A High-Fiber Plant-Based Diet in Myeloma Precursor Disorders – Results from the NUTRIVENTION Clinical Trial and Preclinical Vk*MYC Model

Abstract

Consumption of a western diet and high body mass index (BMI) are risk factors for progression from pre-malignant phenotypes to multiple myeloma, a hematologic cancer. In the NUTRIVENTION trial (NCT04920084), we administered a high-fiber, plant-based diet (meals for 12 weeks, coaching for 24 weeks) to 23 participants with myeloma precursor states and elevated BMI. The intervention was feasible, improved quality of life and modifiable risk factors: metabolic (BMI, insulin resistance), microbiome (diversity, composition), and immune (inflammation, monocyte subsets). Disease-progression trajectory improved (n=2) or was stable. Findings were translated to Vk*MYC mice modeling the myeloma-precursor state, in which a high-fiber diet delayed disease progression through improved metabolism and microbiome composition leading to increased short-chain fatty acid production that reinvigorated anti-tumor immunity and inhibited tumor growth. These effects from fiber consumption were independent of calorie restriction and weight loss. A high-fiber diet is a low-risk intervention that may delay progression to myeloma.

Introduction

Multiple myeloma (MM) is a malignant plasma cell disorder often preceded by the precursor states of monoclonal gammopathy of unknown significance (MGUS) and subsequent smoldering myeloma (SMM). MGUS and SMM affect >3% of the general population >50 years.(1) The risk of progression to MM from MGUS is about 1% annually,(2) and from SMM is about 10% annually.(3) Because monitoring without treatment is standard practice for MGUS and SMM, low-risk interventions to reduce progression could be beneficial.

Obesity(4, 5) and diabetes mellitus(6, 7) are plasma cell disorder risk factors, and obesity increases the progression risk of MGUS.(8–10) Inflammatory and western diets are associated with metabolic disorders and linked to both MGUS(11) and MM in epidemiologic studies.(12–14) High-fiber, plant-based diets (HFPBD) may reduce the risk of metabolic-disorder-associated disease progression through mechanisms reducing inflammation.(15)

We hypothesized that an HFPBD intervention (NUTRIVENTION trial) would be feasible and promote weight loss, potentially favorably altering MM disease trajectory through metabolic-, microbiome-, and immune-based changes. For mechanistic insights, we administered a high-fiber diet (HFD) in transgenic Vk*MYC mice(16, 17) at the asymptomatic phase (murine SMM; mSMM),(18) as untreated mice invariably develop symptomatic murine MM (mMM), mimicking disease progression in humans.(19) We further studied this diet in a MM mouse model (transplant (t)-Vk*MYC) to separate the effects of weight loss from calorie restriction versus fiber.(20) To our knowledge, this is the first interventional trial to assess modifiable dietary risk factors in plasma cell disorders.

Results

A HFPBD is feasible, safe, and improves diet quality and patient-reported outcomes

This was a single-arm, single-center dietary intervention trial in patients with MGUS or SMM and BMI ≥25 (NCT04920084). Participants were shipped 12 self-selected frozen HFPBD meals (6 lunch and 6 dinner items) weekly for 12 weeks. They also received guidance for selecting snacks and breakfasts consistent with a HFPBD (Table S1; Fig S1a). The intervention had no calorie restriction.

Trial baseline characteristics were diverse including median age 62 years (range 40–79), 43% male, 43% from underrepresented populations (Black, Hispanic, or other), 74% obese, and 26% prediabetic/diabetic. Sixty percent had MGUS, and 40% had SMM (Table S2). The median time from diagnosis to baseline visit was 33.8 months (IQR 8.3–84.1) and from baseline to most recent timepoint within 20 months was 18.4 months (IQR 18.2–18.9). Twenty-three participants enrolled, 20 completed the 12-week intervention (13% dropout rate – two for unrelated medical reasons and one for nonadherence with study visits), and 18 completed week (W)52 with data collected at prespecified timepoints (Fig. 1a, S1b–c; Tables S1, S3).

Figure 1. NUTRIVENTION trial design, dietary adherence, and impact on health indicators.

a, Trial design. b, Dietary adherence boxplots by kcal percentage of minimally processed plant foods of total kcal of food consumed (median and interquartile range, IQR). c, BMI boxplots as measured by clinic weights and height (kg/m²) (median and IQR). d, Dietary fiber intake (g/1000 kcal) boxplots (median and IQR). e, Global health status by European Organisation for Research and Treatment of Cancer (EORTC) core Quality of Life Questionnaire (QLQ) C30 linear mixed model score. f. Difficulty following diet. g, Improvement in M-spike paraprotein trajectory in participants 1003 (circles) and 1004 (diamonds). Change in M-spike prior to intervention start (grey) and after intervention start (green). (*p ≤0.05, **p≤0.01, ***p≤0.001, ****p≤0.0001).

The study met its feasibility endpoint at W12, with dietary adherence (≥70% minimally processed plant foods) achieved by all participants and mean body mass index (BMI) ≥5% lower. As a proportion of all foods consumed, HFPBD adherence increased versus baseline at W12 (median 91% vs. 20%, mean 89% vs. 25%; p<0.0001), and remained significantly higher than baseline long term (Fig. 1b, Table S4). Participants had lower BMI by W12 (median −7%; mean −8%; p=0.0001) consistent with lower weight by W12 (median −7.7kg) and at subsequent timepoints (Fig. 1c, Table S4). At W12, 70% of participants achieved >5% BMI reduction, with 40% achieving >10% BMI reduction. The two participants lost to follow up after W12 had the highest BMIs at baseline (50.8 kg/m2 and 49.6 kg/m2). At 24 month follow up, many maintained BMI reduction independently (median −7%), with 71% (10/14) maintaining >5% BMI reduction (Table S3). Among participants with paired CT scans at baseline and W52 (BMI reduction <5% [n=3] or >5% [n=3]), all maintained muscle volume and most lost subcutaneous and visceral fat (Fig. S2a–c, Table S5). Analysis of serial bone marrow (BM) biopsies taken one year apart revealed a trend toward decreased BM adipocyte size with greater BMI loss (Fig. S2d).(21)

Participants increased their median total dietary fiber intake (+12.5 g/1000kcal/day; p<0.0001; Fig. 1d) and Healthy Eating Index score (+14.9 points; p<0.0001) by W12, maintaining higher measures than baseline at subsequent timepoints (Fig. S2e–f, Table S4). At W12, despite the recommendation to eat to satiety, participants had lower daily intake of calories (−478 kcal), fat, protein, sodium, added sugar, sugary beverages, and processed meats, and higher intake of vegetables, fiber, whole grains, and legumes (p<0.05) with no change in total carbohydrates. Participants did not report changes in physical activity (Table S4). Sample menus are provided (Table S6, Fig. S1a).

The diet was well tolerated, with no adverse events attributed to the intervention (Table S7). Participants reported significantly improved global health status/quality of life (p=0.04) (Fig. 1e) and decreased dyspnea (Table S8) by W52. Of the 15 participants who completed a feedback survey at W52 (2 lost to follow-up, 1 died of unrelated causes after W52, 2 did not respond), all self-reported that the intervention was very or somewhat easy to follow (Fig. 1f) and nearly all (14/15) would sign up again; and 13 had improvements in ≥1 symptoms or conditions, most frequently diabetes/pre-diabetes, cholesterol, body weight, self-confidence, and energy level (Table S9, Fig. S2g). Four of 12 participants using prescription medications self-reported stopping them (insulin at W4, bupropion at W4, potassium supplement at W12, hydroxychloroquine at W24), saving a median of $65 monthly (range $20–100). Comments about the intervention were generally positive. (Table S10).

Disease burden trajectory changes

No participants were on anti-myeloma drugs. In this exploratory analysis, given that pre-intervention myeloma paraprotein trajectory (≤3 months) was not available for all participants and low-burden disease (≤0.5g/dL) is not measurable for progression per IMWG criteria(22), trajectory change was evaluable in 8 participants.

Two individuals (1003 and 1004) had improved M-spike trajectory despite gradually progressive disease before the intervention, suggesting a higher risk for progression to MM, and >10% BMI reduction at W52 (Fig. 1g). The other 6 had stable disease (Table S2, Fig. S2h; Supplementary Results).

Additionally, we also evaluated change in paraprotein from baseline to W12 for all 20 participants. We find that 6 participants have a >10% reduction and 3 participants have >10% increase in paraproteins suggesting numerically greater participants had reductions. However, this may not be as clinically meaningful as paraprotein trajectories as levels can fluctuate although this is the period of highest dietary adherence.

Case Studies

Participant 1003 had Mayo low-intermediate risk IgGκ/IgGλ MGUS. In the 5 years before enrollment, he had a 15% BMI reduction (decrease from 34 kg/m²) secondary to a glucagon-like peptide-1 (GLP1) receptor agonist (liraglutide 2015–2021 and dulaglutide 2021 onwards) for diabetes management but a rising M-spike that increased 4-fold (0.24 to 1.21g/dL) during that period. At study baseline his BMI was 28.9 kg/m², and bone marrow plasma cell (BMPC) involvement <5% with a hyperdiploid genomic profile. After initiating the HFPBD on trial he was able to discontinue decades long insulin use at W4 and achieved an additional 19% BMI reduction to 23.5 kg/m² by W52. His M-spike trajectory stabilized (1.25 g/dL at W52) suggested slowing paraprotein progression following trial enrollment (+0.28 g/dL/y decreased to +0.02 g/dL/y; p=0.001), and a BM biopsy 1.5 years later showed 5–9% BMPC involvement (Table S11). These findings indicate that the effects of dietary fiber intake maybe separate from weight loss.

Diagnosed with MGUS 11 years prior to enrollment, participant 1004 had IMWG intermediate-risk IgGκ SMM and had a gradually rising M-spike (+0.11 g/dL/y) from the earliest measurement 5 years prior to enrollment (0.72 g/dL) until enrollment (1.2g/dL). She had 20%−30% BMPC involvement with hyperdiploid genomic profile and achieved a BMI reduction of 13.1% by W52 to 35.5 kg/m² after previously being stable (38.8 kg/m² at baseline and 37.2 kg/m² 5 years prior to enrollment). Her M-spike trajectory (−0.05g/dL/year; p=0.013) also suggested slowing paraprotein progression and a BM biopsy 1 year later showed 10–15% BMPC involvement. Individual biomarkers analyzed at baseline and after intervention at W12 and W52 for patients 1003 and 1004 are provided (Table S11).

Improved insulin resistance

Participants had decreased insulin resistance by W12, including lower insulin concentration (p=0.01; Fig. 2a) and higher adiponectin-to-leptin ratio (p=0.0003; Fig. 2b). Those with elevated hemoglobin A1c (≥5.7%) at baseline had lower levels by W12 (p=0.06), and those with elevated low-density lipoprotein cholesterol (>100 mg/dL) at baseline had lower levels by W12 (p=0.003; Fig. S2i). Participants’ insulin-like growth factor-1 (IGF-1) levels reduced gradually (p=0.008 at W52) (Fig. S2j, Table S12).

Figure 2. Effect of NUTRIVENTION trial on metabolic and microbiome biomarkers of disease.

a, Plasma insulin (mU/L) boxplots by ELISA immunoassay. b, Adiponectin-to-leptin ratio (total adiponectin [ng/mL]/leptin [pg/mL]) boxplots by ELISA immunoassay. c, Gut microbiome α-diversity boxplots by inverse Simpson index by 16S (genus level). d, Butyrate producer relative abundance boxplots by 16S (genus level). e-f, Association of fecal microbiome α-diversity with e, BMI and f, dietary fiber. g, Association of butyrate producers with dietary fiber. h, Carbohydrate active enzymes (CAZymes) that changed in relative abundance from baseline to W4 (p<0.1) on metagenomic sequencing. Trends persist in W12 samples. Boxplots show median and IQR. (*p≤0.05, **p≤0.01, ***p≤0.001).

Increased fecal microbiota diversity and altered composition

Fecal microbiome diversity (α-diversity), as assessed by 16S sequencing, increased by W12 (p=0.021) and remained higher than baseline at subsequent W24 and W52 timepoints (Fig. 2c). The relative abundance of taxa predicted to produce short-chain fatty acid (SCFA) butyrate increased by W12 (median 0.08 vs. 0.03; p=0.015) and remained higher than baseline (W24, 0.11 [p=0.14]; W52, 0.07 [p=0.055]) by 16S (Fig. 2d; Table S13). Microbiome β-diversity changed by W12 (Fig. S3a).

Fecal sample 16S α-diversity correlated negatively with BMI (Fig. 2e), and positively with dietary fiber (Fig. 2f), HEI-2020 score (Fig. S3b), and dietary adherence (Fig. S3c). Fiber intake (Fig. 2g), HEI-2020 score (Fig. S3d), and dietary adherence (Fig. S3e) correlated positively with higher relative abundance of butyrate producers, while BMI did not (Fig. S3f).

By W12 on the intervention, participants’ microbiome composition changed (Fig. S3g) and by 16S sequencing several genera had increased abundance (Monoglobus, Herbinix, Agathobaculum, Faecalibacterium, Roseburia, and Veillonella), while others had decreased abundance (Colinsella, Ruminococcus, Asaccharobacter, Dorea, and Actinomyces) (p<0.05) (Fig. S3h). By shotgun sequencing, certain eubacterium species and Monoglobus pectinilyticus had increased abundance, while Bilophila wadsworthia and others had decreased abundance (p<0.05) (Fig. S3i). Two butyrate producers were significantly increased—Roseburia hominis (p=0.00012) and Agathobaculum butyriciproducens (p=0.048) (Fig. S3j).(23) Additionally, there was a significant increase in acetyl coenzyme A synthesis pathways (an essential precursor of butyrate) from baseline to W4 (p=0.015) and W12 (p=0.014) (Fig. S3k) with a significant positive correlation with the relative abundance of known butyrate producers identified by both methods - 16S and shotgun sequencing (Fig. S3l–m).

We also observed increased abundance from baseline to W4 of multiple carbohydrate-active enzymes (CAZymes) that break down complex carbohydrates. By W12, CAZyme gene carriage decreased somewhat, but pectin (p=0.08), arabinoxylan (p=0.07), and arabinan (p=0.045) remained enriched compared to baseline (Fig. 2h; Table S13).

Further analysis revealed increased concentrations of fecal SCFAs (acetate, butyrate, propionate, and valerate) and decreased branch-chain fatty acids (isobutyrate, 2-methylbutryate [p=0.003], and isovalerate [p=0.01]) at W12, although these trends were not significant (Table S13).

Reduced inflammation and enhanced immunity

Seven participants with elevated baseline C-reactive protein achieved normalization by W52 (p=0.06) (Fig. 3a) and there was a significant reduction in the overall cohort by W52 (n=18; p=0.008; Fig S4a). Median absolute neutrophil count (p=0.007) and median neutrophil-to-lymphocyte ratio (p=0.04) decreased by W12 (Fig. 3b, Table S14). The frequency of several peripheral blood leukocyte subsets changed monthly until W24, including increased classical monocytes and CD33⁺ classical monocytes and decreased CD33^- non-classical monocytes, CD16^-CD86⁺ classical dendritic cells (DCs), and CD33^dim nonclassical DCs (p<0.05; Fig. 3c, Table S15). Using unsupervised hierarchical clustering analysis of relative concentrations of circulating inflammatory biomarkers at W12 compared to baseline, sixty-six percent (10/15) of participants clustered into a “non-inflammatory” category with primarily reduced biomarkers. BMI reduction of ≥5% at W12 was achieved in 80% of this group. In the “inflammatory” cluster (33% participants; n=5), 40% participants achieved ≥5% BMI reduction (Fig. S4b; Supplementary Results). Two participants in the inflammatory cluster who had weight loss >5% at W12 also had diagnoses of metastatic prostate cancer not on treatment (1021) and recent COVID-19 infection (1010).

Figure 3. Effect of NUTRIVENTION trial on immune biomarkers of disease in peripheral blood and bone marrow.

a, C-reactive protein (CRP) boxplots in participants with elevated baseline CRP (dashed line is normal concentration cut off <0.5mg/dL) (n=7). b, Absolute neutrophil count (ANC) (K/μL) boxplots. c, Peripheral Blood Leukocyte Subsets (% change/month) by flow cytometry between baseline and W24 in evaluable patients (n=15) (p<0.1). d, Cell proportions per patient and cell types (myeloid, CD8, and NK cells; logarithmically transformed) by scRNA-seq (n=13; paired samples baseline and W52). Gray and green colors indicate baseline and W52, respectively (Wilcoxon’s test). e, Gene Set Enrichment Analysis (GSEA) according to differentially expressed genes between baseline and W52. Normalized enrichment score (NES) goes from gray to green, if enriched at baseline or W52, respectively. Dot size corresponds to logarithmic-transformed p value, red highlighted when p≤0.1. f, Global interaction landscape between selected cell types: myeloid compartment (GMP, CD14⁺ and CD16⁺ monocytes, and MDSCs) and cytotoxic subset (cytotoxic and exhausted CD8 T and NK^dim/bright cells). Gray and green indicate whether interactions are increased at baseline or W52, respectively. g, TNF-specific interactions between those cell types, both for baseline and W52. The thicker the arrow, the higher the weight of this signaling interaction. Colors reference those used in the global UMAP (Figure S5a). Boxplots show median and IQR. (^#p≤0.1, *p≤0.05, **p≤0.01, ***p≤0.001).

BaMoP, basophil monocyte progenitor; cDC2, classical dendritic cells 2; GMP, granulocyte/monocyte progenitor; HSC, hematopoietic stem cell; MDSC, myeloid derived suppressor; NK, natural killer; pDC, plasmacytoid dendritic cells; TCSM, stem-like memory T cell; TEM, T effector memory; TEM/TEMRA, T effector memory/terminally differentiated effector memory; Tfh, T follicular helper; Th1, T helper 1; Th17, T helper 17; Treg, regulatory T; gdT, gamma delta T.

Exploratory single-cell RNA sequencing (scRNA-seq) analyses of BM mononuclear cell subsets and gene expression at baseline and W52 (N=51,910 cells, Fig. S4c; Table S16) revealed an increase in granulocyte monocyte progenitor (GMP) cells (p=0.04) and CD14⁺ monocytes (p=0.08) (Fig. 3d, S4c–e). We also performed subgroup analysis by BMI and α-diversity. We observed enrichment of GMP and CD14⁺ monocytes at W52 in the weight loss group (Fig. S4f, orange dots), enrichment of CD56^dim NK cells at W52 in participants with high α-diversity at W52, and enrichment of CD56^dim NK cells at baseline in participants with low α-diversity at W52 (Fig. S4g, red dot). We observed over-expressed genes in GMPs, CD14⁺ monocytes, CD16⁺ monocytes, and NK cells at W52 compared to baseline (Fig. S4h). Particularly, we found in subgroup analyses those participants who lost weight or with high α-diversity had more down-expressed genes in CD16⁺ monocytes compared to those whose weight remained stable or with low α-diversity (Fig. S4i–j).

Using gene set enrichment analysis on inflammation and anti-tumor-associated hallmarks, we identified a higher representation of genes associated with TNFα at W52 compared to baseline in the CD14⁺ monocytes (p≤0.1), as well as in GMP and NK CD56^bright cells, although those were not significant (Fig. 3e). The complement pathway, which has been associated with impaired T cell function, was enriched in exhausted CD8 T cells at baseline compared to W52 (Fig. 3e)(24). We found that participants who lost weight had increased TNFα in the CD14⁺ monocytes at W52 compared to baseline (p≤0.1), whereas weight-stable participants did not (Fig. S4k). Interestingly, at W52, we also found increased TNFα in NK CD56^dim and CD56^bright cells in patients with high α-diversity but reduced in NK CD56^dim cells in those with low α-diversity (Fig. S4l).

We also evaluated the global interaction landscape among the myeloid compartment, CD8⁺ T cells (exhausted and cytotoxic), and NK cells (CD56^dim and CD56^bright) using CellChat. The myeloid compartment interacted with exhausted CD8⁺ T cells at baseline but switched to interactions with CD8⁺ cytotoxic T cells and NK CD56^dim cells at W52 (Fig. 3f). Further TNF-specific analysis given the enrichment of the TNFα signaling pathway revealed a similar shift in myeloid and T/NK cell interaction at W52 (Fig. 3g).

Cytotoxic CD8⁺ cells sending and receiving signals increased, while GMP sending and receiving signals decreased (cell interaction heatmap, Fig S4m).

High fiber diet delays progression to MM and prolongs MM survival in mice

Having observed improved or stable M-spike trajectory associations in the single-arm clinical trial, we hypothesized that a diet rich in fiber delays SMM-to-MM evolution. To assess this, we administered a HFD (30% fiber; Table S17) or standard diet (SD, 3.9% fiber; Table S18) to transgenic Vk*MYC mice,(16) which exhibit a murine SMM (mSMM) phenotype at median 37 weeks of age (Table S19). Mice were monitored for disease progression to murine MM (mMM) from W0 until W30 (Fig. 4a, S5a–b). Progression to mMM (Fig. 4b) corresponded to serum M-spike ≥3g/dL (Fig. S5c) and anemia (Hgb <13g/dL; Fig. S5d). Only 56% (5/9) of HFD mice and all SD mice progressed to mMM by W30, a 44% absolute risk reduction and improved median progression-free survival (PFS) (30 weeks [HFD] vs. 12 weeks [SD]; p=0.013; Fig. 4b). Consistent with patients in the clinical trial (Fig. 1g), Vk*MYC mice fed HFD had improved M-spike trajectory (Fig. S5e). The mice were normal weight at baseline (median 27±7 g [HFD] vs. 29.7±3.6 g [SD]); mice on HFD initially lost weight that was regained by W25 (median 28.6±5 g), although the median weight of the HFD group remained lower versus the SD group (36.1±4.5 g) (Fig. 4c, Table S19).

Figure 4. High-fiber diet delays mSMM-to-mMM progression in Vk*MYC mice.

a, Experimental schedule: Vk*MYC mice were monitored for disease appearance and progression by serum protein electrophoresis (SPEP). When developing mSMM (M-spike appearance by SPEP, < 3g/dL), Vk*MYC mice were assigned to high-fiber (HFD) or standard diet (SD) (n=9 mice/group) and were monitored for progression to mMM. b, Progression-free survival (Kaplan-Meier plot) and number at risk of Vk*MYC mice (Log-rank (Mantel–Cox) test; n=9 mice/group). c, Body weight variation of Vk*MYC mice. Each dot represents the mean weight of the mice in the indicated experimental group (multiple unpaired Student’s t-test; n=9 mice/group). d, Experimental schedule for t-VkMYC mice: C57BL/6 mice were fed SD or HFD starting 10 days before tumor injection (Vk12598 MM cells). Disease appearance was monitored by SPEP. e, Percentage of M-spike-bearing (red; SD: n=20/34, 59%; HFD 6/24, 25%) and M-spike-free (grey) t-VkMYC mice 3 weeks after tumor challenge (Fisher’s exact test; data aggregated from 4 experiments). f, Overall survival since M-spike appearance of t-VkMYC mice fed SD (n=12) or HFD (n=10) (log-rank (Mantel-Cox) test). g, Mean body weight variation for each t-VkMYC mouse across 5 weeks of treatment. Data are reported as percentage of initial weight at inclusion (day −10). Each dot represents one mouse (unpaired t-test). h, Caloric intake per wild-type mouse receiving SD or HFD for 50 days (n=5/group, single housed), monitored every 3/5 days (multiple paired t-test; Holm–Šidák correction applied for multiple comparison correction). i, Mean body weight variation for each Vk*MYC mouse in (b) within week 0 to 6. Data are reported as percentage of week 0. Each dot represents one mouse. Black dots indicate mSMM mice not evolved to mMM at the end of observational period (one-way ANOVA). j, Caloric intake per wild-type healthy mouse receiving Ctrl or HFD (n=5 mice/group, single housed) (multiple paired t-test; Holm–Šidák correction was applied for multiple comparison correction). k. Experimental schedule for t-VkMYC mice fed Ctrl or HFD. l, Mean body weight variation for t-VkMYC mice receiving Ctrl (n=8) or HFD (n=7) across 5 weeks of treatment as percentage of initial weight at inclusion (day −10). m, Weight of adipose tissue collected at sacrifice. n, Colon length from the same mice collected at sacrifice. Each dot represents one mouse (unpaired Student’s t-test). o, Percentage of M-spike-bearing (red; Ctrl: n=6/8, 75%; HFD: n=6/24, 25%) and M-spike-free (gray) t-VkMYC mice 3 weeks after tumor challenge (Fisher’s exact test). Data aggregated from 3 experiments. p, Wild-type mice received a high-fat diet inducing obesity (DIO; n=15) for 11 weeks before switching to Ctrl (n= 8) or HFD (n=7). Serum insulin levels were measured at dietary switch (W11; T1) and 8 weeks after (W19; T2). q, Quantification of insulin levels in Ctrl and HFD-treated mice (paired t-test). r, Ratio T2/T1 insulin levels for each mouse (unpaired t-test). s, Body weight variation (multiple paired t-test; Holm–Šidák correction was applied for multiple comparison correction). (Error bars represent ± SD; *p≤0.05, **p≤0.01, ***p≤0.001 ****p≤0.0001).

We conducted similar experiments in mice challenged with bortezomib-resistant Vk12598 cells (t-VkMYC) to model relapsing/resistant MM (Fig. 4d).(25) HFD t-VkMYC mice had delayed disease development (Fig. 4e), and prolonged survival (Fig. 4f). Since survival is measured starting from M-spike appearance, these findings suggest that diets rich in fibers may also limit the aggressiveness of MM when patients are already on a HFPBD. Additionally, HFD t-VkMYC mice had lower weight (Fig. 4g) compared to SD t-VkMYC mice. Wild-type HFD-fed mice also ate less (Fig. S5f) and had lower energy intake compared to SD mice (median 6.3 vs. 12.2 kcal/day; Fig. 4h).

High fiber diet reduces body weight and serum insulin and attenuates disease severity independently from weight loss and calorie restriction in mice

These findings prompted us to investigate the association between weight loss and disease outcome (progression) in HFD transgenic Vk*MYC mice. We observed no significant weight variation between progressors and non-progressors during maximum weight loss (W0–6; Fig. 4i) and after weight stabilization (W7–20; Fig. S5g). Thus, in lean Vk*MYC mice, weight loss did not account for HFD anti-tumor effects.

To rule out a role of reduced energy intake in disease outcome, we fed wild-type healthy mice a HFD or an isocaloric control diet differing only in fiber content (Ctrl; Table S17). Both groups consumed comparable amounts of food (Fig. S5h) and calories (Fig. 4j). We then evaluated Ctrl diet and HFD in t-VkMYC mice (Fig. 4k). Mice on HFD had lower weight (Fig. 4l) and adipose tissue (Fig. 4m) versus mice on a Ctrl diet, demonstrating that fiber favors metabolic rewiring independent from calorie intake. HFD mice also had longer colon length (Fig. 4n), an indirect measure of improved gut health.(26, 27) Compared to Ctrl, the HFD t-VkMYC mice showed delayed disease appearance (Fig. 4o), suggesting that fiber impacts disease trajectory.

Because the NUTRIVENTION trial enrolled participants with elevated BMI, we also investigated whether a HFD had similar effects in t-VkMYC mice fed a high-fat diet inducing obesity (DIO; Table S17; Fig. S5i) and found consistent results (Fig S5j–m), suggesting weight loss and disease trajectory are independent. In the diet-induced obesity wild-type model (Fig. 4p), switching to a HFD reduced serum insulin levels compared to the isocaloric Ctrl diet (Fig. 4q–r). Because Ctrl and HFD had similar weight loss (Fig. 4s), our data support the evidence that the improved metabolic function is independent of weight loss. Supporting the limited role of calorie reduction in HFD benefit, we found that the SD and Ctrl diets (Table S18), differing in caloric load (Fig. S5n), altered neither body composition nor disease aggressiveness in t-VkMYC mice (Fig. S5o–s).

Altogether, we demonstrated the importance of dietary fiber in delaying progression and that calorie restriction alone did not modify mMM trajectory.

High fiber diet alters microbiome and metabolism in Vk*MYC mice and modulates immune and MM cells by increasing SCFAs in t-VkMYC mice

Consistent with human data (Fig. 2c–g), a HFD modulated the composition of gut microbiota (Fig. 5a–b, S6a) by expanding SCFA-producing commensals (Fig. 5c). Because defining SCFA-producing commensals at the genus level may not be accurate, we directly investigated the content of SCFAs in stool from mice exposed to different dietary regimens. Stool acetate, butyrate, and propionate were enriched in HFD mice at W4 (Fig. 5d).

Figure 5. High-fiber diet modulates gut microbiota composition and improve immune response in Vk*MYC mice.

a, Principal coordinate analysis and visualization of β-diversity between the microbiome of standard (SD) versus high-fiber (HFD)-treated mice at baseline and W4 after enrollment. Boxplot shows distances between each sample in the high fiber group and each other sample (PERMANOVA p=0.3 [baseline]; p=0.004 [W4]). b, Baseline to W4 relative abundance (top families) by 16S sequencing for HFD mice. c, Baseline to W4 relative abundance (%) of SCFA-producing bacteria by 16S sequencing in HFD mice (multiple paired t-test). d, Fecal short-chain fatty acid (acetate, butyrate, and propionate) quantification by nuclear magnetic resonance (NMR) from the feces collected 4 weeks after enrollment (unpaired Student’s t-test; n=6 mice/group). e, Neutrophil-to-lymphocyte ratio (NLR) in the peripheral blood of Vk*MYC mice at sacrifice. f, Flow cytometry quantification of immune cells in the bone marrow collected at sacrifice from the same mice described in (a). Each dot represents one mouse. Black dots indicate mSMM mice not evolved to mMM at the end of observational period (unpaired Student’s t-test).

At sacrifice (disease progression or at W30), HFD mice had reduced neutrophil-to-lymphocyte ratio (Fig. 5e) and had comparable BM and spleen cellularity (Fig. S6b,c) versus mice fed SD but had reduced BM Th17 cells, possibly from reduced DCs producing the pro-Th17 cytokine IL-6, reduced bona fide monocytic-myeloid derived suppressor cells (M-MDSCs), and exhausted CD4⁺TIM3⁺PD-1⁺ T cells, as well as increased effector CD8⁺ T cells (Fig. 5f. These findings were restricted to BM and are independent from tumor burden at sacrifice (Fig. S6d), suggesting that a HFD selectively rebalances T lymphocytes within the tumor microenvironment, from pro-tumoral to anti-tumoral. Indeed, a HFD expands IFN-γ⁺ CD8 T cells in the BM of healthy mice (Fig. S6e), supporting the role of dietary fibers in inducing fitter T cell immunity.

Because the HFD promoted the production of SCFAs by the gut microbiota, we investigated whether SCFAs account for the beneficial effect of this diet on mMM in t-VkMYC mice (Fig. 6a). While SCFA supplementation did not alter body weight, adipose tissue, or colon length (Fig. S7a–c), it altered disease aggressiveness (Fig. 6b). Similar effects were observed in obese mice with a trend towards tumor delaying effect of SCFAs in the obese model (Fig. S7d–h). However, SCFA supplementation was not sufficient to reduce serum insulin levels in diet-induced obese mice (Fig. 6c–e). To investigate the mechanistic effects of SCFAs on immune cells, human peripheral blood monocytes were differentiated into dendritic cells in vitro.. Consistent with the HFD (Fig. 5f), SCFAs also limited the activation of human monocyte-derived DCs in response to inflammatory stimuli (Fig. 6f, S7i) without affecting their viability (Fig. S7j). SCFAs also inhibited neoplastic plasma cell proliferation (Fig. 6g). Butyrate accounted for most of the effect of the SCFA mix on mMM cell growth (Fig. 6h, S7k, l). We found similar results when human RPMI8226 MM cells were cultured with butyrate (Fig. S7m).

Figure 6. SCFAs limits disease aggressiveness but are not sufficient to improve metabolic parameters in mice.

a, Experimental schedule for t-VkMYC mice receiving Ctrl diet and plain water (Ctrl) or water supplemented with SCFAs (93 mM acetate; 26.6 mM propionate; 13.3 mM butyrate) (n=8 mice/group). b, Percentage of M-spike-bearing (red: Ctrl: n=6/8, 75%; SCFAs: n=1/8, 12%) and M-spike-free (gray) t-VkMYC mice 3 weeks after tumor challenge (Fisher’s exact test). c, Experimental schedule for diet-induced obese (DIO) mice: wild-type mice received a high-fat diet inducing obesity (DIO; n=16) for 11 weeks before switching to Ctrl (n= 8) or HFD (n=8). Serum insulin levels were measured at dietary switch (W11; T1) and 8 weeks after (W19; T2). d, Quantification of insulin levels in Ctrl and SCFAs-treated mice (paired t-test). e, Ratio T2/T1 of insulin levels for each mouse (unpaired t-test). f, Percentage of human HLA-DR⁺PD-L1⁺ cells gated in CD1a⁺ cells cultured in the presence of indicated SCFAs concentration (acetate:propionate:butyrate 70:20:10 ratio) after an overnight stimulation with E. coli-derived LPS. Data representative of 4 independent experiments with different human donors (one-way ANOVA). g-h, Proliferation index measured through the dilution of CellTrace^™ Violet dye over time compared to t0. (g) Cells conditioned with indicated SCFA concentrations. (h), Cells conditioned with indicated doses of single SCFAs. Each dot represents the mean of three biologically independent experiments each one performed in triplicate (RM-one-way ANOVA). (Error bars represent ± SD; *p≤0.05, **p≤0.01, ***p≤0.001 ****p≤0.0001).

We investigated the anti-tumor activity of a HFD in immunodeficient NSG mice(28) given that fiber-rich diets modulate immunity and its metabolite, butyrate, both modulates immunity and directly affects tumor cells (Fig. S7l, m). The overlapping overall survival curves observed in t-VkMYC NSG mice fed either the Ctrl diet or HFD (Fig. S7n) strongly supports the conclusion that anti-tumor immunity is the most relevant mechanism through which the benefits of a high fiber diet are mediated.

Altogether, the findings demonstrate that a HFPBD exerts anti-tumor activity through mechanisms including improving metabolic fitness, modulating gut microbiota, and releasing SCFAs that directly affect immune cells and neoplastic plasma cells (Fig. 7a–b).

Figure 7. Summary of findings from the NUTRIVENTION dietary interventional trial and preclinical mouse model in precursor plasma cell disorders.

a, Key results. b, Graphical abstract.

Discussion

In this clinical and preclinical study, we demonstrate that diet is a modifiable risk factor that can improve disease biomarkers and potentially delay disease progression to MM, respectively. We provide mechanistic insights into how a HFPBD affects disease trajectory, through metabolic, microbiome and immune mediated mechanisms. This is the first dietary interventional study in plasma cell disorders. Our results highlight the importance of further clinical study of targeted dietary modifications in patients with plasma cell disorders, 90% of whom report having questions about their diet.(29)

Participants treated with the dietary intervention achieved a significant, sustained BMI reduction, reporting satiety despite consuming fewer calories, which is attributable to the satiating effect of dietary fiber.(30) Despite weight loss, participants maintained muscle and lost fat volume, consistent with findings in healthy individuals.(31) Although the participants in our study had a baseline median fiber intake (21g/d),(29) higher than the US average (16g/d), benefits were still observed as participants’ median fiber consumption (35g/d) on intervention exceeded USDA guidelines and in some participants reached 58g/d.(32) It was well tolerated. This clinical trial was not statistically powered to confirm that a HFPBD changes MM natural history, but 2 participants with progressive disease trajectory for 5 years prior to intervention had stabilization of disease on and post intervention suggesting slowing in the rate of progression. Disease remained stable in all other evaluable participants. Most participants sustained dietary changes long-term beyond the intervention, challenging the common belief that a high fiber diet is difficult to maintain. Another benefit was cost savings from discontinuing prescription medications.

We used the Vk*MYC mouse model to evaluate the effect of the intervention on disease outcomes.(16) The mSMM mice fed a high fiber diet had improved median PFS, and 44% never evolved to mMM. We confirmed in lean and obese mouse models that a high fiber diet induced weight loss independent of calorie restriction and because of fiber consumption. We also demonstrated that a HFPBD is more effective than calorie restriction for delaying disease progression. We further showed that the effects of a high fiber diet on tumor growth in t-VkMYC mice (mMM) extend beyond MM prevention to improved survival post diagnosis. This is consistent with an observational study in patients with MM showing that healthier pre-diagnosis dietary patterns correlate with longer survival.(33)

Low circulating adiponectin and high leptin concentrations are risk factors for progression to MM and a higher adiponectin-to-leptin ratio implies greater insulin sensitivity.(4) We previously demonstrated that insulin is a potent MM growth factor in mouse models and that diabetes is associated with worse survival in MM.(34) Here we show that mice receiving the high fiber diet had lower insulin concentrations. Although participants in our trial had no change in carbohydrate intake, we observed improved fasting insulin concentrations and adiponectin-to-leptin ratio, likely because the diet provided complex fiber-rich carbohydrates. These metabolic findings were not seen with SCFA supplementation in mice and suggest that a HFPBD may improve metabolic markers that are negatively associated with disease progression to MM.

Whereas a prior study of a high fiber diet in healthy individuals reported stable fecal microbiome diversity,(35) we found increased diversity beyond the intervention, attributable to changing participants’ dietary patterns rather than solely instructing them to increase fiber. To our knowledge, this is the first study showing an increase in gut microbiome diversity after a HFPBD in a precancer or cancer population as well as the first study to show a sustained increase in gut microbiome diversity that was maintained at W52. Consistent with our findings, that study found increased abundance of genes encoding carbohydrate-utilization enzymes on a high fiber diet.(35) In healthy populations, there is evidence that diet shapes the microbiome and rapidly induces microbial shifts,(36) although long-term data are limited.(37) We demonstrated changes in microbiome diversity and composition up to 1 year through sustained dietary changes. Consistent with our trial, HEI in healthy individuals correlates with microbiome composition and taxa linked to healthy plant-based foods, including butyrate producers Roseburia hominis and Agathobaculum butyriciproducens.(38) Additionally, gut bacteria accelerate MM progression in mice(19); and plant-based proteins correlate with microbiome diversity, butyrate production, and response in MM.(39)

Bacteria facilitate the production of SCFAs that have anti-inflammatory and anti-tumor effects.(40) Our findings indicate that butyrate exerts a direct antitumor effect, consistent with findings in other tumor types.(41, 42) We found that SCFAs reduced the inflammatory function of monocyte-derived DCs, likely favoring a more beneficial immune response as discussed below. Microbiota-dependent SCFA production may be one beneficial mechanism of a HFPBD but not sufficient to reduce adipose tissue or insulin, highlighting the superiority of a HFPBD for metabolic health(43) and its anti-tumor role in MGUS/SMM compared with postbiotic supplementation.

Increased CD16⁺ monocytes and decreased classical monocytes (CD14⁺) have been identified in the BM in MM patients by scRNAseq previously.(44) This skewing increased with disease burden(45) and decreased with treatment response.(46) Trial participants had reduced levels of C-reactive protein, neutrophils, neutrophil-to-lymphocyte ratio, inflammatory non-classical monocytes (CD16⁺) and increased classical monocytes (CD14⁺) after HFPBD intervention. Consistent with human data, fiber supplementation reduced neutrophil-to-lymphocyte ratio and BM CD11b+Ly6C^hi monocytes in mice.(47) Thus, a HFPBD could improve this monocyte imbalance associated with MM. Moreover, healthy individuals whose diets include fermented foods have lower inflammation and fewer non-classical monocytes, consistent with our data.(35)

In humans, this is the first study to identify a global shift in BM interactions of these myeloid cells with exhausted CD8⁺ at baseline to cytotoxic CD8⁺ a year later, with these changes mirrored in TNF specific signaling interactions, suggesting anti-tumor potential through TNFα release by CD14⁺ monocytes to signal CD8⁺ T cells for cytotoxic activity.(48) In mice we found reduced BM Th17 cells, possibly from reduced DCs that produce the pro-Th17 cytokine IL-6. This data mirrors the anti-inflammatory activity SCFAs showed on human monocyte-derived DCs in vitro, strengthening the hypothesis that SCFAs are central to immune modulation. We also found reduced CD4⁺TIM3⁺PD-1⁺ T cells, as well as increased effector CD8⁺ T cells. These results suggest a less immunosuppressive tumor microenvironment, consistent with human results showing a switch in myeloid and T/NK cell interactions at W52 as well as a reduction in DC subsets. These data confirm in mice that the beneficial effects of a high fiber diet are also immune mediated. Our findings are relevant for patients with SMM because BM exhausted PD1⁺TIM3⁺ T cells are associated with disease progression.(49) Additionally, DCs induce expansion of Th17 cells which have pro-MM function in prior studies.(19, 50) Consistent with our findings healthy individuals following a vegan diet have increased CD4⁺ and CD8⁺ effector T-cell responses in a prior study.(51)

Strengths of this study include assessment of a dietary intervention in a clinically relevant population with precancer, diverse racial enrollment, comprehensive biomarker evaluation, and serial BM sampling. Provision of meals without calorie restriction, in addition to counseling, encouraged long-term adherence and behavior change; these patients may also have been more motivated because of the presence of a precancerous state. This study had lower participant burden and greater potential for generalizability and scalability than controlled feeding studies. We also performed BM scRNAseq that has not been previously reported during a dietary intervention. Additional strengths include mechanistic data from preclinical models demonstrating the benefits of a high fiber diet through metabolic rewiring go beyond isolated short chain fatty acid production and calorie restriction induced weight loss.

Limitations include the small sample size trial enrollment bias, and lack of control arm to confirm diet driven changes. The reliance on self-reported dietary adherence data was addressed by collecting multiple recall surveys, as well as food frequency questionnaires which is considered the preferred method. Weaknesses of the mouse model include differences in human and mouse microbiome and metabolism. The dietary interventions used in patients and mice were not completely matched because of technical limitations in rodent dietary formulations. Despite differences in mice and human species we identified consistent beneficial effects.

These findings have the potential for widespread implementation at an individual, institutional and community level. They highlight the need for comprehensive and accessible nutrition education models targeting populations that are most likely to benefit.(52) Medically tailored meals (MTMs) are prepared, home-delivered meals for patients with complex health conditions referred by a medical professional. These are gaining attention with evidence of early adoption in some states and hold promise to improve health outcomes, reduce financial burden, address disparities, and reduce cost of health care.(53, 54) While a fiber supplement may seem like an easier solution it does not provide the wide range of beneficial nutrients that are unique to plant foods such as phytochemicals, antioxidants, vitamins and minerals. Additionally, supplements are typically made of one type of fiber limiting the effects on microbiome diversity and selecting for fewer species(55) although this is a research question that has not been studied specifically in patients with precursor plasma cell disorders to date.

In sum, we demonstrate that high fiber dietary interventions are feasible and provide multifaceted benefits in patients with plasma cell disorders including the potential to favorably alter disease trajectory. These findings support the rationale for a randomized clinical trial of targeted dietary intervention in patients with myeloma and its precursor conditions (NCT05640843).

Methods:

Human Clinical Trial

Trial design:

We conducted a pilot, single-arm, single-center dietary intervention trial (NUTRIVENTION) in patients with MM precursor disorders (MGUS or SMM) and BMI ≥25 (overweight or obese) (NCT04920084).(56) MGUS/SMM could be diagnosed any time prior to the study. The composite primary endpoint was feasibility defined as mean dietary adherence of ≥70% minimally processed plant foods and mean BMI reduction ≥5% at W12. Secondary endpoints included safety, quality of life (QoL), dietary pattern, myeloma markers, and metabolic markers. Exploratory endpoints included microbiome and immune markers. Participants were replaced if they did not complete the 12-week intervention to ensure that 20 participants completed the intervention and were evaluable. Blood and stool samples were collected at baseline, W4, W12, W24, and W52. Bone marrow aspirate and biopsy, as well as imaging (whole-body MRI or CT or PET CT), were performed at baseline and W52.

Ethics statement:

The study was conducted in accordance with ethical guidelines (Belmont Report and Declaration of Helsinki) and approved by Memorial Sloan Kettering (MSK) institutional review board. Written informed consent was obtained from all participants.

Inclusion criteria:

Inclusion criteria were as follows: BMI ≥25; confirmed diagnosis of MGUS or SMM; M-spike (immunoglobulin) ≥0.2 g/dL or abnormal free light chain ratio with increased level of the appropriate involved light chain; secretory MGUS or SMM; age ≥18 years; willingness to comply with all study-related procedures; Eastern Cooperative Oncology Group (ECOG) performance status 0–3; and interest in learning to cook plant-based recipes.

Exclusion criteria:

Exclusion criteria were as follows: already following a HFPBD or vegan diet (ovo-lacto-vegetarian diets were allowed); legume allergy; severe allergies, such as anaphylactic shock to peanuts; concurrent participation in weight loss/dietary/exercise programs; mental impairment leading to inability to cooperate; enrollment onto any other therapeutic investigational study; concurrent pregnancy; known diagnosis of diabetes mellitus only if not regularly following up with an endocrinologist/primary care physician during the trial period; positive hepatitis B virus, hepatitis C virus or human immunodeficiency virus polymerase chain reaction testing; non-English speaking; grade ≥2 electrolyte abnormalities as defined by CTCAEv5.0 that could not be resolved; and any reason that, in the opinion of the investigator, led to a concern regarding the ability of the patient to complete the study safely.

Intervention:

Participants received a HFPBD consisting of 12 self-selected frozen meals (6 lunch and 6 dinner items) weekly, provided by a US-based company called Plantable for 12 weeks. They also received guidance for selecting snacks and breakfasts consistent with a HFPBD. The intervention had no energy restriction, and participants were encouraged to eat to satiety within the study guidelines. All participants received the American College of Lifestyle Medicine (ACLM) whole food plant-based plate handout, and the American Cancer Society (ACS) Diet and Activity Guidelines.(57) A research dietitian at our center and health coaches available through Plantable provided nutrition coaching to participants for 24 weeks (Fig. 1a).

The meals provided had a low glycemic index and contained vegetables, whole grains, and plant-based fats that had undergone minimal processing. Instructions were provided for food storage and reheating. Additional dietary guidance included an emphasis on whole grains, legumes, nuts, seeds, fruits, and vegetables. The intervention was non-energy restricted, with participants encouraged to eat to satiety within the study guidelines. Participants were asked not to count calories but keep a food log for accountability. Participants were encouraged to avoid refined grains, animal products, added sugar, and highly processed foods during intervention period. Participants also had access to an app from Plantable that provided plant-based meal recipes and menu planning. Participants met with the Plantable health coach weekly through telephone visits and had virtual telehealth nutrition counseling sessions with the research dietitian that included clarification of food records every 2 weeks. Participants were asked to take a weekly vitamin B12 supplement of 1000 mcg while on this diet. Vitamin D concentrations were maintained over 30 ng/mL with supplementation if needed. Further details about the foods allowed and intervention structure are provided in Table S1.(58)

Clinical Endpoints and Analyses

Adverse events:

Adverse events were recorded by clinical trial nurses at study visits in accordance with Common Terminology Criteria for Adverse Events (CTCAE v5.0) criteria.

Dietary data collection:

Participants were asked to self-report their intake through food records for 3 days before dietitian visits, which occurred every 2 weeks during the dietary intervention, and at W13, W18, W24, and W52 visits. Dietary adherence was calculated by analyzing 72-hour food records and calculated as the percentage of kcal from whole, minimally processed plant foods out of the total kcal consumed.

Whole, minimally processed plant foods were defined by the professional opinion of the registered dietitian. Foods were considered highly processed if they included substances not normally used in home food preparation, such as hydrolyzed proteins, hydrogenated fats, added sweeteners (both artificial and natural, caloric and noncaloric), bulking agents, artificial flavorings, and such. Examples of processed plant foods included pre-packaged meat substitutes, sweetened plant-based dairy substitutes, sugar-sweetened beverages, pre-prepared baked goods, deep-fried foods, and plant-based cheese and butter alternatives.

All food records were analyzed through the Automated Self-Administered 24-hour (ASA24^®-US-2020) Dietary Assessment Tool developed by the National Cancer Institute, Bethesda, MD.(59) Plantable recipes were provided by the company and analyzed using the Nutrition Data System for Research software version 2023, developed by the Nutrition Coordinating Center (NCC), University of Minnesota, Minneapolis, MN.(60–62) Additional dietary collection was through the full-length 2014 Block Food Frequency Questionnaire (FFQ), a 127-item questionnaire, completed at baseline, W12, W24, and W52.(63) Food records were collected for detailed absolute intake while FFQ was collected to assess dietary pattern changes overtime without the influence of specific day fluctuations in intake and recall bias. The Healthy Eating Index Score–2020 (HEI-2020) was calculated from the combined food record data (ASA24 and NDSR Recipe Files) and Block FFQ as a measure of diet quality and ranged from 0–100, with a higher score indicating better diet quality.(64)

Weight and activity measurements:

For body mass index (BMI) compliance, participants were asked to send weekly self-reported weights through W24 and at the W52 timepoint. If participants did not have a scale available, weights were taken at clinic visits. Weight change was assessed at each timepoint compared to baseline.

The Godin Leisure-Time Exercise Questionnaire (GLTEQ) was used to assess activity at baseline, W12, W24, and W52.(65) The GLTEQ was verbally asked by the research dietitian at the corresponding visit and comprised three questions that assessed the average frequency of mild, moderate, and strenuous exercise in a typical week. The weekly frequencies were used to calculate a numeric total weekly leisure activity score, with 24 or more units indicating active, 14–23 units indicating moderately active, and less than 14 units indicating sedentary.

Post-intervention feedback survey:

Participants who completed the 12-week intervention received a voluntary feedback survey with 30 questions about their time on the diet, barriers to adopting the diet, self-rated health and quality of life, food and beverage consumption, and feedback regarding the intervention. Survey answers were collected and managed using REDCap electronic data capture tools hosted at MSK.(66, 67) All participants who completed W52 on the study were sent the survey.

Quality of life:

Participants completed the European Organization for Research and Treatment of Cancer (EORTC) Quality of Life Questionnaire (QLQ) C30 version 3 at baseline, W4, W12, W24, and W52. The survey was analyzed and scored per standard scoring guidelines and criteria.(68)

Metabolic markers:

Fasting plasma total adiponectin, leptin, insulin, and insulin-like growth factor 1 (IGF-1) were measured by ELISA immunoassays at baseline, W4, W12, W24, and W52 by the Pollak Laboratory at McGill University in Canada. Hemoglobin A1c (HgbA1c), C-reactive protein, and lipid profile were measured using standard laboratory assays at the same timepoints.

Myeloma markers:

The standard myeloma paraproteins — quantitative immunoglobulins, free light chains, serum, and urine protein electrophoresis (M-spike)— were checked at baseline, W4, W8, W12, W24, and W52 on the study.(69) All paraprotein M-spike and free light chain assessments available prior to study intervention and after W52 (up to 20 months before or after baseline) were documented. Serum paraprotein (M-spike in g/dL or free light chain ratio based on disease subtype) trajectory, a measurement of myeloma disease burden over time, was measured by rate of change per month for 20 months before and 20 months after the intervention from baseline. Participants with limited measurements before enrollment (≤3 months) (1009, 1011, 1013, 1014, 1018, 1023) or low disease burden (non-measurable disease per IMWG criteria(22)) at baseline (≤0.5g/dL) (1008, 1010, 1012, 1013, 1018, 1021, 1023) were excluded from trajectory analysis. Participant 1015 had received chemotherapy for breast cancer 14 months prior to baseline that led to improvement of pre-intervention M spike and was also excluded. Data for 2 participants lost to follow up at W12 (1007 and 1010) was excluded. Based on these criteria, disease burden trajectory changes were evaluable in 8 participants (1001, 1003, 1004, 1005, 1006, 1017, 1019, 1022) (Fig S2h).

Body composition:

Low-dose computed tomography scans were obtained from baseline and W52 PET-CT imaging studies. Digital imaging and communications in medicine (DICOM) series of these scans were then processed as previously described using nnU-net, TotalSegmentator and Comp2Comp deep-learning workflows for automated CT segmentation (arXiv 2023.02.23. 06568).(70, 71) Tissue compartment volumes for subcutaneous adipose tissue (SAT), visceral adipose tissue (VAT), intermuscular adipose tissue (IMAT), and muscle tissue (MT) were measured and then summed over T12 and L1-L5 vertebral levels. Muscle percent composition was measured as MT/(MT+SAT+VAT+IMAT) after summation across all vertebral levels.

Bone marrow adipocytes:

Diagnostic bone marrow core biopsies, fixed in formalin, decalcified, and embedded in paraffin, underwent staining with hematoxylin and eosin (H&E). Utilizing a Zeiss Axioscan 7 (Carl Zeiss, Oberkochen, Germany), whole-section digital scans were captured, followed by analysis using the HALO software image analysis platform for quantitative tissue assessment in digital pathology (Indica Labs, Albuquerque, NM, USA). The entire hematopoietic marrow encompassed in the biopsies was scrutinized, with manual exclusion of areas displaying significant artifacts, distortion, or fragmentation that could potentially affect the adipocytes’ size and shape. Adipocytes within the preserved regions were assessed for various metrics, including diameter (μm), perimeter (μm), area (μm²), roundness, and circularity.

Statistical analyses for clinical endpoints:

For each time point, categorical participant characteristics were summarized by frequency, and continuous characteristics were summarized by the median and interquartile range (IQR). Wilcoxon signed-rank test was used to determine the significance of the change in clinical and molecular biomarker measurements from baseline. EORTC QLQ-C30 scores and time from the start of the treatment were estimated using a linear-mixed effect model with a random intercept. The change in monoclonal (M)-spike concentrations as rate of change of M-spike per year (y) measured by slope with 95% CI was calculated for up to 20 months pre-intervention and for up to 20 months from intervention start (baseline) for all values available. A p value for difference in M-spike rates was calculated. P values were not adjusted for multiple testing (unadjusted P < 0.05).

Microbiome Endpoints and Analyses

16S sequencing data processing:

The quality of the raw pooled reads was assessed using FastQC (https://www.bioinformatics.babraham.ac.uk/projects/fastqc/), and then demultiplexed at the sample level. DADA2 was used to further quality trim and filter these reads before using them to call amplicon sequences variants (ASVs).(72) These ASVs were then taxonomically annotated using blastn to compare the sample reads to a 16S ribosomal RNA database. The pipeline used to perform all 16S processing can be found on the public GitHub repository (https://github.com/vdblab/vdblab-amplicon).

Shotgun sequencing data processing:

To prepare for downstream analysis all raw FASQ files underwent deduplication and adapter trimming by bbmap. These reads then underwent a two-step alignment process using bowtie2 and snap aligner 2.0 to the human genome to exclude host-mapping reads. The exact version of the software used can be accessed in the docker files listed below. These steps can be recapitulated by running the “preprocess” stage of our publicly available Snakemake pipeline (https://github.com/vdblab/vdblab-shotgun/tree/main).

16S and shotgun sequencing analyses:

Stool samples were collected and analyzed as detailed previously.(39) The relative abundance of predicted butyrate-producers and microbiome α-diversity were calculated from 16S microbiome profiles in samples collected at baseline, W4, W12, W24, and W52.(73, 74) Predicted butyrate producing bacteria previously identified and published were included.(39, 73–75) The sequencing depths of the 16S samples ranged from 5574 to 1426269, with a median of 87090, which is generally considered adequate for capturing community diversity in human microbiome studies.(76) Additionally, shotgun metagenomic sequencing using baseline, W4, and W12 samples was performed to evaluate species-level change.

Shotgun profiling and microbial pathway identification:

Preprocessed reads were taxonomically profiled using MetaPhlan3. These reads and profiles were then passed through HUMAnN3.0, a pipeline for profiling the presence and abundance of microbial pathways.(77) These steps can be recapitulated by running the “biobakery” stage of our publicly available Snakemake pipeline (https://github.com/vdblab/vdblab-shotgun/tree/main).

CAZyme identification:

Preprocessed host-depleted metagenome reads were assembled using MEGAHIT. Genes were predicted for this assembly using Prokka. run_dbcan(78) was used to annotate these genes as CAZymes and identify potential substrates. Bwa was used to align host-depleted reads to the genes generated by Prokka in order to get an estimate of the Reads per Million for each identified CAZyme. Limiting our analysis to polysaccharide lyase (PL) and glycoside hydrolase (GH) family CAZymes, we then explored which CAZymes were most different in their abundances between baseline and W4. CAZymes identified as most changed (p<0.1) between the baseline and W4 timepoints were plotted and compared to the W12 timepoints.

Exact Version of Docker Files for each software used –

bbmap (docker://staphb/bbtools:38.97),

bowtie2 (docker://ghcr.io/vdblab/bowtie2:2.5.0),

snap aligner (docker://ghcr.io/vdblab/snap-aligner:2.0.1),

MEGAHIT (docker://ghcr.io/vdblab/megahit:5f329c6d24a1480b75145a4c14567a25453b95bf), run_dbcan (docker://baichom1/dbcan3:4.1.4.12),

Prokka (docker://staphb/prokka:1.14.6),

bwa (docker://staphb/bwa:0.7.17).

ShortBRED analysis:

The amino-acid sequences for a collection of Acetyl CoA-pathway butyrate-producing genes(73) was downloaded from the Integrated Microbial Genomes and Microbiomes database(79) and used to create a ShortBRED index.(80) ShortBRED was then used to quantify the relative abundance of these genes in shotgun data in units of reads per kilobase million (RPKM).

Short-chain fatty acids:

Stool short-chain fatty acid concentrations were quantified using gas chromatography-mass spectrometry, at baseline, W4, W12, W24, and W52.(39)

Fecal samples extraction:

Fecal samples (~120 mg) were weighed into 2 mL microtubes containing 1.4 mm ceramic beads (Omni International) and resuspended to a final concentration of 100 mg/mL using 80:20 methanol:water containing acetate-d3, propionate-d5, butyrate-d7, and valerate-d9 as internal standards (Cambridge Isotope Laboratories). Samples were homogenized using a Bead Ruptor (Omni International) at 6 m/s for 3 minutes at 4°C, then centrifuged for 20 minutes at 20,000 × g at 4°C.

Samples derivatization:

100 μL of the sample extracts (fecal) was added to 100 μL of 100 mM borate buffer (pH 10). 400 μL of 100 mM pentafluorobenzyl bromide (PFBBr, Thermo Scientific) diluted in acetonitrile (Fisher), and 400 μL of cyclohexane (Acros Organics) were then added and the reaction vials sealed and heated to 65°C for 1 hour with agitation. Samples were then cooled to room temperature and centrifuged to promote phase separation, and the cyclohexane (upper) phase is transferred to a new vial. Samples were analyzed at 1:100 dilution (prepared with cyclohexane). Calibration curve levels and quality control (QC) samples were prepared in borate buffer (0.1–100 mM for fecal samples and 0.24–120 μM for plasma samples). GC-MS analysis used an Agilent 7890A GC and Agilent 5975C MS detector operating in negative chemical ionization (nCI) mode. Methane was used as the CI reagent gas at 2 mL/min, and a 1 μL splitless injection was made onto a VF-1701ms column (30 m × 0.25 mm, 0.25 μm; Agilent Technologies). For SCFAs quantification, the raw peak areas of acetate (m/z 59) and propionate (m/z 73) were normalized to acetate-d3 (m/z 62) and propionate-d5 (m/z 78), respectively; butyrate and isobutyrate (m/z 87) were normalized to butyrate-d7 (m/z 94); 2-methylbutyrate, valerate, and isovalerate (m/z 101) were normalized to valerate-d9 (m/z 110). Data analysis was performed using the Agilent MassHunter Quantitative Analysis software (Version 10.1, Agilent Technologies).

Statistical analyses for microbiome endpoints:

16S and shotgun sequencing were performed in R 4.1.1. The Inverse Simpson index was calculated for microbial α-diversity for the 16S and shotgun metagenomics data at genus level and species level, respectively. For both 16S and shotgun metagenomics data, Wilcoxon signed-rank test was utilized to conduct paired comparisons of the inverse Simpson index and the relative abundance of butyrate producers between study time points. Utilizing 16S data, Bray-Curtis distance was computed for microbial β-diversity, which is further visualized by the first two principal coordinates after running principal coordinate analysis (PCoA). The global test from the linear decomposition model (LDM) was used to compare the shifts in microbiota, accounting for paired samples at baseline and W12.(81) The differences in taxa relative abundance between baseline and W12 were compared by Wilcoxon signed-rank test for paired samples, where all genera were included for the 16S data and all species for the shotgun metagenomics data. Univariable Generalized Estimating Equations (GEE) models with exchangeable working correlation structure were applied to investigate the linear associations between the longitudinal BMI and nutrition characteristics (fiber intake per 1000kcal, diet adherence, and HEI scores) with longitudinal gut microbiome features (alpha diversity by inverse Simpson’s index and relative abundances of butyrate producers). For CAZymes, statistical significance was assessed by a Wilcoxon signed-rank test. Only those samples that had paired data for both time points were included in the statistical significance tests. Our analysis focused on those CAZymes/substrate pairs (out of 95 CAZyme/substrate pairs identified) found to be most different (unadjusted p<0.1) between the baseline and W4 timepoints. Significance for these CAZymes was reported for those <0.05. The relative abundance of butyrate pathway genes via shortBRED across samples was compared across sample timepoints using a paired Wilcoxon test. The Pearson correlation coefficient was generated to assess the concordance between the relative abundance of putative butyrate producers by 16S sequencing and shotgun sequencing and the relative abundance of genes associated with Acetyl CoA-butyrate production pathways. For SCFAs, Wilcoxon signed-rank test was used to determine the significance of the change from baseline, summarized by the median and IQR. P values were not adjusted for multiple testing (unadjusted p<0.05).

Immune Endpoints and Analysis

Excluded samples:

Participant 1008 had uncontrolled hyperglycemia and was not included in the immune analysis. Samples collected 1 week before or 3 weeks after COVID-19 infection, other infections, surgery, or admission were excluded from immune analysis.

Inflammation:

C-reactive protein and complete blood count with differential were measured using standard laboratory assays at baseline, W4, W12, W24, and W52.

Multiparametric flow cytometry:

Previously frozen peripheral blood mononuclear cells (PBMCs) were thawed, washed, and resuspended in phosphate-buffered saline (PBS), then incubated with Human TruStain FcX Fc receptor blocking solution (Biolegend) and Live/DEAD Fixable Blue Dead Cell Stain (Invitrogen) according to the manufacturers’ specifications for 20 minutes at room temperature (RT), protected from light. The cells were washed once in Flow Wash Buffer (FWB; RPMI 1640 no phenol red + 4% FBS +0.01% sodium-azide) and incubated with the antibody mix for 20 minutes at room temperature in the dark in the presence of Brilliant Staining Buffer (BD). The cells were washed, resuspended in 0.5% paraformaldehyde/PBS, and immediately acquired using a Cytek Aurora 5L flow cytometer (Cytek). The optimal concentration of all antibodies used in the study was defined by titration. Further information about the antibodies can be found on the manufacturer website (Table S20). Analysis was performed with FlowJo v10.8.1. For high-dimensional spectral flow cytometry analysis, each sample was downsampled to 3000 cells using the FlowJo DownSample plugin (version 3.3.1). The samples were concatenated, and Uniform Manifold Approximation and Projection (UMAP) (FlowJo plugin version 3.1; settings: 25 nearest neighbors, 1 minimum distance, 2 components, Euclidean distance metric) was performed using the mean fluorescence intensity (MFI) values from panel markers. Algorithm-assisted clustering was performed using FlowSOM (FlowJo Plugin version 3.0.18) with 30 metaclusters. The analysis yielded 30 leukocyte clusters that were characterized for the expression of lineage and sub-lineage specific markers (Table S21). Patient-specific cluster density was used to compare the relative abundances of leukocyte populations in each patient. The results obtained by unsupervised clustering analyses were confirmed by manual gating.

Plasma inflammatory biomarkers (Olink Analysis):

Changes in plasma secreted inflammatory biomarkers were measured with the Olink Target 96 Inflammation panels (Olink; Uppsala, Sweden) at baseline and W12 in peripheral blood and baseline and W52 in bone marrow supernatant, according to the manufacturer’s instructions. In brief, the method is based on proximity extension assay (PEA) technology, where a pair of oligonucleotide-labelled antibodies bind to the targeted protein, followed by a proximity-dependent DNA polymerization event. The resulting sequence is detected and quantified using standard real-time polymerase chain reaction (PCR). The assay simultaneously measures 92 proteins in each sample and provides normalized protein expression (NPX) data resulting from log2 transformation of the measured relative values. A high NPX protein value corresponds to a high protein concentration, but not an absolute quantification. Only samples where none of the internal controls contained an outlier value and with at least >75% measurements above detection limits in any PEA (n=1) were included in the further analyses. All assay characteristics, including detection limits and measurements of assay performance and validations, are available from the Olink manufacturer’s webpage.

Statistical analysis for immune endpoints:

To assess inflammation, the Wilcoxon signed-rank test was used to determine the significance of the change in C-reactive protein and complete blood counts from baseline, summarized by the median and IQR. For flow cytometry analyses, to account for possible batch effects and missing measurements at each timepoint, linear mixed effect models with random intercept were used to determine associations between flow measurements (i.e., outcomes) and time of measurements (i.e., predictor). These models were fit to each batch separately, and results were aggregated using fixed-effect meta-analysis. To assess secreted inflammatory biomarkers, the difference between W12 and baseline measurements by Olink inflammation panel were analyzed using unsupervised hierarchical clustering on × axis (patients) and y axis (concentrations) in plasma. P values were not adjusted for multiple testing (unadjusted p<0.05). In bone marrow supernatant the Wilcoxon signed-rank test for Olink inflammation panel was used (p<0.1).

Sequencing Endpoints and Analysis

Bone marrow single-cell RNA (scRNA-seq) and ATAC sequencing:

Viably frozen paired bone marrow mononuclear cells from baseline and W52 from 13 participants (1001, 1003, 1004, 1006, 1009, 1011, 1012, 1013, 1014, 1015, 1017, 1022 and 1023) were sorted by fluorescence-activated cell sorting (FACS) for live CD45⁺ single cells (CD45: BD Bioscience, catalog# 563204; DAPI: Millipore Sigma, catalog# 508741) and submitted for single-cell sequencing in 0.04% bovine serum albumin in PBS. They were processed using the Chromium Next GEM Single Cell Multiome Reagent Kit A (catalog no. 1000282) and ATAC Kit A (catalog no. 1000280). We followed the Chromium Next GEM Single Cell Multiome ATAC + Gene Expression Reagent Kits User Guide with the difference that single-cell suspension was prepared as described(82) (DOGMA-seq protocol with Digitonin buffer). Briefly, cells were treated with a digitonin lysis buffer (20 mM Tris-HCl pH 7.4, 150 mM NaCl, 3 mM MgCl2, 0.01% digitonin and 2 U/μl of RNase inhibitor) for 5 min on ice, followed by adding 1 mL of chilled wash buffer (20 mM Tris-HCl pH 7.4, 150 mM NaCl, 3 mM MgCl2, and 1 U/μl of RNAse inhibitor) and inversion before centrifugation at 500g for 5 min at 4 °C. The supernatant was discarded, and cells were resuspended in digitonin wash buffer, followed by counting using trypan blue and DAPI on a Countess II FL Automated Cell Counter. In addition, samples for each patient were pooled together on one lane of 10X Chromium (using Hash Tag Oligonucleotides - HTO).(82) Final libraries were sequenced on an Illumina NovaSeq S4 platform (R1 – 28 cycles, i7 – 8 cycles, R2 – 90 cycles). De-multiplexing, alignment, and matrices generation were performed with Cell Ranger ARC (10X Genomics), using default parameters and the GRCh38 reference genome. Downstream analyses were performed in R (version 4.3.2) with Seurat(83) (version 5.0.1) and CellChat (version 2.1.2) packages.(84)

Matrices with filtered unique molecular identifier (UMI) counts and peaks from cellranger ARC outputs were used as input for Seurat. Furthermore, these were filtered out according to the following criteria: (i) more than 5,000 or less than 500 expressed genes, (ii) more than 20,000 or less than 500 peak counts, (iii) a TSS (transcription start site) enrichment >1, (iv) a nucleosome signal <2, and (v) a percentage of mitochondrial genes ≥25%. Antibody-hashed cells were then demultiplexed taking the union of both Seurat’s HTODemux and the R package scDemultiplex.(85) For more stringent criteria, expression-based doublets were additionally removed using the package R scds(86). All samples were merged into the same object in the final step.

Cluster annotation was performed with Azimuth,(87) projecting the RNA-based data on the bone marrow reference (celltype.l2). Later, the main identified populations were manually curated and/or sub-annotated through Seurat’s function FindClusters and a predefined set of immune makers. Uniform Manifold Approximation and Projection (UMAP) reduction was used for visualization (RunUMAP(…, dims = 1:18, n.neighbors = 10, min.dist = 0.6, spread = 0.9)). Analyses focused on the monocyte compartment comprising granulocyte-monocyte progenitors (GMP), CD14 and CD16 monocytes, and myeloid derived suppressor cells (MDSC); and the cytotoxic subset composed of natural killer (NK) cells and exhausted and cytotoxic CD8 T cells. In addition to studying the effect of diet from baseline to W52, other sub-analyses included additional comparisons. In the first sub-analysis, patients were categorized as “weight loss” or “weight stable” in case they presented (or not) a reduction of at least 5% in the BMI from baseline to W52. In the second one, patients were classified as “16S richer” or “16S poorer” if the difference in the microbiome between baseline and W52 was incremented or not, respectively, in more than 30%.

Gene expression was aggregated by cell type to avoid results biasing for the differential cell numbers. Differentially expressed genes for each separated cell population were identified through DESeq2 Seurat’s implementation (always baseline vs W52, as default parameters). Differentially expressed genes and gene activity scores with p≤0.05 were then used for Gene Set Enrichment Analysis (GSEA) based on the Cancer Hallmark’s gene sets,(88) using the R package clusterProfiler(89) (with default parameters except minGSSize = 3, version 4.10.0).

Relying on CellChat package and using ICELLNET’s interactions database, cell-to-cell interactions and signaling pathways were inferred from transcriptomic data.(90) The global Seurat’s object was split by timepoint (baseline and W52) and individually passed to CellChat’s workflow (default parameters); the resulting objects were merged again by timepoint. TNF signaling pathway was selected in accordance with previous GSEA based on RNA-seq differential analysis.

Statistical analysis for sequencing endpoints:

Cell subtype proportions between baseline and W52 were compared using the Wilcoxon signed-rank test. For visualization purposes (Fig. 3e), cell proportions were pseudo-logarithmically transformed (log-10 and sigma 0.01) to show a smooth transition to a linear scale around 0 values. After pseudo-bulk data aggregation, differentially expressed genes were iteratively identified for each cell population based on a negative binomial distribution model (i.e., DESeq2 Seurat’s implementation, keeping default parameters). GSEA analysis was based on differentially expressed genes with p≤0.05, reducing the minimum group size (minGSSize) to a value of 3. For the graphic representation, hallmarks with p≤0.1 were highlighted.

Mouse Experiments

Vk*MYC, t- Vk*MYC and healthy wild type models:

Animal procedures were approved by the institutional animal ethics committee (Italian Ministry of Health authorization No. 208/2023-PR and n. 259/2025-PR). All mice were bred and maintained in conventional or specific pathogen-free animal facilities under 12hour light/12hour dark cycle, at temperatures of 23±2 degree Celsius and 50%±10 humidity. Mice were euthanized using carbon dioxide (CO₂) inhalation followed by confirmation of death in accordance with institutional animal care guidelines. All mice described in Fig. 4,5 and Fig. S5–S8 were on a C57BL/6 genetic background. In Vk*MYC (Tg(Igkv3–5*-MYC)11Plbe, MGI ID #7430707) transgenic mice the activation of the transcription factor MYC, whose locus is found rearranged in half of human MM tumors including SMM,(16, 17, 91) occurs sporadically through the exploitation of the physiological somatic hypermutation process in germinal center B cells. Within a year, although with variable intensity, all mice develop a monoclonal plasmacytosis confined to the BM, a measurable serum M-spike, and progressively show typical end-organ damage. Vk*MYC mice were screened by real-time PCR to identify experimental Vk*MYC^+/− mice. Monoclonal plasmacytosis was monitored for each mouse starting at 25 weeks of age by serum protein electrophoresis (see below) and included in the study at the phase of murine SMM (i.e., M-spike <3 g/dL, no evidence of anemia or end-organ damage) mimicking human SMM phase.(18) M-spike was quantified as γ-globulin/albumin ratio 0.2 ± 0.09 (i.e., M-spike <3 g/dL; Fig. S5a). All mice were born from 09/2022 to 11/2022 and included in the experiment from 05/2023 to 08/2023 (median age at inclusion control arm: 37±4 weeks; high-fiber arm: 37±3 weeks). All mice were randomly assigned to the control or treated arm and monitored for disease progression until the sacrifice of the last enrolled mouse (02/2024). Disease progression was monitored by serum M protein quantification in peripheral blood every other week. Sacrifice occurred either at M-spike ≥3 g/dL (i.e., γ-globulin/albumin ratio 0.5 ± 0.16; Fig. S5b), symptomatic disease (i.e. presence of anemia with hemoglobin <13g/dL; Fig. S5c), corresponding to SMM-to-MM evolution, or at the end of the observational period. Hemoglobin levels and complete blood count were obtained by Idexx Procyte analyzers (IDEXX, Italy)

t-VkMYC mice were generated by injecting i.v. Vk12598 cells (RRID: CVCL_D5FT) from MM-bearing Vk*MYC mice into non-irradiated WT recipient mice.(20) After MM cells injection, tumor burden can be observed in the BM but also in the spleen. t-Vk*MYC mice also develop anemia, renal impairment, clonal proteinuria, and bone destruction.(20) Disease progression was monitored weekly by serum M protein quantification in peripheral blood and mice were sacrificed at M-spike appearance. t-VkMYC mice received selected diet composition according to the experiment and starting 10 days prior tumor injection and monitored for weight variation. To generate diet-induced obese (DIO) mice, a high-fat diet was administered to t-VkMYC mice 34 days before tumor injection, followed by diet switch to either HFD or its isocaloric Ctrl. At the time of injection mice were randomly assigned to receive dedicated diets as detailed below until their sacrifice occurring at M-spike appearance.

Healthy wild-type mice (C57BL/6) were single-housed and fed dedicated diets for 50 days. Food intake for each mouse was quantified every 3 to 5 days. To measure caloric intake, wild-type healthy mice were single-housed.

NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ (NSG) mice(28) were fed dedicated diet for 10 days and injected i.v. with Vk12598 cells. Disease progression was monitored as described above.

Mouse diets:

Vk*MYC, t-VkMYC, wild-type healthy and NSG mice were randomized and assigned to dedicated diets:

1. HFD: Purified high-fiber diet (D15092907; formulated by Research Diets, Inc.) macronutrient content: 10%/kcal fats, 20%/kcal proteins, 70%/kcal carbohydrates. Soluble inulin fiber in the high-fiber diet account for 12% of the total kcal intake (328 g for 4057 kcals).

2. Ctrl: Purified matched control diet for HFD (D12450J; formulated by Research Diets, Inc.) macronutrient content: 10%/kcal fats, 20%/kcal proteins, 70%/kcal carbohydrates. Carbohydrates derived from corn starch.

3. SD: Standard diet refers to laboratory chow diet (VRF1 (P) Data Sheet 210114; SDS Diets; 3.85% fibers/weight of crude fibers).

4. DIO: Purified high-fat diet (D12492; formulated by Research Diets Inc.) for diet-induced obesity t-VkMYC model. Macronutrient content: 60%/kcal fats, 20%/kcal proteins, 20%/kcal carbohydrates.

The nutritional profiles of all used diets are listed in Tables S17 and S18. All diets were provided ad libitum.

SCFAs treatment:

Mice on a control diet (D12450J) received a mixture of SCFAs in their drinking water at a total concentration of 133mM (SCFA) or plain water (Ctrl). The solution was prepared starting from powders: sodium acetate (Sigma-Aldrich, Cat# S5636, CAS No 127–09-3), Sodium propionate (Sigma-Aldrich, Cat# P1880, CAS No 137–40-6), sodium butyrate (Sigma-Aldrich, Cat# 303410, CAS No 156–54-7). The SCFA ratio was determined based on quantified SCFA levels found in the stools of HFD Vk*MYC mice (93 mM acetate; 26.6 mM propionate; 13.3 mM butyrate, at a 70:20:10 molar ratio) (Fig. 5d). The solution was refreshed b.i.w.

M-protein quantification by serum protein electrophoresis:

Peripheral blood of Vk*MYC mice was periodically collected. Semi-automated electrophoresis was performed on the Hydrasys instrument (Sebia, Lissex, France) according to the manufacturer’s instructions. The use of Hydrasys densitometer and Phoresis software (Sebia, Lissex, France) for scanning resulting profiles provided accurate relative concentrations (percentage) of individual protein zones. M-spike concentrations were calculated as total γ-globulin/albumin ratio (γ/albumin).

Serum insulin quantification:

Mouse Insulin ELISA kit 1 (Mercodia, Cat# 0–1247-01) was used to measure serum insulin according to the manufacturer protocol. Samples were measured at paired time-points for each mouse.

Mouse fecal short-chain fatty acid quantification:

Feces were resuspended in buffer (PBS pH 7.4, 10% D2O, and 0.01% NaN3; ratio 1:10 mg feces:μL buffer). Samples were centrifuged for 5’ at 10000rpm and supernatants were collected. 50μM of 3-(Trimethylsilyl)-propane-1-sulfonic acid, as internal chemical shift reference, was added to samples and nuclear magnetic resonance (NMR) spectra acquired. NMR spectra were recorded at 298K on a Bruker Avance 600MHz spectrometer equipped with a triple resonance cryoprobe. 1D-1H-NMR spectra (noesypr1d) were recorded with an acquisition time of 2s, 128 transients and a relaxation delay of 6s. Spectral window was set to 12ppm. Spectra were processed with zero filling to 132k points, and apodized with an unshifted Gaussian and a 1 Hz line broadening exponential using Mnova 14.0 (Mestrelab Research).

Mouse microbiome 16S sequencing processing:

Total DNA was extracted from fecal samples of 24 mice using the QIAamp PowerFecal Pro DNA Kit (Qiagen, Cat# 51804),following manufacturer’s instructions. The V3-V4 region of the 16s rRNA gene was amplified starting from 500 ng of extracted DNA using the AccuPrime Taq DNA Polymerase (Invitrogen, Cat # 12339016),, the following primers: V3–16S-Fw: TCG TCG GCA GCG TCA GAT GTG TAT AAG AGA CAG CCT ACG GGN GGC WGC AG; V4–16S-Rev: GTC TCG TGG GCT CGG AGA TGT GTA TAA GAG AGA CAG GAC TAC HVG GGT ATC TAA TCC; and the amplification protocol: 94°C for 2 min, 35 cycles of 94°C for 30 sec, 56°C for 30 sec, 68°C for 1 min, and finally stored at 4°C. Amplicons were purified using the AMPure XP beads (Beckman Coulter, Brea, USA). A second PCR step was performed for indexing and add Illumina sequencing adapters to each sample. The Nextera XT Index Kit (Illumina, Cat# FC-131–1001) and the KAPA HiFi HotStart PCR Kit (KAPA Biosystem, Cat# KR0369) were used following the amplification protocol: 95°C for 3 min, 8 cycles of 95°C for 30 sec, 55°C for 30 sec, 72°C for 30 sec, 72°C for 4 min, and then stored at 4°C. A second purification step with AMPure XP beads was performed to clean up the samples for preparing the final library. Purified DNA is quantified using the Qubit double-stranded DNA (dsDNA) HS Assay Kit on a Qubit 2.0 (Thermo Fisher, RRID: SCR_020553) and then diluted and pooled following Illumina protocol. Sequencing is performed using the MiSeq Illumina platform (RRID: SCR_016379) with a 600 (2 X 300)-base pairs (bp) paired-end read protocol.

Mouse microbiome sequencing analyses:

Sequences with high-quality score and length >250bp were used for the taxonomic analysis with QIIME2 (Quantitative Insights Into Microbial Ecology 2 v. 2024.2) software (RRID:SCR_021258). For all samples a coverage rate >99% was obtained. Sequences were clustered into OTUs based on a 97% similarity threshold using UCLUST algorithm. Operational Taxonomic Units (OTUs) were picked by the de novo OTU picking method using the command pick_open_reference_otus.py. Representative sequences were aligned against Greengenes database (gg_13_8 release, RRID:SCR_002830). Taxonomy was assigned to identified OTUs by using the Ribosomal Database Project (RDP) classifier (RRID:SCR_006633). The number of observed OTUs and Shannon index were calculated to determine α-diversity. β-diversity was calculated and generated by using principal coordinates analyses (PCA), with the derivation of weighted UniFrac and unweighted UniFrac distance matrices. PERMANOVA and Kruskal-Wallis analyses were used for assessing the statistical significance of β and α-diversity respectively.

Isolation of mouse single-cell suspension and flow cytometry:

Mouse bones were harvested, and epiphyses were cut to collect BM cells. Peyer’s patches were removed from the small intestine of the same mice and gently disaggregated with the help of tweezers. The spleen was removed and single-cell suspension was obtained through 70 μm cell strainer. Single-cell suspensions were labeled with fluorochrome-conjugated monoclonal antibodies (Table S22). Cells were also assessed for intracellular cytokine production after 6 hours at 37°C of stimulation with Phorbol Myristate Acetate (PMA) (PMA; Sigma-Aldrich, Cat# 524400 CAS No 16561–29-8) and ionomycin (MedChemExpress, Cat# HY-13434, CAS No 56092–81-0). Brefeldin A (BFA; MedChemExpress, Cat# HY-16592, CAS No 20350–15-6) was added to the samples during the last 5 hours of culture. After incubation, cells were washed and stained for surface markers 15 minutes at 4°C, fixed and permeabilized with Fixation/Permeabilization Kit (BD-Bioscience; Cat# 554714) or with 2% paraformaldehyde (CAS No 30525–89-4) and saponin (CAS No 8047–15-2) solution. Cells were then washed and stained for intracellular markers 30 min at 4°C and acquired by CytoFLEX LX Flow Cytometer (Beckman Coulter, RRID: SCR_025067). Data were analyzed using the FlowJo software v10.9 (Treestar Inc; RRID:SCR_008520). Exemplification gating strategy for the analyzed organs is reported in Fig. S8a–b.

Statistical analysis for mouse experiments:

Mice were matched for age and sex. Randomization was performed for in vivo experiments (RRID:SCR_002798). Data were analyzed with GraphPad Prism version 9. Data are presented as mean ± standard deviation of the mean, individual values as scatter plot with column bar graphs. Paired or unpaired t-test was applied. Adjusted p value was performed by Holm-Šídák method. Differences were considered significant when p<0.05. N values represent biological replicates. Survival curves were compared using the log-rank test (Mantel–Cox). All the statistics and reproducibility are reported in the figure legends (Fig. 4–6; Fig. S5–S7). Bray-Curtis distance was computed for microbial β-diversity, which is further visualized by the first two principal coordinates after running principal coordinate analysis (PCoA). PERMANOVA test was used to compare the shifts in microbiota across the control and intervention mice groups (Fig. 5a and Fig. S6a).

In vitro experiments

Cell proliferation assay with SCFAs including butyrate:

GFP-expressing 5TGM1 cells ((RRID:CVCL_VI66) were stained with CellTraceTM Violet Cell Proliferation Kit (ThermoFisher Scientific; Cat# C34557) and seeded in the presence of SCFA-conditioned media (1mM or 0.1mM concentration). SCFAs mix was composed of acetate, propionate, and butyrate with the following molar ratio: 70:20:10. Cells were collected at 24, 48 and 72 hours and proliferation was quantified by dye dilution through flow cytometry. Proliferation in the presence of medium conditioned with butyrate (0.1mM), or propionate (0.2mM) or acetate (0.7mM) was also assessed. Values represent the mean of three biologically independent experiments.

In vitro induction of human monocyte-derived DCs:

CD14⁺ monocytes were isolated by magnetic sorting (CD14 MicroBeads, human, Miltenyi Biotec, Cat# 130–050-201) from healthy donor’s peripheral blood mononuclear cells (PBMCs) and cultured in RPMI (Lonza Bioscience) with 10% FCS, GM-CSF (50ng/ml; Peprotech, Inc; Cat 300–03) and IL-4 (25ng/ml; Peprotech, Inc; Cat# 200–04) supplemented with SCFAs at concentrations of 0.1mM; 0.5mM and 1mM. SCFA ratio was determined based on quantified SCFA levels found in the stool of Vk*MYC mice. DC activation was induced as described above.

The multiple myeloma cell line RPMI8226 (RRID: CVCL_0014) was acquired from ATCC and cultured as recommended (www.atcc.org/products/ccl-155). For the proliferation assay, 2×10⁵ cells per well in 48 wells plates (Corning) were cultured with DMEM media, supplemented with 10% FBS as well as sodium butyrate at the 5μM and 10μM concentrations. PBS was used as control. Proliferation was assessed in the live-cell analysis instrument, Incucyte S3 (www.sartorius.com; RRID:SCR_023147).. Cell proliferation per well was normalized to time 0 and the readout was performed every 24 hours. Values are representative of two independent experiments, each with duplicate wells. Statistical analysis was one-way ANOVA, followed by multiple comparison Dunnett tests (Prism).

statement of significance:

A high fiber plant-based diet in participants with precursor plasma cell disorders and an elevated body mass index improved metabolic, microbiome and immune biomarkers of disease. In a subset, it may delay progression to myeloma. In mouse models a high fiber diet delayed progression to myeloma independent of calorie restriction.

Data Availability Statement

The data generated in this study are available upon request from the corresponding author. Analysis code is available in GitHub – https://github.com/mskcc-microbiome/2025_shah_nutrivention_pilot.