudinal changes of the MRD status, which is currently ongoing.

Ethics statement

The study was approved by the Institutional Review Board (IRB) of the First Hospital of Jilin University (approval # 2018–087). It adhered to the Declaration of Helsinki and the Good Clinical Practice guidelines. All patients had given written informed consent to the use of clinical data.

Data availability statement

The data that support the findings of this study are available from the corresponding author upon reasonable request.

CRediT authorship contribution statement

Weiling Xu: Writing – review & editing, Visualization, Validation, Resources, Methodology, Investigation, Formal analysis. Xinyue Liang: Writing – review & editing, Validation, Resources, Methodology, Investigation, Formal analysis, Data curation. Shanshan Liu: Writing – review & editing, Validation, Resources, Investigation. Xingcheng Yi: Writing – review & editing, Visualization, Software, Methodology. Mengru Tian: Writing – review & editing, Validation, Resources, Methodology, Investigation. Tingting Yue: Writing – review & editing, Validation, Investigation. Yingjie Zhang: Writing – review & editing, Validation, Investigation. Yurong Yan: Writing – review & editing, Validation, Investigation. Maozhuo Lan: Visualization, Software, Methodology. Mengtuan Long: Writing – review & editing, Visualization, Software, Methodology. Nan Zhang: Writing – review & editing, Validation, Investigation. Jingxuan Wang: Writing – review & editing, Validation, Investigation. Xiaoxiao Sun: Writing – review & editing, Validation, Investigation. Rui Hu: Writing – review & editing, Validation, Investigation. Yufeng Zhu: Writing – review & editing, Validation, Investigation. Xintian Ma: Writing – review & editing, Validation, Investigation. Yue Cheng: Writing – review & editing, Resources, Methodology. Jiayi Xu: Writing – review & editing, Resources, Methodology. Yun Dai: Writing – review & editing, Writing – original draft, Visualization, Validation, Supervision, Resources, Project administration, Methodology, Investigation, Funding acquisition, Formal analysis, Data curation, Conceptualization. Fengyan Jin: Writing – review & editing, Writing – original draft, Validation, Supervision, Resources, Project administration, Methodology, Investigation, Funding acquisition, Formal analysis, Data curation, Conceptualization.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.