The research was conducted in accordance with the PRoBE design concept (prospective specimen collection and retrospective blinded evaluation) (26). This study was approved by the Ethics Committee of Xijing Hospital of the Fourth military medical university (KY20192070), and the participants provided their written informed consent to participate in this study.

Patients who were newly diagnosed with AL amyloidosis between July 2018 and February 2021 in Xijing hospital (Xi’an, China) were screened. The inclusion criteria included patients: I) were biopsy-proven primary systemic AL amyloidosis; II) ≥18 years of age. The exclusion criteria were patients: I) had digestive diseases or systemic diseases such as diabetes and hypertension; II) have been treated with chemotherapy; III) used antibiotics or probiotics within 3 months before sampling; IV) in the period of pregnancy.

Healthy volunteers who were age-, gender-, and body mass index (BMI)-matched with AL amyloidosis patients from the physical examination center of Xijing hospital were enrolled. The inclusion criteria were individuals: I) had a normal value of kidney and liver function, routine blood, feces, and urine tests, fasting blood glucose, blood lipids, and blood pressure; and II) ≥18 years of age. The exclusion criteria were individuals: I) administered antibiotics or probiotics three months before enrollment; II) had a history of any chronic diseases or acute infection.

The AL amyloidosis patient registration form was designed to record the demographic and clinical characteristics at kidney biopsy including gender, age, height, weight, blood pressure, medical history, pathological results, biochemical indexes, and therapies. All indicators above were uniformly tested and issued by the Laboratory Department of Xijing Hospital and extracted from the patient medical record system. Clinical organ involvement was assessed according to international consensus criteria (27), and the Mayo 2012 staging system was used for stratification (28).

The final follow-up date was March 31, 2022. Overall survival (OS) was defined as the time from diagnosis to death or the last follow-up. Patients who were alive at the last follow-up were censored at that date. Progression-free survival (PFS) was defined as the period from treatment initiation to disease progression, relapse, or death from any cause.

All patients received subcutaneous bortezomib at a dose of 1.3 mg per square meter of body-surface area, and dexamethasone at a dose of 40 mg orally or intravenously once weekly for 28 days each cycle. The assessment of hematological and organ response was performed according to the validated response criteria published by the International Society of Amyloidosis (29, 30). A ≥25% eGFR decrease was considered as the criterion for renal progression. All response assessment was performed on an intent-to-treat (ITT) basis.

Each individual provided a fresh tail stool sample at 06:30–08:30 hours. Fecal samples were quickly placed in sterile specimen tubes, and transferred to a -80°C cryogenic refrigerator for further analysis. DNA extraction was carried out as previously described by our team (31).

DNA libraries were constructed, and the sequencing was performed on an Illumina MiSeq platform by Shanghai Mobio Biomedical Technology Co. Ltd., China. The 16S ribosomal RNA (rRNA) gene sequence that targeted the V3-V4 region was applied to identify the bacterial taxon, which was detailed described in a previously published article by our team (31). Original Illumina read data for all samples were stored in the NCBI Sequence Read Archive (SRA) database under accession number PRJNA825339 and PRJNA574226.

Amplicon sequence variants (ASVs) were identified with the DADA2 algorithm. The representative sequences for each ASV were annotated using the SILVA reference database (SSU138). QIIME feature classifier was used for species annotation, which was the species annotation plug-in of the QIIME 2 analysis process, and adopted the classify-sklearn algorithm. Alpha diversity metrics (ACE estimator, Chao 1 estimator, Shannon-Wiener diversity index and Simpson diversity index) were assessed by using Mothur v1.42.1. The non-parametric Mann-Whitney U test was used to test for significant differences between two groups. A comparison of multiple groups was done using a nonparametric Kruskal-Wallis test. Bray-Curtis, weighted UniFrac, unweighted UniFrac, and Jaccard-binary dissimilarity were calculated in QIIME. Principal coordinate analysis (PCoA) plots and PERMANOVA which were used to test for statistical significance between the groups using 10,000 permutations were generated in R (version 3.6.0) package vegan 2.5-7. The linear discriminant analysis (LDA) effect size (LEfSe) (32) was used to detect taxa with differential abundance among groups (lefse 1.1, https://github.com/SegataLab/lefse). A heatmap plot of the key ASVs identified by random forest models was generated by using the ‘pheatmap’ package of the R program. Probability of disease (POD) index was defined as the ratio between the number of randomly generated decision trees that predicted sample as “AL amyloidosis” and that of healthy controls. The identified optimal set of ASVs was finally used for the calculation of POD index for both the training and the testing cohort. A receiver operating characteristics (ROC) analysis was performed to measure the quality of the classification models by the R software package pROC. PICRUSt2 v2.4.1(https://github.com/picrust/picrust2/wiki) (33) was used to predict functional abundances in the Kyoto Encyclopedia of Genes and Genomes (KEGG) PATHWAY database based on 16S rRNA gene sequences. Spearman’s rank correlation was performed to explore correlations between ASVs and the clinical characteristics of patients with AL amyloidosis.

We recruited 27 patients with AL amyloidosis and 27 healthy controls (HCs). Process and flow chart of this study were demonstrated in Figure 1 . The baseline clinical characteristics of these two groups were shown in Table 1 . Age, gender, and BMI were matched between these two groups. There was also no significant difference between two groups at serum creatinine level, while albumin was considerably lower in patients with AL amyloidosis than that in healthy controls. The median baseline difference between the involved and uninvolved free light chain (dFLC) levels was 142 mg/L (range, 15 to 1201), and the median proteinuria was 2780 mg/24 h (range, 102 to 7905). A total of 18 patients (66.7%) had two or more organs involved; 81.48% of patients had kidney involvement, and 62.96% had heart involvement. The majority of patients (96.29%) were classified as lambda isotype and approximately half of patients (48.15%) had a Mayo 2012 stage of III or higher.

The rarefaction curve revealed the number of ASVs of each sample at different sequencing quantities, which indicated that the amount of sequencing data is large enough to reflect the vast majority of microbial information in the samples when the curve approaches plateau ( Figure 2A ). In total, 3,629,337 usable raw reads were obtained from 54 stool samples. After quality filtering and assembly of overlapping paired-end reads, 1,892,625 high-quality reads were generated and 976 ASVs were obtained. The average number of sequences per sample was 67,209 ± 20,151 (range 35,925–120,795, Supplementary Table 1 ). A total of 729 ASVs were shared among the two groups according to the Venn diagram, while 168 ASVs were unique for AL amyloidosis, and 79 ASVs were specific for HCs ( Figure 2B ). No significant differences in community richness (estimated by Chao and ACE indices) and diversity (measured by Shannon and Simpson indices) were observed between AL amyloidosis and healthy controls ( Supplementary Table 2 ). The gut microbial communities in patients with AL amyloidosis and the HCs were clustered separately, indicating a marked differentiation in taxonomic composition based on Jaccard-binary distances (Adonis, p = 0.0155, Figure 2C ; Supplementary Table 3 ). In the ANOSIM analysis based on Jaccard-binary distances ( Figure 2D ; Supplementary Table 3 ), the difference between the two groups was significantly greater than that within the groups, further confirming the value of comparison between these two groups (R = 0.055, p = 0.0162).

Based on the taxonomic classification of the ASVs, the relative microbial abundance of 54 samples at different levels (phylum, class, order, family, and genus) was identified. The phylum and genus levels of fecal microbial composition in each sample from two groups were shown in Supplementary Figures 1A, B .

At the phylum level, Firmicutes, Bacteroidota, and Proteobacteria were the three dominant populations in two groups, accounting for up to 90% of sequences on average ( Figure 3A ). Compared with HCs, Actinobacteriota and Verrucomicrobiota were enriched (3.22% vs. 1.15%, 2.20% vs. 0.52%, respectively) in patients with AL amyloidosis, while Bacteroidota was considerably reduced in patients with AL amyloidosis (31.20% vs. 42.85%, all p < 0.05, Figure 3B ; Supplementary Table 4 ).

At the genus level, Bacteroides, Faecalibacterium, and Escherichia-Shigella in AL amyloidosis, while Bacteroides, Faecalibacterium, and Prevotella in the HCs, were three dominant populations in two groups, each accounting for up to 5% of the sequences on average ( Figure 3C ). By comparison, we observed that a total of 21 genera had significantly different abundance between the two groups, and the relative abundance of the top 8 genera was presented in Figure 3D . Specifically, six genera, namely, Bifidobacterium, [Eubacterium]_coprostanoligenes_group, Akkermansia, Enterobacteriaceae_unclassified, Streptococcus, and Collinsella were enriched in the AL amyloidosis group (all p < 0.05), whereas Faecalibacterium and Tyzzerella were remarkably decreased in AL amyloidosis group (all p < 0.05, Figure 3D ; Supplementary Table 5 ).

In addition to the differences at the above different levels, we analyzed gut microbiota using the LEfSe approach to identify the specific taxa associated with AL amyloidosis. The histogram of LDA value distribution ( Figure 4A ) revealed that 35 microbial biomarkers clearly distinguished patients with AL amyloidosis and HCs (LDA > 2.5, all p < 0.05, Supplementary Table 6 ). Meanwhile, a cladogram ( Figure 4B ) depicting the fecal microbial structure and prevalent bacteria revealed the significant alterations in taxa between AL amyloidosis patients and HCs. Compared with the healthy group, patients with AL amyloidosis had a markedly higher abundance of several bacterial taxon chains within the phylum Verrucomicrobiota and Actinobacteriota. Specifically, p-Verrucomicrobiota.c-Verrucomicrobiae.o-Verrucomicrobiales.f-Akkermansiaceae.g-Akkermansia, p-Actinobacteriota.c-Actinobacteria.o-Bifidobacteriales.f-Bifidobacteriaceae.g-Bifidobacterium, p-Actinobacteriota.c-Coriobacteriia.o-Coriobacteriales.f-Coriobacteriaceae.g-Collinsella, as well as genus Eggerthella of family Eggerthellaceae in phylum Bacteroidota, were enriched in AL amyloidosis. Whereas in phyla Bacteroidota and Proteobacteria, two taxon clades, p-Bacteroidota.c-Bacteroidia.o-Bacteroidales and o-Pseudomonadales.f-Pseudomonadaceae.g-Pseudomonas, were substantially less abundant in AL amyloidosis patients (LDA > 2.5, all p < 0.05).

To elucidate the functional and metabolic alterations of gut microbiota between AL amyloidosis and HC groups, PICRUSt2 analysis was used to predict functional abundances based on 16S rRNA gene sequences. We observed a substantial increase in functional abundance of 7 KEGG pathways in patients with AL amyloidosis compared with HCs, including one pathway at level 2, metabolism of other amino acids, and 6 pathways at level 3, namely, retinol metabolism, glutathione metabolism, naphthalene degradation, taurine and hypotaurine metabolism, zeatin biosynthesis, and apoptosis. Whereas, the genes of four pathways including metabolism of terpenoids and polyketides at level 2, and biosynthesis of ansamycins, synthesis and degradation of ketone bodies, glycerolipid metabolism at level 3 were significantly decreased in AL amyloidosis patients (all p < 0.05, LDA > 2.5, Figures 5A, B , and Supplementary Table 7 ).

To explore correlations between ASVs and the clinical characteristics of patients with AL amyloidosis, Spearman’s rank correlation was performed. The oblique triangle heatmap indicated a taxon-taxon correlation between gut microbiota and clinical parameters. A total of 17 solid lines represented strong correlations (p < 0.01) between 9 ASVs and 7 clinical parameters and may be the focus of research ( Supplementary Table 8 ; Figure 6A ). For example, ASV605 (Streptococcus) was negatively correlated with systolic blood pressure (SBP) and positively correlated with ASV244 (Christensenellaceae_R-7_group), ASV496 (Marvinbryantia) and ASV638 (Bifidobacterium). In the genus Christensenellaceae_R-7_group, ASV518 and ASV895 were negatively correlated with albumin (ALB) and total protein (TP), and ASV518 was also positively correlated with age.

As for AL amyloidosis-unique clinical parameters, a significant positive correlation existed between mayo 2012 stage and ASV649 (Alistipes) (ρ=-0.38, p = 0.049), and ASV167 (Bacteroides) (ρ=-0.61, p < 0.001). Moreover, lambda restricted had a negative correlation with ASV530 (Oscillospiraceae_UCG-005) (ρ=-0.39, p = 0.04). Significant positive correlations also existed between dFLC and ASV922 (Veillonella) (ρ=0.42, p = 0.027), and between ALP and ASV581 (Lachnospiraceae_unclassified) (ρ=0.40, p = 0.037, Supplementary Table 9 ; Figure 6B ).

According to the above significant difference in gut microbiota between AL amyloidosis and HCs, we assessed the potential use of gut microbiota-based signatures in the diagnosis of AL amyloidosis. A total of 63 ASVs were used for random forest modeling and biomarker selection, which were significantly different between the two groups (Mann-Whitney U test, p < 0.05) and their corresponding abundance was more than 0.1% in any sample. A five-fold cross-validation curve of the random forest model revealed that the 5 ASV-based markers were identified as the optimal marker set ( Figure 7A ). The relative abundance of the 5 ASV-based markers in each sample was presented in Supplementary Table 10 . In the stochastic decision forest model, the distribution of ASV importance was demonstrated by the mean decrease in accuracy and mean decrease in the Gini coefficient ( Figure 7B ).

In the training phase, the probability of disease (POD) value was significantly increased in the AL amyloidosis samples compared with the control samples (p = 1.3×10^–4, Figure 7C ; Supplementary Table 11 ). The POD index achieved an AUC value of 0.8549 with 95% CI of 0.731 to 0.9789 between AL amyloidosis and HCs cohorts (p = 0.0001, Figure 7D ).

In the validation phase, 9 HCs and 9 AL amyloidosis were used to validate the diagnostic efficacy of the POD for AL amyloidosis. Each POD of each patient was estimated and the relevant results are shown in the online Supplementary Table 12 . The average POD value was significantly higher in the 9 patients with AL amyloidosis than that in 9 controls (p = 0.034, Figure 7E ), and the POD attained an AUC value of 0.8025 (95% CI 0.5771-1, p = 0.0339) between AL amyloidosis and controls ( Figure 7F ).

Subgroup analysis was further done to reveal the association between the characteristics of gut microbiota at baseline and disease severity, treatment response, and prognosis in 27 patients with AL amyloidosis. All the following subgroups were matched between age, gender, and BMI ( Supplementary Table 13 ).