O-Antigen Modulates Infection-Induced Pain States

The molecular initiators of infection-associated pain are not understood. We recently found that uropathogenic E. coli (UPEC) elicited acute pelvic pain in murine urinary tract infection (UTI). UTI pain was due to E. coli lipopolysaccharide (LPS) and its receptor, TLR4, but pain was not correlated with inflammation. LPS is known to drive inflammation by interactions between the acylated lipid A component and TLR4, but the function of the O-antigen polysaccharide in host responses is unknown. Here, we examined the role of O-antigen in pain using cutaneous hypersensitivity (allodynia) to quantify pelvic pain behavior and using sacral spinal cord excitability to quantify central nervous system manifestations in murine UTI. A UPEC mutant defective for O-antigen biosynthesis induced chronic allodynia that persisted long after clearance of transient infections, but wild type UPEC evoked only acute pain. E. coli strains lacking O-antigen gene clusters had a chronic pain phenotype, and expressing cloned O-antigen gene clusters altered the pain phenotype in a predictable manner. Chronic allodynia was abrogated in TLR4-deficient mice, but inflammatory responses in wild type mice were similar among E. coli strains spanning a wide range of pain phenotypes, suggesting that O-antigen modulates pain independent of inflammation. Spinal cords of mice with chronic allodynia exhibited increased spontaneous firing and compromised short-term depression, consistent with centralized pain. Taken together, these findings suggest that O-antigen functions as a rheostat to modulate LPS-associated pain. These observations have implications for an infectious etiology of chronic pain and evolutionary modification of pathogens to alter host behaviors.


Introduction
The mechanisms by which infections induce pain are not understood. Although inflammation may cause or contribute to pain [1,2], few studies have examined infection-associated pain. UTI is the second-most common infectious disease, afflicting half of all women during their lifetime and costing billions of dollars in annual treatment costs in the U.S. [3,4]. Most UTIs are caused by UPEC and result in pelvic pain and voiding dysfunction [5]. Interactions between UPEC and the urothelium that lines the bladder lumen result in local cytokine/chemokine production and accumulation of urinary IL-6 and IL-8 [6,7]. IL-8 secretion, in turn, drives a robust innate immune response characterized by an influx of neutrophils that clear bacteria from the bladder. Yet it remains unclear how these events relate to pain.
An initial study of pain in murine UTI found that UPEC resulted in peripheral thermal sensitivity, measure of pain, that was abrogated in TLR4-deficient mice [8]. Since TLR4 mediates bladder responses to LPS, a major inflammatory mediator, these findings suggested UTI pain is due to inflammation. However, we recently examined pain during acute UTI by quantifying mechanical allodynia of the pelvic region, since visceral pain is manifested as cutaneous hypersensitivity [9]. Although UPEC induced TLR4-dependent acute allodynia, allodynia was not correlated with inflammation. Indeed, LPS purified from an asymptomatic bacteriuria (ASB) E. coli strain and from UPEC elicited similar inflammatory responses, but UPEC LPS produced allodynia whereas ASB LPS did not. These findings suggested that TLR4-dependent infection pain is a function of LPS structure independent of inflammation.
LPS is comprised of three major domains: the acylated lipid A component embedded in the outer membrane, a core oligosaccharide, and the O-antigen polysaccharide. The lipid A component is the chief mediator of inflammation through its interactions with TLR4 [10,11,12]. However, gram-negative bacteria derive significant diversity from differential forms of LPS, and E. coli generate much diversity at the level of O-antigen, where more than 180 serotypes have been identified. Although this diversity suggests functional importance for interactions between E. coli and hosts, the known functions of O-antigen are largely limited to enhancing bacterial robustness [13,14,15]. Since UPEC and ASB LPS elicit markedly different pain responses despite similar inflammation, we hypothesized that O-antigen modulates E. coli infection pain. Here, we demonstrate that O-antigen modulates TLR4-dependent pain phenotypes of E. coli, suggesting a novel target for pain control.

Transient bacterial infection induces acute or chronic allodynia
We previously generated an O-antigen biosynthesis mutant in UPEC by deleting the gene encoding O-antigen ligase, waaL, from the archetypal cystitis isolate NU14 [16]. NU14DwaaL is highly attenuated in murine UTI and represents a convenient strain to test the role of O-antigen in pain. Pain behavior was quantified using the classic approach of tactile allodynia in response to mechanical stimulation with calibrated von Frey filaments applied to the pelvic region [9]. Responses of jumping, abdominal retraction, or pelvic grooming were quantified relative to baseline, and mice that were infected with NU14 via transurethral catheter exhibited acute allodynia (Fig. 1A). Allodynia was greatest at 2 days post-infection and decayed over approximately two weeks, consistent with our previous observations [17], and serial NU14 infections also induced allodynia.
In contrast to wild type UPEC, serial infection with NU14D-waaL induced chronic allodynia following a second infection and third infection (P,0.05 and P,0.01, respectively) despite evoking no response to an initial infection (Fig. 1A). Allodynia was specific to the pelvic region and not observed in the hind paw, and the absence of altered body mass suggested that overall health was not affected (Tables S1 and S2). Adaptive immune responses were next examined as potential mediators of chronic pain following serial infection. Rag1-deficient mice lacking T and B cell responses developed chronic pain in response to DwaaL, and adoptive transfer of CD90+ splenocytes from mice with chronic allodynia did not induce allodynia in recipients, thus excluding adaptive immune mechanisms (Fig. S1). We previously reported that DwaaL is rapidly cleared from the bladder after an initial infection [16], and the waaL mutant was undetectable at 14 days following a third infection, whereas the majority of NU14-infected mice retained detectable bladder colonization despite resolution of pain (Fig. 1B). Residual NU14 colonization is consistent with previous reports of stable intracellular bacterial populations of UPEC in murine UTI [18,19], but the absence of detectable DwaaL suggests that chronic allodynia is not due to residual bacteria. Together, these findings

O-antigen modulates chronic allodynia
To exclude the influence of UPEC virulence factors in UTI pain, we examined allodynia in response to infection with the K-12 E. coli strain SW874 that bears a deletion of the entire wz* gene cluster encoding O-antigen biosynthesis genes and lacks UPEC virulence factors [20,21]. SW874 caused chronic allodynia that persisted at least 35 days following a rapidly-cleared, single infection (Figs. 1C and D). Complementing the O-antigen defect in SW874 with the plasmid pWQ288 encoding Klebsiella pneumoniae O2a antigen [22] abrogated allodynia ( Fig. 1C, P,0.05 . Despite the disparate pain phenotypes, clearance was similarly rapid for SW874 with/without O-antigen (45.5643.8 CFU/ bladder for SW874 and 55.2636.9 CFU/bladder for SW874 pWQ288 at 35 days after infection, P.0.05). Moreover, four of ten SW874-infected mice had no detectable bladder colonization at the conclusion of the experiment while exhibiting allodynia of 459%6299 of baseline, so pain may persist even after complete bladder sterilization. These findings suggest that transient E. coli infection can cause chronic pain and that O-antigen modulates E. coli pain phenotypes in pathogens and non-pathogens alike.
To confirm that LPS underlies the differential pain phenotypes of E. coli, we examined responses to bladder instillation of purified LPS. As previously reported [17], LPS from NU14 and ASB E. coli strain 83972 caused transient allodynia and no allodynia, respectively ( Fig. 2A). In contrast, a single instillation of SW874 LPS induced allodynia that persisted significantly above baseline throughout the 14-day time course ( Fig. 2A, P,0.05). Similar to infection with intact bacteria, purified DwaaL LPS did not induce allodynia following a single instillation. Together, these data support a role for O-antigen modulation of pain and demonstrate that such modulation is not restricted to the O18 serotype of NU14 or exclusive to E. coli serotypes.

Allodynia phenotypes are not associated with inflammation
UPEC generate TLR4-dependent cytokine secretion and neutrophil influx [6,23,24], so it is possible that differential inflammatory processes underlie allodynia phenotypes modulated by O-antigen. Indeed, in a limited-inoculum model mimicking early infection, we previously observed that DwaaL induces greater neutrophil influx than parental NU14 [16]. To exclude differential inflammation as the determinant of differential allodynia, we quantified bladder pathology and neutrophil influx. Scoring of tissue sections by a blinded reviewer of bladders obtained 6 hours or 14 days after instillation of 83972, NU14, DwaaL, SW874, or saline revealed similarly elevated inflammation scores for all E. coli relative to saline at 6 hours after instillation that was resolved by 14 days following instillation (Figs. 3A-C, P,0.05).
Since the magnitude of acute allodynia induced by NU14 is independent of neutrophil influx [17], urinary neutrophil myeloperoxidase (MPO) was quantified to determine if chronic allodynia is somehow associated with differential neutrophil influx. A transient MPO increase was induced similarly in mice infected with 83972 or SW874 despite the markedly difference allodynia phenotypes (Fig. 3D). Serial infection with NU14 or DwaaL also induced transient increases in urinary MPO (Fig. 3E), yet MPO was not significantly different between strains at any time point or between any single infection and serial infection. Thus, differential allodynia phenotypes that are modulated by O-antigen do not

Infection-induced chronic allodynia is centralized
Chronic pain can result from a continuous peripheral stimulus or can be neuropathic as a result of enhanced sensory processing within the central nervous system [1]. Since SW874 and DwaaL induce chronic allodynia persisting after bacterial clearance, we hypothesized that E. coli-induced chronic allodynia is associated with persistant centralizedchanges in excitability. Using an isolated murine preparation of the sacral spinal cord, the primary location of bladder sensory and micturition reflex circuits [25,26,27], spontaneous firing was quantified in ventral roots as a measure of coupling sensory and motor functions (Fig. 4A). This approach quantifies motor output of the flexion (withdrawal) reflex, a reflex that has long been used as a pain index [9,28].
Spontaneous sacral activity was quantified using pCLAMP software to identify discrete action potential characteristics and thereby determine the number of spontaneously firing neurons within the flexion reflex of a spinal cord preparation (Fig. 4B). NU14 acute allodynia was associated with significantly higher spontaneous activity relative to saline controls, and this activity returned to baseline in mice with resolved allodynia at 14 days after infection, suggesting an absence of central effects (Fig. 4C, P,0.05). Spinal cords of mice with chronic allodynia from a SW874 or DwaaL infection also exhibited significantly higher spontaneous firing relative to saline controls (P,0.001, P,0.05, respectively). These data indicate that E. coli-induced sacral spinal hyper-excitability is associated with allodynia and suggest that chronic allodynia has a central component.
To further examine spinal hyper-excitability, we quantified short-term depression (STD) of motor output, a measure of normal plasticity [29], by stimulating dorsal inputs and recording ventral outputs (Fig. 4A). Individual dorsal roots were stimulated with five consecutive current pulses (P1-P5) at 40 millisecond intervals at multiple stimulus intensities ( Fig. 4D; 1-to 5-fold of the minimum current intensity evoking a response). For spinal cords of control mice, sequential stimuli evoked diminishing responses relative to the first pulse (P1), consistent with STD of dorsal horn sensory inputs (Fig. 4D). Responses to pulses P2-P5 were then normalized to P1 to facilitate comparisons between groups. Spinal cords from mice with acute allodynia induced by NU14 infection did not exhibit deficits in STD for any pulse or at any gain (Fig. 4E). In contrast, significant STD deficits were observed in spinal cords of mice infected with SW874 and DwaaL that exhibited chronic allodynia (Figs. 4E-G). Both SW874 and DwaaL spinal cords exhibited significant STD deficits at 26 stimulus intensity and at P3 and/or P4 (Fig. 4E). These STD deficits at P3 and P4 were also manifested at multiple stimulus intensities (Figs. 4F and G). These findings suggest that chronic allodynia induced by transient E. coli infection is associated with persistant centralized hyper-excitability characteristic of chronic pain.

UPEC elicits diverse tactile allodynia responses via Oantigen
To further characterize O-antigen modulation of pain, we deleted the NU14 wz* O-antigen gene cluster [30] to generate NU14Dwz. For genetic complementation, low-copy plasmids containing the wild type wz* clusters of NU14 and 83972 were isolated from genomic libraries. The morphology of gram-negative bacterial colonies on agar is classically described as smooth in appearance for strains expressing O-antigen, whereas strains lacking O-antigen or expressing minimally-glycosylated O-antigen appear as colonies with a rough morphology (Fig. 5A). Wild type NU14 exhibited smooth morphology, whereas NU14Dwz appeared rough. Complementation with the NU14 wz* gene cluster plasmid rescued the smooth phenotype. Although the 83972 wz* plasmid did not rescue the NU14 wz* phenotype, this is consistent with the absence of a typeable serotype for 83972 and a rough phenotype [31].
Serial infection with wild type NU14 or with NU14Dwz complemented with the NU14 wz* plasmid elicited only acute allodynia (Fig. 5B). In contrast, NU14Dwz bearing a control plasmid that contains a region of the human X chromosome induced chronic allodynia that was significantly elevated at day 14 following infection (P,0.05). Expression of the 83972 wz* cluster in NU14Dwz entirely abrogated allodynia. These findings suggest that the pain phenotype of a strain can be modulated across a spectrum by O-antigen.

Discussion
The role of O-antigen in host responses is understudied, and a role for O-antigen in modulating pain was unknown. These data demonstrate that allodynia elicited by murine UTI corresponds with spinal cord hyperexcitability, and this UTI pain response is modulated by O-antigen expression. While limited studies suggest that O-antigen modulates inflammatory responses [32], and UTI pain is TLR4-dependent, this pain appears separable from inflammation because UTI pain is independent of bladder pathology, neutrophil influx, or TLR4 expression on hematopoietic lineages ( [17] and this study). TLR4-induced hypersensitivity appears specific for a given strain, since a single DwaaL infection did not render mice more sensitive to subsequent instillation of NU14, 83972, or capsaicin, a known activator of bladder sensory afferents (Table S3). Conversely, a pilot experiment examining allodynia in response to an initial NU14 infection did not reveal any influence by subsequent DwaaL infection relative to saline (not shown), further suggesting the specificity of inducing chronic pain. Subtle differences in initial inflammatory responses may be a cofactor contributing to differential sensory responses, since cytokines can modulate neural responsiveness [33]. However, inflammation was not different among strains with different pain phenotypes (Fig. 3). Therefore, TLR4 mediates UTI pain, and pain intensity and duration are determined a priori by initial TLR engagement that is subject to modulation by O-antigen.
Together, our data suggest a model whereby O-antigen acts as a rheostat that modulates LPS-associated pain across a spectrum (Fig. 5C). At its most benign with ASB strain 83972, this continuum spans from pain suppression [17] to a mild transient pain induced only by serial infection with 83972Dwz* (Fig. S3). In contrast, NU14 can exhibit a range of pain phenotypes depending upon O-antigen status: a null phenotype, acute pain, an ''adaptive chronic'' pain elicited by serial infection, or chronic pain following a single infection of NU14Dwz*. SW874 is similarly capable of either null or chronic pain phenotypes. These strain-specific differences in pain phenotypes are subject to modulation by Oantigen but are defined by other genetic factors. The LPS core oligosaccharide is one potential factor, but an in silico analysis based on PCR primers that discriminate among the five E. coli LPS core oligosaccharides [34] identified R1 as the LPS core of 83972 as (Table S4). R1 is common among UPEC and is expressed by NU14 [35], so R1 is associated with strains with pain phenotypes at either end of the spectrum. Nor is the chronic pain phenotype unique to R1 because SW874 expresses the K-12 core. Nonetheless, differences in O-antigen modulation of pain may mediate health and disease because gram-negative bacteria typically produce a distribution of LPS species varying in the extent of glycosylation [36,37]. So it is possible that bacteria modulate pain phenotype during pathogenesis and/or commensal association.
The mechanism by which E. coli induce TLR4-dependent chronic pain may differ from other pain models. Initial infection with DwaaL presumably induces hypersensitivity to a subsequent stimulus, perhaps by known mechanisms that upregulate sensory responses [38], whereas SW874 may induce such events in a single infection. This is reminiscent of studies where instillation of yeast zymosan into neonatal rats induces adult hypersensitivity and increased bladder content of pain-related neuropeptides [39,40]. This is also consistent with recent findings by Farmer and colleagues where serial Candida infection resulted in increased peptidergic nociceptor and sympathetic fiber density in vulvar tissue, demonstrating a mechanism for peripheral hypersensitivity to infection [41]. However, that same peripheral mechanism is unlikely to mediate chronic pain in our UTI model because we did not observe increased sensory nerve fiber density in bladders of mice with chronic allodynia (data not shown).
Many microbial species coexist with human hosts without inducing pain, while different strains of the same species cause symptomatic responses for pathogens as diverse as E. coli and Candida albicans. Clinical data suggest prior infection as an etiology or sensitizing factor for chronic pain syndromes. For example, interstitial cystitis/painful bladder syndrome and vulvodynia are associated with a history of UTI and candidiasis, respectively [42,43]. Our data thus contribute to an emerging picture of an infectious basis for at least a subset of individuals afflicted with chronic pain. We also observed TLR-dependent allodynia induced by gram-positive S. milleri oral bacteria (Rudick et al, in preparation). Similar to E. coli, S. milleri strains also exhibited a spectrum of pain phenotypes, from null to acute to chronic allodynia, indicating that TLR-dependent pain is a general phenomenon that extending beyond E. coli. Therefore, we suggest microbes evolved surface features that modulate host behaviors to their advantage through differential activation of pattern recognition receptors. Such modulation is likely widespread and extends beyond pain to other neurally-regulated processes. Indeed, T. gondii relieves rodents of their innate fear of cat predation by specifically altering fear of feline odors, and this is correlated with parasite cysts in the amygdala [44]. But while T. gondii also has been correlated with altered human behaviors, the mechanism of T. gondii behavioral modulation is unknown [44,45,46,47]. Since T. gondii stimulates TLR9 [48,49], we speculate that T. gondii may modulate behavior by a TLR-mediated process analogous to Oantigen-mediated pain modulation. However, in contrast to behavioral change with T. gondii cysts, data presented here also indicate that durable effects may persist after microbial clearance.
In summary, O-antigen modulates E. coli pain phenotypes to alter pain duration and severity during infection. TLR4-dependent pain may persist long after bacterial clearance, is independent of inflammation, and can result in chronic pain -supporting the possibility of an infectious etiology for chronic pain.

Ethics statement
All animals (mice) were housed in Northwestern's Center for Comparative Medicine and were cared for only by trained facility personnel. Mice were maintained in enriched enviroments and monitored frequently to confirm good health and absence of visible signs of stress. All experimental protocols were approved by the Northwestern IACUC and follow NIH guidelines to minimize any stress or pain, and all scientific personnel were appropriately trained and certified, including euthanasia training. All experiments were designed to minimize the number of mice needed to obtain statistically significant findings, and experiments were terminated as soon as possible.

Mice and bacteria
Female mice were purchased from JAX at 6-12 weeks of age and maintained in the Center for Comparative Medicine and used in Northwestern IACUC-approved protocols. Bone marrow chimeras were generated by gamma irradiating recipient mice with 1000 rad total from a Cs-137 source and reconstituting with 10 7 bone marrow cells from age-matched donors; recovery after 12 weeks of reconstitution was confirmed by LPS induction of CD80 and CD86 on splenic macrophages.
NU14 is a cystitis strain of O18 serotype, 83972 is an asymptomatic E. coli strain of a non-typeable serotype, and SW874 is a K-12 E. coli strain [21]. Targeted wz* deletion mutants of NU14 and 83972 were constructed using l-Red mutagenesis with strain-specific primers spanning the gnd and JUMP-start sequences [50,51]. Complementation plasmids were isolated from fosmid libraries of NU14 and 83972 in pCC1FOS (EPICENTRE Biotechnologies). A plasmid containing a 43 kb region of the human X chromosome was also isolated using the kit positive control DNA. Mutants and fosmids were confirmed by DNA sequence analysis.

Murine UTI and pelvic pain behavior
Bacteria were cultured under static conditions and used to infect mice by instillating 10 ml containing 10 8 CFU in a non-reflux UTI model [52]. Allodynia was quantified in response to von Frey filament stimulation to the pelvic region or the hind paw by a blinded tester [17]. Briefly, mice were adapted to the test chamber environment (5-10 min.) and von Frey filaments were applied 10 times each to the pelvic region, moving the stimulus with each successive fiber application to avoid wind up. For paw sensitivity, filaments were applied to the hind paw in a modified ''up-down'' stimulus protocol to establish a 50% threshold.

Pathology scoring and MPO assay
Mice were instilled with saline or infected with 10 8 CFU E. coli. Bladder sections (10 mm) were stained with hematoxylin-eosin and evaluated by a blinded reviewer. Sections were scored on a scale of 0-5 for edema, leukocytic influx, and disruptions in urothelial integrity. Data points reflect the mean scores of individual mice from an evaluation of 3 non-serial sections. MPO was quantified in urine by ELISA (Hycult Biotech).

Sacral spinal cord recording
The spinal cord was prepared as previously described [26,27]. Briefly, the spine was opened at upper lumbar under deep anesthesia, perfused with artificial cerebro-spinal fluid (ACSF), and then transected at the middle of the lumbar enlargement immediately after sacrifice, and the distal part of the cord with dorsal/ventral roots was transferred to a dish containing ACSF. After separating sacral ventral roots 1-3 (S1-S3) and dorsal roots on each side of the cord and removing all other roots, six recording electrodes mounted on the ventral roots (S1-S3 on each side) were connected to six DAM 50 amplifiers (WPI) in differential mode with 10006 gain, high-pass filtering at 300 Hz, and low-pass filtering at 20 kHz. Amplifier outputs were transferred to an A-to-D interface (DT1322A, Molecular Devices), and signals were digitized at 50 kHz and acquired using pCLAMP v9.1 software (Molecular Devices). Spontaneous action potentials were recorded for 30 min. For each animal, the firing units per root were the average from all recorded roots.
To record evoked motor outputs in response to dorsal stimulation, the stimulation threshold was first determined by adjusting the intensity of a 0.2 ms current pulse. Then 5 pulses at 25 Hz with intensities of 1, 1.5, 2, and 5 times the threshold stimulus were used. The evoked peak-to-peak compound action potential (coAP) was averaged for each pulse, P2-P5, relative to the coAP at P1.

Statistical approaches
Results were expressed as means 6 standard errors of the mean. Data were analyzed with the Student t test or ANOVA followed by a Kruskal-Wallis test, by the Dunn post test, or with analysis of variance, followed by Dunnett's post test; Prism software, version 5 (GraphPad), was used, as appropriate. Differences were considered statistically significant at P,0.05. Figure S1 Adaptive immune responses do not underlie DwaaL-induced chronic allodynia. (A) Donor +/+ were mice serially infected with DwaaL exhibited chronic allodynia (n = 5). (B-D) CD90+ splenocytes from donor mice in (A) or saline-treated control donors were transferred to naïve recipients (n = 4 saline, n = 6 DwaaL), and allodynia was quantified for four days after transfer. Pelvic sensitivity (D) and paw sensitivity (E) were unaltered by transfer of CD90 splenocytes. (F) Chronic allodynia following a second infection with DwaaL was similar in +/+ (n = 9) and Rag1 2/2 mice (n = 9). (G) At 14 days following a third infection with DwaaL, bladders were harvested from mice, homogenized, and plated onto selective agar to quantify colonization. Except for a single Rag1 2/2 mouse, no colonization was detected. (TIF) Figure S2 Chronic allodynia from DwaaL infection requires TLR4. Mice (n = 10) were infected serially with DwaaL, and tactile allodynia was quantified in +/+ or TLR4 2/2 B6 mice. Allodynia following second and third infections was significantly reduced in TLR4 2/2 mice relative to +/+ mice (P,0.01). (TIF) Figure S3 O-antigen modulation of 83972-associated pelvic responses. Tactile allodynia of mice infected with 83972 or 83972Dwz with/without a plasmid encoding the wz* cluster of K. pneumoniae in pWQ288 in response to two sequential infections (n = 10). 83972Dwz induced significant allodynia in response to serial infection that was not observed in mice receiving serial infection with 83972Dwz/pWQ288 (P,0.05).

(TIF)
Table S1 Paw sensitivity determined by 50% threshold (*p,0.05). Tactile allodynia of the hind paw was assessed to determine the specificity of pelvic responses. No significant differences were detected. (DOC)   BLAST analyses with conserved primer sequences known to identify E. coli sequences identified homologous sequences representing the R1 core. The region ''amplified'' in silico was 547 bp and corresponded to a known R1 core enzymatic function, gylcosyl transferase. (DOC)