Transplantation of fecal microbiota from patients with inflammatory bowel disease and depression alters immune response and behavior in recipient mice

Gut dysbiosis is closely associated with the occurrence of inflammatory bowel disease (IBD) and psychiatric disorder. Here, to understand the difference of gut microbiota composition and physiological effect between IBD patients with (IBD/D+) or without depression (IBD/D−), we analyzed the fecal microbiota composition of patients with IBD with (/D+) or without depression (/D−) and healthy volunteers (HVs) and examined the effects of these fecal microbiota transplantations (FMTs) on the occurrence of systemic inflammation and anxiety/depression in mice. FMTs from patients with IBD/D+ or IBD/D− caused IBD-like colitis in the transplanted mice: they increased the myeloperoxidase activity, IL-1β and IL-6 expression, and NF-κB+/CD11c+ cell population in the colon. Transplantation of the IBD/D+ patient feces (IBD/D+-F) caused IBD-like colitis more strongly than that of IBD/D−-F. FMTs from patients with IBD/D+ also caused anxiety-/depression-like behaviors, increased the NF-κB+/Iba1+ and lipopolysaccharide (LPS)+/Iba1+ cell populations, and decreased the BDNF+/NeuN+ cell population in the hippocampus. They increased LPS levels in the blood. FMTs from patients with IBD/D− caused anxiety-like, but not depression-like, behaviors. α-/β-diversities and composition of gut microbiota in IBD-F were different from those of HV feces (HV-F). The Enterobacteriaceae and Enterococcaceae populations and LPS levels were higher in the IBD-F than in the HV-F. The Enterococcaceae population was higher in IBD/D+-F vs. IBD/D−-F. However, the transplantation of HV-F into mice previously transplanted with IBD/D+-F significantly reduced depression-like behaviors, NF-κB+/Iba1+ and LPS+/Iba1+ cell populations in the hippocampus, LPS levels in the feces and blood, and IL-1β expression in the colon. These findings suggest that the outbreak of depression/anxiety may be dependent on the systemic inflammation with a leaky gut through the gut dysbiosis-attributable overproduction of bacterial LPS and suppression of tight junction protein expression in patients with IBD.


Results
Fecal microbiota transplantations from patients with IBD caused colitis as well as anxiety/ depression in the transplanted mice. To understand the role of gut microbiota in the occurrence of IBD and anxiety/depression, we transplanted the fecal microbiota of HVs or patients with IBD/D + (score ≥ 11 on HADS-D) or IBD/D − (score ≤ 6 on HADS-D) into specific pathogen-free mice and examined their effects regarding the occurrence of colitis and anxiety/depression. The FMTs from patients with IBD/D + or IBD/D − all caused colitis in mice: they significantly induced colon shortening; upregulated myeloperoxidase activity, IL-1β and IL-6 expression and increased the NF-κB + /CD11c + cell population in the colon ( Fig. 1a-g). In contrast, the FMT from HVs did not cause colitis. In particular, the stenosis was a significant occurrence in mice with FMTs from IBD/D + patients vs. mice with FMTs from IBD/D − patients. After segregation of the IBD population into patients with patients with UC or CD, the FMTs of patients with UC or CD all caused colitis in mice, regardless of the presence/absence of depression: they increased colon shortening; upregulated myeloperoxidase activity, IL-1β and IL-6 expression; increased the NF-κB + /CD11c + cell population in the colon (Supplementary Figure S1). Although the occurrence of IBD-like colitis between FMTs from UC and CD patients was not significantly different, the stenosis occurrence was higher in mice with FMTs from patients with CD than in one with UC, but was not different between ones transplanted with the feces of HVs and UC patients.
The FMTs from patients with IBD/D + or IBD/D − all also significantly increased anxiety-like behaviors in the transplanted mice in the EPM, LDT, and MB tasks ( Fig. 2a-e). Moreover, depression-like behaviors in the TST and FST were significantly increased by the FMTs from patients with IBD/D + , but not those from patients with IBD/D − . The FMTs from patients with IBD also increased NF-κB + /Iba1 + , and LPS + /Iba1 + , and IL-1R + cell populations, as well as IL-1β expression in the hippocampus, while the BDNF + /NeuN + cell population reduced (Fig. 2f,g). They also increased corticosterone, IL-1β, IL-6, and LPS levels in the blood (Fig. 2h-k). After the segregation of the IBD population into patients with UC and CD, the FMTs from patients with UC and CD caused depression-like behaviors in mice (Supplementary Figure S2). The neuroinflammatory markers IL-1β and IL-6 and the NF-κB + /Iba1 + cell population were induced in the hippocampus of mice with FMT from patients with UC and CD. However, the FMT from HVs did not significantly affect neuroinflammation markers.
Fecal microbiota composition in patients with IBD/D + and IBD/D − . The FMTs from patients with IBD significantly caused colitis and anxiety/depression in transplanted mice compared with those from HVs. Therefore, to understand the role of gut microbiota in the occurrence of depression and colitis, we investigated the fecal microbiota composition of the HV-F, IBD/D + -F, and IBD/D − -F using pyrosequencing (Fig. 3, Supplementary Figures S3 and S4). The α-diversity estimated operational taxonomic unit (OTU) richness, but not Shannon's diversity index, was lower in the IBD-F than in the HV-F (Fig. 3A,B). Moreover, the OTU richness was lower in the IBD/D + -F than in the IBD/D − -F. The β-diversity and bacterial community were significantly different between IBD/D + -F and IBD/D − -F. At the phylum level, the Proteobacteria population was higher in the IBD-F compared with HV-F, while the Bacteroidetes population was lower in the IBD-F. At the family level, the populations of Enterobacteriaceae and Enterococcaceae were higher in the IBD-F compared with HV-F (Fig. 3C). The Enterococcaceae and Lactobacillaceae populations were higher in the IBD/D + -F than in the IBD/D − -F. In particular, the fecal LPS level was significantly higher in the IBD-F than in the HV-F (Supplementary Figure S5). It was weakly, but not significantly, higher in the IBD/D + -F than in the IBD/D − -F. www.nature.com/scientificreports/ Next, to understand what kinds of gut microbiota are associated with the occurrence of anxiety or depression, we analyzed the correlation between the gut microbiota composition and depression index in patients with IBD (hospital anxiety and depression scale: HADS-A and HADS-D) (Fig. 4). The Enterobacterales_f (R 2 = 0.302, p = 0.010), Enterococcus (R 2 = 0.220, p = 0.032), and Lactobacillaceae (R 2 = 0.110, p = 0.142) populations were positively correlated with the depression index in patients with IBD (HADS-D) (Fig. 3A). The Enterobacterales_f (R 2 = 0.344, p = 0.005), Enterobacterales_g (R 2 = 0.344, p = 0.005), Enterobacterales group (R 2 = 0.344, p = 0.005), and Bacteroides uniformis (R 2 = 0.346, p = 0.005) populations were positively correlated with the anxiety index in patients with IBD (HADS-A) (Fig. 3B). In addition, even if two IBD/D + patients thought to be outliers were excluded, the correlation between the composition of gut bacteria, except Enterococcus faecium group, and depression index was not different to those included them (Supplementary Figure S6).
Fecal microbiota composition in mice transplanted with IBD/D + -F or IBD/D − -F. IBD-F transplantation significantly caused colitis and anxiety/depression in transplanted mice compared with HV-F transplantation. Therefore, we analyzed the gut microbiota composition of the feces of mice transplanted with IBD/ D + -F or IBD/D − -F (Fig. 5, Supplementary Figures S7 and S8). The α-diversity (OUT richness), but not Shannon's diversity index, was significantly higher in IBD-F than in HV-F (Fig. 5A). The β-diversity was also significantly different between the gut microbiota compositions of mice transplanted with HV-F and IBD-F. Compared with the bacterial composition at the phylum level, the Proteobacteria population was higher in the feces of mice transplanted with IBD-F than in those of mice transplanted with HV-F (Fig. 5B). At the family level, the Enterococcaceae, Bacteroidaceae, and Prevotellaceae populations were increased by the FMTs from IBD-F vs. HV-F. Furthermore, the Enterococcaceae populations were higher in the IBD/D + -F than in the IBD/D − -F (Fig. 5C). The HV-F transplantation also modified gut microbiota in the transplanted mice compared with normal control mice: it increased the populations of Coriobacteriaceae and Acholeptasmataceae, not Enterobacteriaceae. The FMTs from patients with IBD increased the LPS level compared with those from HV (Fig. 5D).

Discussion
Gut microbiota is closely involved in the occurrence of psychiatric disorders through the activation of the MGB axis 29 . Many studies have focused on the role of gut microbiota in the outbreak of neuropsychiatric disorders [29][30][31] . Klebsiella and Lachnospira, which belong to the Firmicutes phylum, are higher, while the Escherichia and Bifidobacterium populations are lower, in the feces of patients with depression than in the feces of healthy individuals 32 . The Enterobacteriaceae and Alistipes populations are over-represented in patients with depression 32 . The induction of depression by stressors triggers colitis and increases the gut Proteobacteria population and bacterial LPS production in mice 11,12 . The FMT from stress-stimulated mice also causes colitis with depression in the transplanted mice 11 .
The fecal microbiota transplantation (FMT) from IBD patients causes the IBD-like colitis in germ-free mice 26,27 . The FMT from Chrna7 KO mice causes depression-like phenotypes and systemic inflammation in www.nature.com/scientificreports/ antibiotics-treated mice via the subdiaphragmatic vagus nerve 33 . The FMT from mice with colitis occurs colitis as well as depression in conventional mice 11 . The FMT from NLRP3-deficient mice alleviates depressive-like behaviors by regulating astrocyte dysfunction in antibiotics-treated mice 34 . The FMT from healthy individuals alleviates IBD in clinic studies 27,28 . These findings suggest that the microbiota are closely connected with the occurrence of IBD and neuropsychiatric disorders, such as depression, through the microbiota − gut − brain (MGB) axis. Several studies have demonstrated that the high prevalence of commensal gut Proteobacteria including Enterobacteriaceae is closely associated with IBD, including UC and CD, overexpression of gut bacterial LPS 25 . However, the character of gut microbiota associated with the occurrence of IBD and depression is not sufficiently understood. www.nature.com/scientificreports/ In the present study, the FMTs from patients with IBD caused colitis and depression in the transplanted mice: they induced the NF-κB + /Iba1 + , LPS + /Iba1 + and IL-1R + cell populations and IL-1β and IL-6 expression in the hippocampus and myeloperoxidase activity and IL-1β and IL-6 expression and suppressed tight junction protein expression in the hippocampus and colon. After the segregation of patients with IBD/D − and IBD/D + , the FMTs from patients with IBD/D − or IBD/D + all caused colitis and anxiety in the transplanted mice, while the FMTs from patients with IBD/D + , but not IBD/D − , caused depression. Parsing anxiety-like behaviors is different from depressive behaviors in humans 35 . The high rate of patients with anxiety disorder has symptoms of depression. Therefore, evaluating anxiety-and depression-like behaviors in mice may be not sufficient to understand the multiple etiologies of patients with anxiety and/or depression 35 . Nevertheless, it is helpful to understand the physiological effects of gut microbiota on the occurrence of anxiety and/or depression through the activation of MGB axis activation. These findings suggest that gut microbiota are closely involved in the occurrence of anxiety and depression.
FMTs from patients with IBD/D + or IBD/D − also induced myeloperoxidase activity and IL-1β and IL-6 expression in the colon. The FMTs from patients with IBD/D + increased the stenosis score and decreased tight junction protein expression more strongly than did the FMTs of patients with IBD/D − . However, after the segregation of patients with UC and CD, the FMTs from patients with UC and CD caused colitis with anxiety/depression in the transplanted mice. Interestingly, the occurrence of colon stenosis was observed at a greater extent in mice transplanted with the feces of CD patients than in those transplanted with the feces of patients with UC, as patients with IBD 36 . These results suggest that the gut microbiota consists of IBD-inducing, -suppressing, and -irrelevant microbes and that some IBD-inducing microbes may cause anxiety with or without depression. www.nature.com/scientificreports/ We also found that the gut microbiota composition of IBD-F was significantly different from that of HV-F: the α-/β-diversities of IBD-F were lower than those of HV-F, and the β-diversity was different between IBD-F and HV-F, as reported previously [37][38][39] . Although the α-diversity was not significantly different between IBD/ D − -F and IBD/D + -F, the β-diversity was different. IBD-F exhibited a higher abundance of the Proteobacteria population compared with HV-F, while the Bacteroidetes population was lower in IBD-F. In particular, IBD/D + -F exhibited a higher abundance of Enterobacteriaceae, Enterococcaceae, and Lactobacillaceae compared with IBD/ D − -F. Among these gut bacteria, the Enterobacterales_f and Enterococcus populations were positively correlated with the depression index in patients with IBD (HADS-D) and the Enterobacterales_f, Enterobacterales_g, Alistipes, and Enterobacterales group populations were positively correlated with the anxiety index in patients with IBD (HADS-A). These results suggest that the IBD with anxiety/depression may be induced in the intestine by Enterobacteriaceae, Enterococcaceae, and Lactobacillaceae.
The fecal microbiota compositions of mice transplanted with HV-F and IBD-F partially matched those of patients with HV-F and IBD, respectively. However, those of mice transplanted with HV-F was significantly different to those of mice transplanted with IBD and normal control mice. The Proteobacteria population, including Enterococcaceae, was increased in the feces of mice transplanted with IBD-F than it was in those transplanted with HV-F, similar to those of IBD-F itself. The Enterococcaceae, Lactobacilliaceae, and Enterobacteriaceae populations were positively correlated with the anxiety/depression index of patients with IBD (HADS-A and HADS-D). However, the Bacteroidaceae, Prevotellaceae, and Helicobacteraceae populations were increased in the feces of mice transplanted with IBD-F compared with mice transplanted with HV-F. Jang et al. reported that the FMTs from mice with or without depression shifted the gut microbiota composition, which persisted for more than 15 days in specific pathogen-free mice 11 . Lee et al. reported that the incidence of cognitive impairment by exposure to a human gut bacterium, Paenalcaligenes hominis, was severe 10-fold in germ-free mice more than in SPF mice 21 . www.nature.com/scientificreports/ These results support the suggestion that the gut microbiota of IBD-F comprises IBD-inducing microbiota, such as Enterobacteriaceae and Enterococacceae, which may be transient and attachable in the gastrointestinal tract. The content of LPS was higher in the IBD-F than in the HV-F. Furthermore, the fecal and blood LPS levels were higher in IBD-F-gavaged mice than they were in HV-F-gavaged mice. The FMTs from IBD patients significantly suppressed tight-junction protein expression in the colon and hippocampus. They also suppressed BDNF + / NeuN + cell population in the hippocampus, while the NF-κB + /Iba1 + , LPS + /Iba1 + and IL-1R + cell populations and IL-1β and IL-6 expression increased. Among the IBD-F, the levels of IBD markers such as IL-1β , IL-6, LPS, and NF-κB + /Iba1 + cells were induced more strongly by IBD/D + -F than by IBD/D − . Exposure to Escherichia coli increases the hippocampal NF-κB + /Iba1 + and LPS + /Iba1 + cell populations and blood and fecal LPS levels and decreases the hippocampal BDNF + /NeuN + cell population and tight-junction protein expression, leading to depression 11,37 . Oral gavage of Escherichia coli or LPS causes colitis and neuroinflammation in mice by suppressing tight-junction protein expression 21 . An IL-1R antagonist mitigates depression 40 . Exposure to stressors such as high-fat diet and ampicillin increases the Proteobacteria population and gut bacterial LPS production 11,41,42 . They induce gastrointestinal inflammation and suppress tight-junction protein expression, leading to a leaky gut, which accelerates the translocation of gut bacteria and their by-products, such as LPS, across the intestinal mucosa. In the present study, we found that the transplantation of IBD-F significantly increased the LPS levels in the blood and feces and caused neuroinflammation with colitis in mice compared with HV-F transplantation. Moreover, IBD patients suffer from leaky gut 43 . Excessive absorption of LPS across the leaky gut can cause systemic inflammation, leading to the neuroinflammation in the brain 44,45 . In the present study, we found that anxiety-/depression-like behaviors and their related markers BDNF expression and NF-κB activation were more severely fluctuated in the FMTs from patients with IBD/D + than in ones from patients with IBD/D − . These results suggest that the FMT from patients with IBD/D + may cause colitis with a leaky gut, which may accelerate the absorption of fecal LPS into the blood and suppress the NF-κB-mediated BDNF expression in the brain, resulting in the occurrence of anxiety/depression through the regulation of neuroinflammation.
However, the FMT from HV suppressed IBD/D + -F (UC/D + -F and CD/D + -F)-induced anxiety-and depression-like behaviors, NF-κB + /Iba1 + , LPS + /Iba1 + , and IL-1R + cell populations, hippocampal IL-1β and IL-6 expression in the hippocampus, blood corticosterone, IL-6, and LPS levels, and colonic myeloperoxidase and IL-6 expression in mice, while the IBD/D + -F (UC/D + -F and CD/D + -F)-suppressed BDNF + /NeuN + cell population increased in the hippocampus. Furthermore, the FMT from HV significantly reduced the IBD/D + -F-induced Enterococcus sp. population and bacterial LPS production in the feces. He et al. reported that the FMT from HV prevented the relapse of seizures and CD activity score after withdrawing the antiepileptic drugs in a patient with CD and epilepsy 46 . Goo et al. reported that the FMT for normal control mice alleviated cognitive deficits and social withdrawal symptoms observed in autism spectrum disorders in Fmr1 KO mice with autistic-like behaviors by increasing the population of Akkermansia muciniphila 47 . These results support the hypothesis that gut microbiota are closely connected with the occurrence of IBD and neuropsychiatric disorders, such as depression, by the regulation of the MGB axis activation.
In conclusion, the outbreak of depression/anxiety may be dependent on the systemic inflammation with a leaky gut through the gut dysbiosis-attributable overproduction of bacterial LPS and suppression of tight junction protein expression in patients with IBD..

Materials.
A De Man, Rogosa, and Sharpe (MRS) medium and Sabouraud dextrose agar (SDA) were purchased from BD (Franklin Lakes, NJ). An antibody for NF-κB was purchased from Cell Signaling Technology (Danvers, MA). Antibodies for CD11c, Iba1, and BDNF were purchased from Abcam (Cambridge, U.K.). Alexa Fluor 488 and Alexa Fluor 594 were purchased from Invitrogen (Carbsband, CA). Enzyme-linked immunosorbent assay (ELISA) kits for TNF-α, IL-6, and IL-10 were purchased from Ebioscience (Atlanta, GA). A QIAamp Fast DNA stool mini kit was purchased from Qiagen (Hilden, Germany).  Table S1). They are not a vegetarian or a meatarian. The stools were collected from volunteers, if antibacterial medications had not been administered within 3 month. Furthermore, HVs had not received analgesics and anti-inflammatory drugs within 3 months. However, most of IBD patients received steroids, immuno-modulators, such as azathioprine or methotrexate, and/or biologics/small molecules for therapeutic purpose. All patients with IBD enrolled in the study were > 13 years of age at the diagnosis of IBD. All diagnoses were confirmed by previously established international criteria based on clinical, endoscopic, histopathological, and radiological findings 48 . Psychometric tools (Patient health questionnaire-9 [HQ-9] and hospital and anxiety depression scale [ADS] were assessed according to the method of Song et al. 49 The study protocol and consent forms for the stool collection were approved by the Committee for the Care and Use of Clinical Study of the Medical School of Kyung Hee University (IRB File No., KHUH 2018-03-006-018 and KHUH 2018-12-004-003). All participants agreed to participate in the study and signed informed consent before the initiation of the study. In a patient aged under 18 (n = 1), informed consent was obtained from a parent. All experiments related to the usage of human feces were conducted in compliance with the principles of the Declaration of Helsinki and the Korean Good Clinical Practice guidelines.

Volunteers. Volunteers
Animals. C57BL/6 mice (male, 6 weeks old, 19-21 g) were purchased from Koatech Inc. (Seoul, Korea). Mice were maintained in plastic cages with the 5-cm raised wire floor, which was designed to protect mice for feeding the feces, under a controlled condition (temperature, 20 ○ C-22 ○ C; humidity, 50% ± 10%; light/dark cycle,  First, to understand the effects of gut microbiota on the occurrence of depression and colitis, each fecal microbiota suspension of the IBD patients or HVs (10 mg feces [wet weight]/kg/day) was orally gavaged into three mice once a day for 5 days, respectively.
Second, to understand the anti-depressive effects of gut microbiota, the fecal microbiota suspension of patients with IBD/D + was orally gavaged into twelve mice once a day for 5 days. From next day after the final gavage of the fecal microbiota suspension of patients with IBD/D + , the fecal microbiota suspension from HVs (10 mg feces [wet weight]/kg/day) or vehicle was orally gavaged once a day for 5 day.
Anxiety-/depression-like behaviors were measured on the next day following treatment with the fecal or bacterial suspension 11,37 . Twelve hours after the final behavioral task, mice were euthanized by the inhalation of CO 2 . Colons and brains were removed for the enzyme-linked immunosorbent assay (ELISA) and immunoblotting assay from mice and stored at -80 °C for ELISA and immunoblotting. Colons and brains were removed for the immunofluorescence assay from mice that were transcardially perfused with 4% paraformaldehyde.
Behavioral tasks. The elevated plus maze (EPM) task was assessed in a plus-maze apparatus (consisting of two enclosed [30 × 7 cm] and two open [30 × 7 cm] arms with 20-cm-high walls extending from a central platform [7 × 7 cm]), as previously reported 11 . The light/dark transition (LDT) task was assessed in a light/dark box apparatus (45 × 25 × 25 cm, consisting of two chambers made of black and white Plexiglass [floor] and polywood [walls] connected by an opening [7.5 × 7.5 cm] located at floor level in the center of the dividing wall), as previously reported 11 . The marble burying (MB) task was assessed in a smooth, opaque plastic cage (30 × 35 × 13 cm) with a 5-cm layer of sawdust and 30 marbles on top of it (five rows or marbles regularly spaced at 2-cm away from the walls), as previously reported 11 . The number of marbles buried for 30 min was counted. The tail suspension test (TST) was assessed on the edge of a table, at 30-cm above it, for 5 min, as previously reported 35 . Mice were judged to be immobile, when they did not move and hung passively. The forced swimming test (FST) was assessed in a round transparent plastic jar (20 × 40 cm 3 ) containing fresh water (25 °C) to a height of 25-cm for 5 min, as previously reported 37 . Mice were judged to be immobile, when they remained floating in the water without struggling. Immunofluorescence assay. The brains and colons were post-fixed with paraformaldehyde, cytoprotected in 30% sucrose solution and cryosectioned. Sectioned tissues were immunostained (Supplementary Methods) and observed by a confocal microscope, as reported previously 21 . ELISA and immunoblotting. Brain and colon tissues were homogenized with radioimmunoprecipitation assay lysis buffer (Biosesang Inc., Seongnam, Korea) containing a protease inhibitor cocktail and a phosphatase inhibitor cocktail at 4 °C. Cytokine levels were determined in their supernatants using ELISA kits (eBioscience, San Diego, CA) according to the method of Jang et al. 11 Corticosterone, IL-1β and IL-6 levels were measured using ELISA kits 11 . Limulus amebocyte lysate (LAL) and myeloperoxidase activity assays. Fecal and blood endotoxin levels were assayed using an LAL assay kit (Cape Cod Inc., E. Falmouth, MA) according to the method of Jang et al. 11 Colonic myeloperoxidase activity was assayed according to the method of Jang et al. 11 . Effects on the occurrence of anxiety/depression in the EPM (a) and LDT tasks (b), TST (c), and FST (d). Effects on the BDNF + /NeuN + , NF-κB + /Iba1 + , LPS + /Iba1 + , and IL-1R + cell populations (e) and IL-1β expression (f) in the hippocampus (f). Effects on the corticosterone (g), IL-1β (h), and LPS levels (i) in the blood. (j) Effects on the myeloperoxidase (MPO) activity in the colon. (k) Effects on the fecal Enterococcus sp. population. (B) Effects in IBD/D − -F (CD/D − )-gavaged mice. Effects on the occurrence of anxiety/depression in the EPM task (a) and TST (b). (c) Effects on MPO activity in the colon. (d) Effects on the fecal Enterococcus sp. population. HF, IBD/ D + -F, or IBD/D − -F suspension was orally transplanted in mice once a day for 5 days. Control mice (CON) were treated with vehicle (saline) instead of fecal suspension. From the next day, HV-F suspension was gavaged in IBD/D + -HV and IBD/D − -HV mice once a day for 5 days. Con, IBD/D − , and IBD/D + mice were treated with vehicle (saline) instead of fecal suspension. Data values were indicated as mean ± SD (n = 6). Fecal Enterococcus sp. population was analyzed by using qPCR. Means with same letters are not significantly different (p < 0.05). All data were analyzed using ANOVA with Tukey's multiple comparisons test.