Subsets of mononuclear phagocytes are enriched in the inflamed colons of patients with IBD

GSEA analysis comparing inflamed vs. non-inflamed tissue was performed to evaluate immune cell enrichment, including both innate and adaptive cell types, at different Crohn’s colonic locations in the colon. Overall, we observed immune cell enrichment at all locations of inflamed colon (Fig. 1a). Mononuclear phagocytes, especially monocyte, M1, aDC and pDC were among the most significantly enriched cell types in the ileum and descending colon. As a positive cellular control, distal colon was also enriched in Th1, Th2, CD4+ or CD8+ tissue effector memory cells amongst others. The enrichment of mononuclear phagocytes was confirmed by single sample enrichment analysis (ssGSEA) which showed that those myeloid cells were significantly enriched in inflamed colon from most locations (Fig. 1b, c). Among all the dendritic cell types, the enrichment of aDC, pDC and DC are significantly different between inflamed and non-inflamed colons in most locations. Among all the macrophage and monocyte signatures, the enrichment of monocytes and M1 are significantly different between inflamed and non-inflamed colons in most locations. Interestingly, M2 macrophages shows an opposing enrichment and are significantly enriched in non-inflamed ileum and sigmoid tissue.

In a separate UC data set with endoscopic scores, mononuclear phagocytes are enriched in advanced biopsies with mayo endoscopic scores of 2 or 3, comparing to normal control (Fig. 2a). ssGSEA confirmed that the enrichment scores of all DC subsets and monocyte in advanced patient samples are significantly higher than that of normal control (Fig. 2b, c). Macrophage and M1 macrophage enrichments show no significant difference between advanced samples and normal controls. However, their enrichment scores are significantly higher in advanced samples than those in samples with lower score (Fig. 2c). On the other side, M2 macrophages are significantly enriched in the control colon comparing to that in those severely diseased colon tissues. Similar to CD samples, we again observed that some T cell subset enrichments correlate with disease severity. One such subset is Th1, which is known to produce IFN-γ and play an important role in gut pathology. Overall, these data suggest that myeloid cell populations, including M1 macrophages and aDC, accumulate in inflamed regions and are associated with disease severity.

Current biological treatments for IBD, while effective, are not optimal. Although more than half of patients respond to treatment, many lose response or become resistant over time [29]. The mechanisms which lead to loss of response are not known. A better understanding of these mechanisms could enable novel therapeutic target identification. We determined immune cell type enrichments in samples from responding and non-responding patients including both pre- and post- treatment conditions. Out of the three independent infliximab treatment data sets, two strongly support mononuclear phagocytes (aDC, Monocytes, M1) as being among the top enriched cell types in the non-responding patients in both pre- and post- treatment samples (Fig. 3a, b), while the other dataset shows enrichment in the post treatment samples (Fig. 3c). Similarly, in vedolizumab-treated samples (Fig. 3d), there is a clear mononuclear phagocytes enrichment in non-responders before treatment, and they are further highly enriched after treatment. In order to investigate whether non-responders have more severe disease, we investigated the disease scores in pre-treatment samples for both infliximab and vedolizumab where data are available. There is no difference between responder vs. non-responder disease scores, which indicates that disease severity is not a factor that leads to treatment resistance.

To evaluate the prediction performance of each cell type for treatment non-response, we used ROC to measure the true-positive rates against the false positive rates at various thresholds of prediction scores and calculated the AUC score. Table 1 lists the top 10 immune cells sorted by AUC score. All 10 cell types predict non-responders better than random (i.e. all AUC scores > 0.5). Scores are higher in infliximab response prediction, especially for CD patients. Among the top ranked immune cell types, macrophage M1, monocytes and aDC are frequently found with higher AUC scores, which suggest higher prediction power.

We measured the enrichment of different myeloid subsets in IBD colon biopsies using the adopted ten gene signatures representing 6 DCs (DC1 to DC6) and 4 monocytes (Mono1 to Mono4) from Villani et al. [28], and assessed their enrichment in the datasets discussed above. For the DC subsets, conventional DC1, known for antigen cross presentation to CD8 T cells, was defined by CLEC9A instead of the classical surface marker CD141. DC2 and DC3 were both defined by the myeloid cell marker CD1C, but like DC1 shared a common conventional DC (cDC) progenitor. However, DC2 and DC3 differed from one another as DC3 can also originate from monocyte precursors (Mono1, CD14+ classical monocytes). DC4 was defined by absence of DC1/DC2/DC3 phenotyping markers CD141 and CD1C while retaining the generic myeloid marker CD11C. DC4 was not connected to the common cDC progenitor but instead to Mono2 (CD16+ classical monocytes). DC6 were the classical pDC, while the new DC5 showed features of both conventional and plasmacytoid DC. DC5 was marked by AXL + SIGLEC6+. Finally, they reported two new monocyte subsets: Mono3 and Mono4. Our results show that these DC and monocyte signatures are enriched in CD patient inflamed vs non-inflamed colon samples (Fig. 4a). Specifically, the Mono3 signature is highly enriched in the ileum, and Mono2, DC4 and DC6 signatures are highly enriched in the descending colon. For the UC samples with different disease scores, DC2, DC4, and Mono4 are highly enriched in the severe disease (Fig. 4b).

We further assessed the ten gene signatures in the patient samples with treatment response vs. resistance. In samples with infliximab treatment, most of those gene signatures are enriched in non-responders, and DC1 and DC5 are slightly less enriched (Fig. 5a, b, c). In samples with vedolizumab treatment, while most signatures are enriched in non-responders after treatment, DC4 is the most enriched (Fig. 5d).

To evaluate the connection of DC and monocyte subtypes and explore their function, we determined the correlation between subset expression in IBD data sets (Fig. 6). Most subsets, with the exception of DC5 and potentially DC6, show a high correlation with each other in both CD and UC patients (CD ρ > 0.6, UC ρ > 0.8). Most cell type expressions are strongly correlated (ρ > 0.5) with a panel of proinflammatory gene signature (IL1B, IL1A, IL23, IL12A, IL12B, TNFA, IL6, IL8, IL18, IFNA1, IFNB1), except for DC5 in CD.

Next, we assessed the prediction performance of each DC, monocyte subset for treatment non-response. ROC was used to measure the true-positive rates against the false positive rates at various thresholds of prediction scores (Additional file 1: Figure S1). All ROC AUC scores are above 0.5 (random). Similarly, we found that the scores are higher in infliximab treatment prediction, especially for CD patients (Additional file 1: Figure S1 d).

Chemokines are chemotactic cytokines that control immune cell migratory patterns and positioning, and are critical for their development and homeostasis [30]. We evaluated chemokine expression and identified a set whose expression was up-regulated in IBD colon biopsies, and the up-regulation pattern was retained only in non-responders after treatment (Fig. 7 and Additional file 1: Figure S2 a). Among those cytokines, CXCL1/2/3/5/6, CCL2/3/4/24 are reported to be involved in myeloid cell trafficking [30]. It provides additional evidence for the enrichment of myeloid subsets in inflamed and treatment resistant samples that we observe. Other chemokines (CXCL9/10/11/13, CCL18/22) with a similar pattern are involved in Th1/Th2 response and migration. This is consistent with T cell involvement in disease pathogenesis. To further quantify the chemokines involved in myeloid cell trafficking [30], we examined their median expression in related data sets. Shown in Additional file 1: Figure S2 b, c, there is a distinct up-regulation pattern of those chemokines in non-responder vs. responder population, in both pre- and post- treatment groups for either infliximab or vedolizumab.

In order to delineate the cell types in the colon tissue microenvironment contributing to dys-regulation of those myeloid chemokines, we leveraged a recent scRNA-seq data from human stromal cells [31]. CXCL1 and CXCL2 was up-regulated while CCL2 was not. Expression of other chemokines are either rare (CXCL3/5/6, CCL3/4) or not detectable (CCL24) (Additional file 1: Figure S3). These observations suggest that colonic stromal cells may contribute to the expression of CXCL1/2, while other cells—such as epithelial cells—are likely contributing to the chemokine dys-regulation observed in mixed colon tissues.

In vivo murine model of colitis, amongst other features, provide a platform to enumerate mononuclear phagocytes in situ during inflammation development. In addition, advancements in transcriptomics and other technologies allow the selective ablation of particular mononuclear phagocytes and understand their roles in pathology using these murine models of disease. However, surface proteins are not necessarily shared between murine and human cells. Hence, we focused on standard mononuclear phagocyte phenotype in the mouse colon relying on their broad functional classification. Myeloid cell isolation procedures and subsequent phenotype analysis by Flow Cytometry needed to divide them into two fractions: intraepithelial lymphocyte fraction (IEL) and lamina propria fraction. Both fractions taken together, allowed us to consider myeloid subset changes in the entire colon.

The DSS induced colitis model depicts chronic pathology to a greater extent than the often used acute DSS model (Fig. 8). The overall pathology initiation is similar between the two models, as both display epithelial barrier injury exposing immune cells to luminal pathogens. However, the cellular activity is reported to be different. Using the chronic DSS model, we found inflammation expanded total and potentially activated MHCII+CD11c hi/lo macrophage subsets, in both IEL and LP fractions. Total cDC and all DC subsets, including DC1 and DC2, were more enhanced in the IEL fraction than in the LP fraction. Interestingly, we also found that cDCs, with features of both DC1 and DC2, were enhanced in both fractions. DC2 is the myeloid DC subset in the mouse, expressing CD11b, and well known for T cell activation. DC1 s expressing CD103 have been shown to generate and sustain Tregs and cross present antigen to CD8 T cells. Recently, murine DC1 were also shown to interact with intestinal epithelial cells, thereby helping them maintain barrier integrity in an acute DSS colitis murine model [32]. Lastly, we observed inflammation-induced expansion of pDCs which have been reported to be pathogenic in an acute DSS colitis model [2], but not in more spontaneous IBD models [33] and induced into tolerogenic cells by commensal bacterial antigen thereby conferring protection [34].

Three transcriptomic studies (GSE100833, GSE73661, GSE16879) for samples from both CD and UC patients, representing different locations of the gut, disease severity and treatment response (infliximab, vedolizumab), were downloaded from GEO database and used in the analysis. For each data set, we standardized the transcriptome data across patients by quantile-normalization.

scRNA-seq data for stroma cells from UC patients were downloaded (GSE114374). Cells were QC, selected and normalized by using Seurat package (v2.3) [52].

From GSE16879, two separate data sets represent infliximab treatment study in CD and UC. In CD patients, there are 12 responders vs. 7 non-responders. In UC patients, there are 8 responders vs. 16 non-responders. From GSE73661, one study contains infliximab treatment in UC patients including 8 responders vs. 15 non-responders; the other study contains vedolizumab treatment in UC patients including data from 16 responders vs. 25 non-responders before treatment at week 0; and 13 responders vs. 6 non-responders after treatment at week 52. Detailed information regarding patient and biopsy can be found in [53, 54].

Immune cell type gene signatures were from xCell study which were compiled from over thousands of sorted human cell type transcriptomes from various resources [25]. There are 165 gene signatures with number of gene > = 15, representing a total of 31 immune cell types which were included in the analysis (iDC and NKT gene signatures were removed due to small number of genes).

DC, monocyte subset gene signatures are from scRNA-seq study of human blood DC, monocytes [28].

Two sample population GSEA analysis [26] was performed for each gene signature. Normalized enrichment score and p values were averaged for the same cell type.

Single sample enrichment analysis were based on the analysis of xCell [25].

Spearman correlation analysis was performed for each pair of DC, monocyte subset (after removing common genes) for their expressions in either UC or CD transcriptomic data using R v3.5.1. The heatmaps were generated using gplots v3.0.1.1 (https://​cran.​r-project.​org/​web/​packages/​gplots/​index.​html) with default parameters. Similarly, their potential function in inflammation was also assessed by evaluating their correlation with a proinflammatory gene signature. The expression value of each signature was represented by the median value of all member in the specific data set.

Response prediction performance for each classifier (cell type) was implemented in ROCR package. The predicted value of each gene signature was the median value of all members in the data set. For cell type with multiple gene signatures, the final AUC score is the average of individual ones.

Female Bl/6 J mice of around 7 weeks of age were purchased from Jackson Laboratories and housed in the animal facilities of Takeda California and maintained under regular specific pathogen free condition. Following 1 week of acclimatization, for some mice, water was replaced with 3% DSS for 3 cycles, each cycle consisting of 5 days. We performed in vivo experiment with 3 untreated and 6 DSS treated mice. All the 9 animals were the same gender, same date of birth and arrived at the animal facility at the same time and house for similar time period. Animals were randomly picked and put into 3 cages of 3 mice per cage. Mean wt. at the start was 18.8 +/− 1.1 g. In between two consecutive cycles, these DSS-fed mice were put on regular water for 15–20 days. All three DSS cycle were initiated before noon. Body weights were monitored on a regular basis. No special husbandry condition was employed. All mice were sacrificed on day 55 after the start of the first round of DSS by using CO2 based euthanasia as per our animal facility regulations. DSS treatment was performed in parallel i.e. two bottles were placed in two cages. While sacrificing for cell isolation, experimenters were blinded of the order of sacrifice. No adverse events were observed on course of the experiment. The colons of mice were removed from MLNs and other mesenteric structures, cut open longitudinally and fecal matters removed. Then the opened colon was transversely cut into small pieces of approximately 1 cm in length and intraepithelial and lamina propria fractions were extracted following protocol supplied by the manufacturer (LP dissociation kit, Miltenyi Biotec) with a few modifications. Briefly, small pieces of colon were incubated in pre-digestion solution (HBSS buffer containing 10 mM Hepes buffer, 5 mM EDTA, 5% FBS and 1 mM DTT) and continuously but gently shaken using a MACSmix Tube Rotator at 37 °C. This step was repeated for 2 rounds. After each round, tubes the tissue in pre-digestion solution were vigorously shaken for approximately 10 s and the content was passed through 70 μM filter to get the IEL fraction. Following two rounds of pre-digestion solution, the remaining tissue was washed in HBSS buffer containing only 10 mM Hepes buffer (passing through 70 μM filter to add to the IEL fraction) to remove the EDTA. Thereafter, the colon pieces were incubated in digestion solution (HBSS with Ca2++ and Mg2++ buffer containing 10 mM Hepes buffer, 5% FBS and requisite amount of enzymes provided in the kit) and gently shaken using a Max rotor at 37 °C. Finally, tissue pieces were disintegrated using appropriate program of the gentle MACS dissociator and passed through 70 μM filter to get the LP fraction. The whole experiment was repeated as a part of a different experiment which was not part of the current project.

All experiments were ethically performed according to our institutional guidelines. We are an AAALAC accredited organization using The Guide for the Care and Use of Laboratory Animals (Guide) as the primary standard for our animal care and use program. We are also a registered research facility with the California Department of Health Services and hold a Certificate of Approval to Keep and Use Laboratory Animals.

Single cell suspensions from IEL and LP were labelled with a cocktail of flurochrome conjugated antibodies against surface antigens purchased from either BD Biosciences (San Jose, CA) or Biolegend (San Diego, CA): anti-CD45 FITC, anti-CD19 PerCP Cy5.5, anti-Siglec H Pacific Blue, anti-I/A I/E BV605, anti-CD11b BV 650, anti-CD3e BV786, anti-B220 BUV496, anti-EpCam APC, anti-F4/80 AF700, anti-Ly6c APC-Cy7, anti-CD11c PE-CF594, anti-CD103 PE-Cy7 and Zombie aqua kit. Cells were fixed with 1% PFA prior to acquisition in Flow Cytometry. Cells were acquired using a 18 color BD LSRFortessa (San Jose, CA) and analyzed with FlowJO software (Ashland, Oregon).