Circulating miRNAs as Auxiliary Diagnostic Biomarkers for Multiple Myeloma: A Systematic Review, Meta-Analysis, and Recommendations

Multiple myeloma (MM) is a hematologic malignancy characterized by aberrant expansion of monoclonal plasma cells with high mortality and severe complications due to the lack of early diagnosis and timely treatment. Circulating miRNAs have shown potential in the diagnosis of MM with inconsistent results, which remains to be fully assessed. Here we updated a meta-analysis with relative studies and essays published in English before Jan 31, 2021. After steps of screening, 32 studies from 11 articles that included a total of 627 MM patients and 314 healthy controls were collected. All data were analyzed by REVMAN 5.3 and Stata MP 16, and the quality of included literatures was estimated by Diagnostic Accuracy Study 2 (QUADAS-2). The pooled area under the curve (AUC) shown in summary receiver operating characteristic (SROC) analyses of circulating miRNAs was 0.87 (95%CI, 0.81–0.89), and the sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), and diagnostic odds ratio (DOR) were 0.79, 0.86, 5, 0.27, 22, respectively. Meta-regression and subgroup analysis exhibited that “miRNA cluster”, patient “detailed stage or Ig isotype” accounted for a considerable proportion of heterogeneity, revealing the importance of study design and patient inclusion in diagnostic trials; thus standardized recommendations were proposed for further studies. In addition, the performance of the circulating miRNAs included in MM prognosis and treatment response prediction was summarized, indicating that they could serve as valuable biomarkers, which would expand their clinical application greatly. Systematic Review Registration https://www.crd.york.ac.uk/prospero/display_record.php?RecordID=234297, PROSPERO, identifier (CRD42021234297).


INTRODUCTION
Multiple myeloma (MM), the second most common hematological malignancy (1), develops from monoclonal gammopathy of undetermined significance (MGUS) and smoldering multiple myeloma (SMM), through the malignant transformation of long-lived plasma cells deriving from memory B cells and plasma-blasts (2). Normal interactions between plasma cells and their environment (the bone remodeling chambers) enable the stability of normal plasma cell genotypes and phenotypes, which may be disrupted by multiple factors at a precursor stage of MM (MGUS, pre-MGUS state) leading to MM tumorigenesis (3). The risk of progression for MGUS is about 1% yearly and 10% for SMM in the first 5 years (4). Early treatment is of benefit, inhibiting disease progression but with nonnegligible side effects (5), and a novel biomarker with high accuracy for early diagnosis is imperative.
As a member of non-coding RNAs, miRNAs are small RNA molecules that function as negative regulators of gene expression. The decrease or increase of specific miRNAs in MM is associated with the dysregulation of the target gene expression, reflecting their impact on tumor-suppression or promotion (6,7). Circulating miRNAs have been explored as valuable tools for various tumor diagnoses and prognoses by several studies (8). Similarly, the diversification of circulating miRNA expression in MM has been investigated, indicating that miRNAs could serve as potential diagnostic biomarkers (9,10). However, not every miRNA was eligible, and there was significant heterogeneity among the findings. We updated a meta-analysis to address whether the circulating miRNAs could be promising biomarkers for early detection of MM with the latest evidence and try to resolve the problems that contributed to heterogeneity in previous studies.

MATERIALS AND METHODS
This systematic review and meta-analysis was conducted following the guidance of the PRISMA 2020 Statement: an Updated Guideline for Reporting Systematic Reviews (11), and registered on PROSPERO prior to the start.

Study Selection
Studies that met the following criteria were included: 1) all the MM patients were diagnosed according to standard diagnostic criterion; 2) control subjects were analyzed concurrently; 3) miRNA measured by qRT-PCR and the process clearly described; 4) same outcome: sensitivity, specificity, AUC; 5) specimens are limited to serum or plasma; and 6) sample size was given. The exclusion criteria were as follows: 1) different control groups; 2) reviews and meta-analysis; 3) abstracts, editorials, conference papers, and letters without valid data; 4) repeated articles; 5) not conducted on humans; 6) sample size was insufficient; 7) non-English literature. To avoid any selection bias, two independent authors decided whether to include or exclude a study when they reach a consensus; otherwise, a third author reviewed the article again and resolved the disagreement.

Quality Assessment of Literature and Data Extraction
The quality of the selected studies was assessed according to QUADAS-2 (Quality Assessment of Diagnostic Accuracy Studies-2) on RevMan (version 5.3, London, UK, RRID: SCR_003581) by two authors and illustrated in a graph. Also, basic information from eligible articles/supplementary materials (first authors, publication year, specimen, sample size, miRNAs, expression mode, sensitivity, specificity, AUC, patient information including Ig isotype, stage, study design, gender structure, and age range) was extracted by two authors independently using a standardized form and then reviewed by a third author in detail.

Statistical Analysis
The data of true positives (TPs), false positives (FPs), true negatives (TNs), and false negatives (FNs) of 32 miRNAs extracted from 11 individual studies were analyzed in REVMAN 5.3 and STATA (version MP16, Texas, USA, RRID : SCR_012763). The sensitivity, specificity, positive likelihood ratio (PLR), negative likelihood ratio (NLR), diagnostic odds ratio (DOR), and area under the curve (AUC) were pooled with the random effects meta-analysis model. Total diagnostic accuracy was assessed by summary receiver operating characteristic (SROC) curve with the sensitivity and specificity data of each qualified study. To evaluate the existence of heterogeneity, I 2 over 50% and/or P-value of Q-test under 0.05 were set as statistically significant. Meta-regression was executed with univariable model and multivariate model respectively. Subgroup analyses were employed to dissociate the heterogeneity among the studies using the random-effects inverse-variance model with DerSimonian-Laird estimate of tau². All P-values less than 0.05 were considered statistically significant.

RESULTS
Literature Selection, Quality Assessment, and Study Characteristics As the study search and selection process show in Figure 1, a total of 898 articles were initially obtained; 897 studies were identified from Cochrane Library, PubMed, Embase, Web of Science, SinoMed, CQVIP database, Wan Fang Data, China National Knowledge Infrastructure (CNKI), and Clinical Trials.gov, and 1 was acquired from reference lists of articles cited in this paper. All of the articles were imported into Zotero. After carefully screening for the title. abstract, and full-text, 117 Chinese articles, 360 duplicates, 14 patent, 25 reviews, 8 conference abstracts, 349 with irrelevant themes, 5 with different types of specimen, and 10 without available data were excluded. Finally, 32 miRNAs from 11 articles were included for this meta-analysis. In comparison to the previous version of review (10), one report of a study for non-circulating specimen (12) and seven reports of four articles published in Chinese were excluded; and one report of a new study was included (13). The main information of circulating miRNA reports extracted from the qualified studies is displayed in Table 1. Among these, six extra reports of the studies included in the previous version of review were found from the articles and supplementary materials (14,18,21,22). Characteristics of MM patients and healthy controls are available in Table 2. All the eligible articles were published before Jan 31, 2021 containing 627 MM patients and 314 healthy controls. The expression levels of miRNAs in serum (n = 27) or plasma (n = 5) were detected by qRT-PCR. Of these 32 studies, three studies evaluated miRNA clusters, whereas the others evaluated individual miRNA.
Quality assessment of qualified literatures by QUADAS-2 tool reveal that the overall quality of the studies included was acceptable, but with a few unignorable flaws and uncertainties ( Figure 2). Some of the literatures did not clarify the study design, the patient information (stage/Ig isotype/prior to any treatment), and whether there was an appropriate time interval between the reference standard and miRNA detection. Besides, the absence of suspected cases in some of the literatures may lead to bias in the evaluation of diagnostic power.

Diagnostic Performance Evaluation
All the data were included to calculate the combined diagnostic accuracy of miRNAs by STATA MP 16 software. As Forest plots present, the sensitivity and specificity of miRNAs in various studies were significantly heterogeneous (sensitivity: I 2 = 76.86%, p < 0.01; specificity: I 2 = 62.94%, p < 0.01). By recombining the data with random effect models, the pooled sensitivity and specificity were 0.79 (95%CI, 0.73-0.83) and 0.86 (95%CI, 0.81-0.89) respectively   Figure 3B). The pooled DOR was 22 (95%CI, 14-35) using a random effect model ( Figure 3C). Figure 4A, demonstrating that a small number of miRNAs (No. 9/18/19/20/21/31/32) contribute to the confirmation or exclusion of MM, whereas the others appear to be underachieving. The pre-test and post-test probabilities were evaluated by Fagan's nomogram, presenting that when the prior probability was 20%, post-test probabilities were 54 and 6% for positive and negative circulating miRNAs in MM patients, respectively ( Figure 4B).

Sensitivity Analysis and Heterogeneity Exploration
Then, a sensitivity analysis was performed to evaluate the stability of the combined DOR. After subtracting the No. 18 study with the maximal deviation, the DOR value changed from 22 to 19, presenting that there was no significant change in the results ( Figure 5A). Similarly, other studies, excluding one, showed at the time consistent combinatorial DOR without significant fluctuations. However, the fixed-effect model had a greater influence on the pooled DOR than the random-effect model (DOR, 17 vs. 22), indicating that an improper data analysis method may have a great impact on the pooled results. Among these miRNAs, miR-4254 was the most promising diagnostic biomarker with 0.80 sensitivity (95%CI, 0.71-0.87), 0.98 specificity (95%CI, 0.90-1.0) and 252 DOR (95% CI, . Bivariate boxplot revealed that some of the studies with relatively higher AUC (No.18/21/31) showed stronger heterogeneity ( Figure 5B). Meanwhile, publication bias was not found by Deek's funnel plot asymmetry test (P = 0.73) ( Figure 5C). Further heterogeneity analysis by meta-regression according to "miRNA cluster", " detailed stage or Ig isotype", "newly diagnosed or untreated", "determined cohort study design", "sample size>30", "serum or plasma" and "ethnicity" were conducted with univariable model, demonstrating that "miRNA cluster", "detailed stage or Ig isotype" and "determined cohort study design" reached statistical significance. Subsequently, these three variables were analyzed by meta-regression with multivariate model, and there were significant differences between the "miRNA cluster" subgroups and the "detailed stage or Ig isotype" subgroups ( Table 3). Then subgroup analyses based on sensitivity, specificity, AUC, and DOR were performed using the random-effects inverse-variance model with DerSimonian-Laird estimate of tau² to investigate the specific heterogeneity existing within and between subgroups, and the results are displayed in Table 4. The diagnostic parameters of miRNA clusters were better than the single miRNA subgroup, whereas the parameters of the detailed patient information subgroup and determined cohort study design subgroup were inferior than their contrasts. The DOR-based subgroup analysis results are shown in Figure 6.

Circulating miRNAs in MM, MGUS, and SMM
In order to investigate the ability of the included circulating miRNAs to discriminate MM from MGUS/SMM, we summarized their distributional and differential diagnostic data among those disease states. Between the MM and MGUS groups, the levels of miR-20, miR-15a, and miR-92a were significantly different. In ROC analysis, the individual miRNAs did not exhibit substantial discriminative performance, which would be improved by the combination of miRNAs and other clinical parameters. The combination of miR-107, miR-15a-5p, and hemoglobin gained the best differential performance with AUC = 0.954 (14), and the combination of miR-1246 and miR-1308 ranked second with AUC = 0.725 (21). For MM and SMM, the expression level of miR-92a was significantly different, but the differential diagnostic value remains to be verified (20). For MGUS and SMM, Manier et al. found significant differences in the expression levels of miR-107, miR-92a, and miR-125a in circulating exosomes (24); however, the included serum or plasma miRNAs have not shown any differential diagnostic value ( Table 5).

DISCUSSION
Although survival of MM patients have been improved with the rapid advances of therapeutic strategy and supportive care, myriad patients still suffer from relapsed/refractory MM, entailing a reliable biomarker for early diagnosis. The remarkable impact of miRNA on protein expression is emerging with the discovery that about one-third of human encoding genes are regulated by miRNAs (26). Pioneering studies have found a mass of specific miRNAs carried by circulating microparticles significantly distinct from their maternal cells, depicting an interesting transfer pathway for the gene-regulating function of miRNAs from microparticles releasing cells to the target cells (27).
MicroRNAs transfer between MM plasma cells and bone marrow microenvironment, enabling the development and metastasis of malignancy through messenger RNA destruction or translation inhibition (28,29). Circulating microRNAs have been applied in hematological diseases as diagnostic biomarkers due to their reliable stability and non-invasive properties (30). In recent years, the important role of circulating miRNAs in MM diagnosis has received considerable attention; however, the results vary and leave questions open (10). We updated this meta-analysis to figure out whether the circulating miRNAs could be promising means for early detection of MM with the latest evidence. The pooled sensitivity, specificity, AUC, PLR, NLR, and DOR were 0.79, 0.86, 0.87, 5, 0.27, and 22, respectively, consistent with the previous meta-analysis (10). Summary LRP and LRN for Index Test suggest a small number of studies have relatively high value for diagnosis conformation and/or exclusion. The Fagan nomogram of post-test probability also indicated that circulating miRNAs were of relatively good diagnostic value but still had room for improvement. Researchers found that the tests based on serum or plasma did not show significant differences, nor between ethnic groups; and our meta-analysis confirmed it.
However, the quality assessment of literatures presents a few ignorable flaws and uncertainties, which are somewhat inconsistent with the results of the previous meta-analysis (10), possibly because of the different versions of evaluation tools or the strictness in our assessment. We found that some of the literatures did not specify the study design, patient information (stage/Ig isotype/treatment information), or the time interval between the reference standard and miRNA detection. Besides, the absence of suspected cases in some of the studies may lead to bias in evaluation.
The combined diagnostic power did not fluctuate significantly in sensitivity analysis. The DOR value only changed from 22 to 19 even when the study with the greatest deviation was  eliminated. To dissociate the sources of heterogeneity, a publication bias assessment was performed and no publication bias was found (p = 0.73). Inevitably, the exclusion of non-English literatures may also lead to selection bias. However, significant differences were found in the "miRNA cluster" and "detailed stage or Ig isotype" subgroups through meta-regression with univariable model and multivariate model (p < 0.05), but not in the "cohort study" and "newly diagnosed and untreated"  subgroups, which may also be influenced by the limited number of studies included in each subgroup. Besides, no significant differences were found in the "specimen" and "ethnicity" subgroups, consistent with previous studies. These results reveal that the study design and the enrolment of patients and healthy controls may have an impact on the diagnostic value of the index to be assessed, and a standardized recommendation is imperative. At present, the efficacy of early treatment determined by the free light chain ratio for the precursor-stage of MM (MGUS or SMM) to improve longevity and health-related quality of life is still unclear (31,32), and distinguishing symptomatic multiple myeloma from those conditions is of great importance. In our summary, not much distributional or differential diagnostic evidence of circulating miRNAs in MM, MGUS and SMM were found. Although miRNA expression levels differed significantly, no individual miRNA exhibited excellent differential diagnostic ability, indicating that this area needs to be further explored with more meticulous design for subject enrolment. Moreover, a variety of diseases such as inflammation (33), cardiovascular diseases (34), or other non-cancerous illnesses (35) may also alter miRNA profile and level, which should be taken into account by researchers when setting up control groups.
The past decade has seen remarkable achievements in the understanding of miRNAs in MM, including their various targets, effects, and dysregulation modes in disease development and progression. Some miRNAs, such as miR-20a, miR-19a, miR-92a, and miR-214-3p act as oncomiR playing important roles in antiapoptosis, proliferation, migration, and invasion. Other miRNAs, including miR-15a-5p, miR203, miR144, miR199, miR-483-5p, miR-34a, miR-33b, miR-202, and miR-29a act as tumor suppressors. Also, some miRNAs are involved in the development of bone marrow microenvironment in MM; for example, miR143 and miR-29a promote angiogenesis and osteoblast differentiation (36,37), miR199 neutralizes the oncogenic effect of bone marrow stromal cells (38), and miR-92a is essential for B cell development (39). These results provide valuable resources for the investigation of the etiology and treatment methods for MM. Interestingly, the role of miR-125b-5p in MM is still controversial since it has the ability to inhibit the growth and survival of MM cell and promote apoptosis and autophagy-associated cell death by targeting IRF4 and its downstream effector BLIMP-1 (40), but miR-125b-5p may generate counterproductive effects through different target genes or signaling pathways; it promotes MM progress by increasing p53 mRNA and protein and attenuating MM cell death in response to dexamethasone (41). Further exploration of miRNAs may enrich our perspective. More details are listed in Table 6.
Due to the important role of miRNAs in the tumorigenesis and progression of MM, circulating miRNAs have also shown their prognostic ability in MM, expanding their clinical application for identifying high-risk MM patients. As shown in Table 7, several studies have demonstrated that patients with different levels of miR-483-5p, miR-744, let-7e, miR-19a, miR-92a, miR-33b, miR-214, or miR-20a appeared with quite different survival rates.
More interestingly, the expression levels of miR-19a, miR-744-5p, miR-143-5p, and miR-92a, had significantly different outcomes in bortezomib-based treatment (22,(65)(66)(67). Patients with higher serum levels of miR-214 had extended overall survival upon bisphosphonate-based therapy (68). In patients with autologous stem-cell transplantation, the lower level of miR-15a on day 0 was associated with the shorter time to engraftment, and miR-20a decreased at complete remission (69,70). These results demonstrate that the expression patterns of circulating microRNAs are valuable markers for predicting treatment response. See Table 8 for details.

Strength and Limitations
The main strengths of this meta-analysis are the follows: 1) this is the first time that "cohort study design", patient "stage or Ig isotype" and "newly diagnosed or untreated" information of MM patients were included in the subgroup analysis, summarizing the possible influencing factors of the current results; 2) it uses extensive but rigorous search strategies to optimize the quality of included literature; 3) it is stricter in assessing the quality of the included literature, providing a more detailed summary of included patients and healthy donors; 4) it structures a comprehensive review of the current understanding of circulating miRNAs in multiple myeloma, including the value of diagnosis, differential diagnosis, prognosis, and therapyguiding. The major limitations are the following: 1) due to the insufficiency of miRNA studies included, the pooled diagnostic power of miRNA may be of some deviation, and the results of the subgroup analysis may also be biased to some extent; 2) some miRNAs are also dysregulated in other hematologic diseases; for example, miR-143, miR-144, and miR-199 are also underexpressed in child acute lymphoblastic leukemia, suggesting that circulating miRNAs may not be independent diagnostic biomarkers of MM, but can only be used as auxiliary and discriminatory diagnostic biomarkers (74).

Recommendations
For diagnostic markers, randomized controlled trials are hard to achieve due to the limitations on patient choice and the invasiveness of the gold standard test; many studies have adopted case-control study designs. However, the reproducibility of miRNA-based studies in the diagnosis of MM is conducive to the feasibility of clinical application.
Optimizing the flow of miRNA diagnostic tests according to the following suggestions would enable researchers to conduct systematic reviews and/or meta-analyses and draw more practical conclusions.

Patient Selection
"newly diagnosed or untreated patient" should be regarded as one of the criteria for the inclusion of diagnostic test, for circulating miRNA profile and level may change due to treatment. Pathological stage or IG isotype information of MM patients should be provided in detail, since the progression and Ig isotype of MM may affect circulating miRNA expression.

Study Design
Case-control study should not be perceived as a valid design for investigating the diagnostic value of miRNAs in MM patients. A cohort study may be a more applicable design initially, owing to the limited source of patients. Randomized controlled trials could be taken into consideration after the diagnostic power has been qualified.

Control Groups
To attenuate selection bias and avoid overestimation of diagnostic value, researchers should consider setting up control groups for suspected cases, individuals with MM precursor state, and patients with other conditions (such as inflammation,  (75). Besides, the incidence of MM is more prevalent in black race than in white, and higher in males than in females (76). Hence, the composition of age, ethnicity and gender of the control group should be consistent with that of the experimental group.

Sample Size
A sufficient sample size should be rigorously calculated in advance and can be achieved through collaboration between specialized institutions.

Specimen and Storage
Based on the long-term storage stability, circulating miRNAs have reliable performance as biomarkers. Both serum and plasma samples can be used to detect circulating miRNAs since there is no significant difference between serum or plasma-based tests. The researchers found that preservation at −20°C barely influenced the total amount of miRNAs for at least 2-4 years, with only slight changes in the concentration of individual miRNAs; in addition, storage at −80°C is even better (77).

Reference Standard
There are two widely used diagnostic criteria for patients with multiple myeloma and MUGS, one from the National Comprehensive Cancer Network (NCCN) and the other from the International Myeloma Working Group (IMWG) (78,79).
To avoid ambiguity and bias, researchers should clearly define the reference criteria.

Experimental Flow and Result Interpretation
Strict concealment measures should be observed throughout the study, including grouping, detection, and interpretation of test results. In addition the instruments, reagents, operating procedures, and the cutoff value of the test should be determined before the validation process and be detailed in the paper to avoid subjective bias and to ensure reproducibility. Except for the data on sensitivity, specificity and AUC, the direct presentation of true positive (TP), false positive (FP), true negative (TN), and false negative (FN) data would be of great benefit for future systematic evaluation.

CONCLUSION
Through the unremitting efforts of researchers, miRNAs have been confirmed to be implicated in many pathophysiological processes of MM; however, the exact regulation mechanism remains to be fully elucidated. Much attention has been given to the diagnostic value of circulating miRNAs in MM over the past decade. This meta-analysis reveals that miR-4254 has the best potential to be a biomarker for MM diagnosis, and miRNA cluster might be a good choice to optimize the utilization. Successfully unraveling the diagnostic value of circulating miRNAs in MM will depend on multicenter large-scale studies with rigorous process design and a broad enrolment in patient and control groups.

DATA AVAILABILITY STATEMENT
The original data presented in the study were derived from the included article/supplementary material. Further inquiries can be directed to the corresponding author.

AUTHOR CONTRIBUTIONS
JZ designed the study, assessed the quality of the manuscript. YX and PX were responsible for study searching, data extraction and meta-analysis implementation. LZ reviewed the results of each step and resolved any differences in the evaluation opinions of other authors. YX, PX, and LZ co-drafted the manuscript. All authors contributed to the article and approved the submitted version.

FUNDING
This project was supported by Youth Foundation of National Natural Science Foundation of China (81802075/H2003).