Microbiome homology between vagina and tumors is linked with outcome after concurrent chemoradiotherapy
To further assess the relationship between chemoradiotherapy and the microbiome, patients were grouped as having either a good response (clinical complete response) or a poor response (partial response or stable disease) after CCRT. No significant differences were found in baseline clinicopathologic factors (age, TNM disease stage, HPV infection) between the good-response and poor-response groups, but the two groups differed in likelihood of disease recurrence at 4 years after treatment (Table 1). Although no significant difference in alpha diversity was found in vaginal or intratumoral microbiome between good and poor responders relative to CCRT (Fig. S1 A, C), beta diversity of the intratumoral microbiome, not the vaginal microbiome, was siginificantly difference between good and poor group (p < 0.05, Fig. S1 B, D). It indicated that intratumoral microbiome play more important role in the outcome of CCRT. To further investigate the possible relationship between intratumoral microbiome and vaginal microbiome in CCRT, dissimilarity of microbiome between vagina and cervical tumors were calculated for each patient, and Bray-Curtis analysis revealed that the composition of the microbiome was more similar among patients with a poor response compared with patients with a good response to CCRT (Fig. 2A), suggesting that the homology of cervical tumoral and vaginal microbiota could be related to outcome after CCRT.
We next used SourceTracker to predict the microbial translocation between the cervical tumor and vagina in the same subject. The results showed that the proportion of intratumoral microbes potentially originating from vagina were significantly different in good and poor response group (P < 0.01, Wilcoxon test), but the vaginal bacteria potentially originating from tumor was not significant in these two groups. It suggested that microbes transferred from the vagina to the cervical tumor were correlated with the outcome of CRT (Fig. 2B, C). To reconstruct the broader relationships between vagina and cervical tumor, we calculated the correlation coefficient for each homological ASV and found that most of the ASVs were positively correlated with poor response (71/100) (Fig. 2D). ASVs with |rgood-rpoor|> 0.7 were showed in Fig. 2E and their related proportion in cervical tumor and vagina were calculated. 75.8% ASVs in cervical tumor and 97.8% ASVs in vagina were vagina-tumor co-existing ASVs (Fig. 2F).All the results indicated that taxon shared in between the vaginal and cervical tumor niches could be associated with response to CCRT.
Specific bacteria located in both the vagina and cervical tumors were associated with clinical characteristics and immunomodulatory proteins
To further discover which specific bacteria that co-exist in vagina and cervical tumors may affect clinic outcome, we used Wilcoxon tests and found that the co-located bacteria (at ASVs level) was correlated with therapeutic effect (response to CCRT), HPV infection status, metastasis status, disease stage, and likelihood of recurrence within 4 years. First, g_Streptococcus_s_Streptococcus-anginosus_63 and Corynebacterium_simulans_167 in tumor, and g_Escherichia.Shigella_216, g_Enterococcus_55 in vagina were more abundant in poor response group than good response group. on the other hand, the abundance of three ASVs (g_Methylobacterium.Methylorubrum_310, g_Porphyromonas_s_Porphyromonas_somerae_32 and g_Peptostreptococcus_92) in tumor and three ASVs (g_Fenollaria_111, g_Prevotella_s_Prevotalla_timonensis_22 and g_Petoniphilus_s_Peptoniphilus_lacrimails_106) in vagina were higher in good response group (Fig. 3A). Several ASVs in the co-existing (shared) microbiome were associated with HPV infection; for example, g_Moryella_s_Moryella_indoligenes_141, g_porphyromonas_s_porphyromonas_somerae_33, g_Dialister_Veilonellaceae_bacterium_69 in tumor, g_Ureaplasma_s_Ureaplasma_parvum_39, g_Ureaplasma_38 in vagina were highly enriched in patients with high-risk HPV infection, and g_Fusobacterium_Fusobacterium_gonidiaformans_124 in tumor was increased in patients with low-risk HPV infection (Fig. 3B). In addition, the abundance of g_prevotella_s_prevotella_timonensis_24 were much higher in patients with high-risk HPV infection, both in tumor and vagina (Fig. 3B). g_Prevotella_s_Prevotella_bivia_26, g_Prevotella_s_Prevotella_timonensis_22 and g_Dialister_68 in tumor, g_Prevotella_s_Prevorella_disiens_17 and g_Anaerococcus_115 in vagina were decreased in patients with N0M0 disease, and the abundance of g_Gemella_s-Gemmella_asaccharolutica_45 in vagina was increased in patients with metastatic disease (Fig. 3C). Levels of two ASVs (g_Escherichia_Shigella_215 and g_Enterococcus_55) in tumor were increased and only one ASV (g_Precotella_s_Prevotella_timonensis_22) in vagina was decreased in patients who experienced recurrence vs those who did not (Fig. 3D). Finally, g_Moryella_s_Moryella_indoligenes_140, g_Prevotalla_s_Prevotella_bivia_27 and g_Anaerococcus_115 in tumor and g_Dialister_68 in vagina were more abundant in those with stage II disease compared with those with stage III/IV disease (Fig. 3E).
To further investigate the relationship between the microbiome and immune signatures, we collected peripheral blood samples from each patient before CCRT and assessed levels of immunomodulatory proteins by enzyme-linked immunosorbent assay (ELISA). We found that three co-stimulatory factors (ICOS, CD28, and LAG3) were expressed at high levels in the good-response group (Fig. 4A), and that several types of shared bacteria in good responders (Fig. 4B) and poor responders (Fig. 4C) were strongly correlated with these immune factors (Table S1). The results above indicated that specific bacteria in tumor/vagina were corelated with the clinic characteristics and systemic immunity in patients with cervical cancer.
Integrated prediction model combining microbial and immunological signatures could stratify patients by response to chemoradiotherapy
To further investigate the relationship between bacterial diversity, the immunomodulatory protein panel, and treatment response, we used Multi-omics factor analysis and found that by integrating vaginal taxa, cervical tumor taxa, and immunological signatures with outcomes after CCRT (good vs poor), we discovered that Factor1 contributed most to classify patients as having good or poor response (Fig. 5A, B). Taxa in cervical tumor and immune factors were the main component in Factor1(Fig. 5B). Moreover, the Factor 1 index was significantly higher in poor response group compared to good response group (Fig. 5C), implying that the microbiome (Fig. 5D) and immune factors (Fig. 5E) in Factor 1 were strongly associated with the outcome of CCRT. We next crossed the taxa located in the vagina, tumors or both with Factor1(Fig. 5F), and found that four taxa (g_Bosea_305, g_Brevundimona_289, g_Sphingobiume_s_Sphingobiume_xenophaum_333 and g_Ralstonia_256) that were co-located at both the vagina and cervical tumors were strongly corelated with the immune factors such as CD28, ICOS and LAG3 (Fig. 5G). Specifically, g_Sphingobium_s_Sphingobium_xenophagum_333 in tumor and g_Ralstonia_256 in vagina were identified as key for achieving the best predictive function, with AUC = 0.984 and AUPR = 0.917 (Fig. 5H). Hence bacteria that were co-located in the vagina and cervical tumor were critical for predicting good vs poor outcome after CCRT for cervical cancer.