Time-dependent displacement of commensal skin microbes by pathogens at the site of colorectal surgery

Objective: To characterize perioperative changes of the skin microbiome in patients undergoing elective colorectal surgery and to determine this relationship with surgical site infections (SSIs). Summary of Background Data: Despite the prevalence and signicance of SSIs, their pathogenesis remains poorly understood. Although the complexity of the human skin microbiome has been the subject of recent studies, it is not known whether alterations among commensal microbes contribute to the incidence of SSIs. Methods: Skin swabs of the abdominal wall and chest wall from 60 study subjects were collected before and after colorectal surgery, in addition to intraoperative samples including subcutaneous fat and colonic contents. Bacterial 16S rRNA gene sequences were sequenced and analyzed. Results: Alpha diversity on the skin decreased in the perioperative period but later recovered at the postoperative clinic visit. Alpha diversity of the subcutaneous fat increased signicantly between the beginning and end of these operations, with an increase in abundance of gut microbes also seen within luminal contents after colon resection. In the early postoperative period, the abundance of Enterobacteriaceae increased at the site of surgical incision, with a concomitant decrease in commensals including Corynebacterium and Propionibacterium. Only one patient developed a wound infection. Incisional skin swabs from this patient demonstrated a sharp postoperative increase in the abundance of Enterococcus, which was later cultured from wound drainage at the time of SSI diagnosis. Conclusions: In most patients, we observed a transient postoperative loss of skin commensals at the surgical site, which were replaced by potential pathogens and anaerobes from the gut. We postulate that real time monitoring of the skin microbiome, in parallel with improved knowledge with the gut microbiome and gastrointestinal surgery, could provide actionable ndings about the pathogenesis of SSIs.


Introduction
Surgical site infections (SSIs) remain an expensive cause of morbidity after colorectal surgery 1 . It is commonly said that SSIs are preventable, but this idea remains largely unproven 2 . This notion can be traced back to speculations from Lister and Halsted that infections result from wound contamination 3 .
Today, the problem of SSIs persists despite the ubiquity of novel products developed to maintain intraoperative sterility and barriers. Just in the last decade, studies have assessed the role of preoperative skin preparation 4 , antibiotic coated sutures 5 , adhesive drapes 6 , wound protectors 7 , and absorbable antibiotic coated sponges 8 left within wounds at closure. The failure to eradicate SSIs suggests that our understanding of SSI pathogenesis is incomplete.
It is generally recognized that infections represent a context-dependent host-microbe interaction 9 , but this relationship is not well understood with regard to SSI pathogenesis. The focus on contamination may underestimate the importance of host factors and non-microbial environmental factors. The outsized emphasis on reducing intraoperative contamination may be because it is easier to operationalize than interventions regarding the myriad genetic and environmental determinants of surgical outcomes. Still, it is known that some patient characteristics increase the risk of SSIs, including in ammatory bowel disease 10 , obesity 11 , smoking 12 , and procedures such as ERCP 13 . It is not fully known why such host factors increase the risk of SSI.
The skin microbiome is a host factor that may play an underappreciated role in the pathogenesis and prevention of wound infections. Knowledge of the skin microbiota lags behind the gut 14 , but the existing literature makes it clear that the skin harbors clinically relevant microbial communities 15,16 that interact with the host immune system 16 . Recent experimental studies have demonstrated contributions of skin microbes to neoplasia 17 , in ammation 18 , and the pathogenesis of infection. Other studies suggest that wound healing and in ammation are also impacted by both commensal and pathogenic skin bacteria [22][23][24][25][26][27] .
To date, investigations of the microbiome of human wounds have focused upon chronic wounds, e.g. diabetic ulcers. Few studies have chronicled the bacteriology of surgical incisions [28][29][30][31] . In this study, we utilized a high-throughput approach to characterize the skin microbiota of adults undergoing colorectal surgery. To our knowledge, there are no other studies of this subject in colorectal surgery, where the incidence of SSI remains as high as 20% in some studies 32 . We collected a large number of skin swabs and fecal samples during the perioperative period to clarify perioperative changes in the skin microbiota.
Additionally, we sought to determine if postoperative skin dysbiosis contributed to SSI.

Patient Selection
Adults undergoing outpatient evaluation prior to colorectal surgery at University of Pittsburgh Medical Center were recruited to participate in this study (University of Pittsburgh PRO15080161) from January 2016 through March 2018. Patients were eligible for participation if they were scheduled to undergo resection of the small or large intestine.

Patient Factors
All patients underwent surgery via an enhanced recovery protocol 33 after administering a chlorhexidine bath on the night before surgery. All subjects undergoing total colon or left colon surgery or proctectomy underwent mechanical bowel preparation prior to surgery; subjects undergoing right or transverse colon surgery underwent mechanical bowel preparation based on surgeon preference. No oral antibiotic regimen was used based on institutional protocol except for six patients early in the study who received neomycin and metronidazole. Chlorhexidine solution was used as intraoperative skin preparation unless an allergy had been noted. Preoperative intravenous antibiotics (ceftriaxone and metronidazole) were administered; if necessary, appropriate antibiotic prophylaxis was chosen based on patient allergy pro le.
During surgery, laparotomy or extraction site wound protectors were utilized unless the incision was too large to accommodate the device. All operative personnel changed gloves and used clean surgical instruments for abdominal wall closure. Following fascial closure, subcutaneous tissues were irrigated with sterile saline. Sterile wound dressings of gauze with tape were left in place until postoperative day 2; following dressing removal, incisions were redressed only in the presence of wound seepage or if the incision was in a pannus fold. Signs of wound infection (erythema, purulent drainage) were assessed daily by the clinical team.
Sample Collection and Preparation of Bacterial DNA for 16S rRNA amplicon sequencing For a detailed description of sample collection and microbiome analyses, see eMethods. Brie y, skin swabs of the largest abdominal incision and of the anterior chest wall were collected at preoperative clinic visits, during the perioperative hospitalization, and at postoperative clinic visits. Rectal swabs, fecal samples, and/or ileostomy e uent were collected during the perioperative hospitalization. Additional sample types included intestinal contents harvested from the surgical specimen, swabs of the subcutaneous tissue during opening and closing of the abdominal incision, and wound drainage if present.

Comparison of microbial communities across body sites
We collected 990 total samples from 60 patients. Of these, 882 samples (eTable 1) passed quality ltering after sequencing and ampli cation. As expected, alpha diversity (observed OTUs metric) was higher in fecal samples, rectal swabs, and colonic luminal contents than in skin swabs (eFigure 1). Among skin swabs, alpha diversity was similar between chest swabs and abdominal swabs at all time points except during the hospital stay ( Fig. 1). On POD 1 and 2 we observed signi cantly lower alpha diversity within abdominal swabs. Interestingly, alpha diversity did not differ signi cantly before and after skin preparation with chlorhexidine (p = 0.2616). However, 16S rRNA analyses of bacterial DNA cannot discern between live and dead bacterial cells, so this does not necessarily indicate that skin decontamination was inadequate.
As also expected, beta diversity analysis (principle coordinate analysis of community composition) demonstrated that skin samples formed a discrete group from fecal samples and luminal contents, but considerable overlap existed between abdominal skin swabs and chest wall swabs for either preoperative, hospital stay or postoperative clinic visit ( Fig. 2A). Differences between sites were signi cant by PERMANOVA analysis, as were pairwise PERMANOVA comparisons between sites, though with only modest R-squared values (Abdomen vs Fecal, R 2 = 0.15437; Chest vs Fecal, R 2 = 0.21417).
Temporal changes in the perioperative microbiome Among chest swabs, alpha diversity decreased during the time between preoperative clinic visit and day of surgery, perhaps re ecting the self-administered chlorhexidine bath on the night before surgery. After surgery, alpha diversity in chest swabs quickly returned to preoperative levels despite perioperative antibiotics and subsequently did not change signi cantly (Fig. 1). By contrast, alpha diversity near the abdominal incision decreased on the day of surgery, remained signi cantly lower during hospitalization until day 3, and slowly returned to normal by the postoperative visit (Fig. 1).
Beta diversity analyses indicated that preoperative clinic, perioperative, and postoperative clinic skin swabs overlap considerably in PCoA space when visualized as a scatter plot. This was true for swabs of both the abdomen (Fig. 2B) and the chest (Fig. 2C). Another way to visualize differences in beta diversity differences between time points is to consider the distribution of weighted UniFrac distances between abdominal skin sample groups. We found that, while distances between sample group comparisons were small, the distances between preoperative and postoperative clinic samples were even smaller than distances of either group to postoperative hospital stay samples, though these statistical differences did not reach signi cance (Fig. 3A).

Intraindividual variation of skin microbiome
To quantify the extent of variation within patients, we calculated distances for each subject between preoperative samples, postoperative clinic samples, and the earliest postoperative hospital stay sample (Fig. 3B). Interestingly, we found that preoperative and postoperative clinic visits samples from the same patient were more similar than comparisons between all preoperative clinic (all patients) and all postoperative clinic visits samples (all patients), illustrating the resilience and personalized nature of the skin microbiome even in the perioperative period.
We also tracked UniFrac distances over time for abdominal and chest skin samples by comparing samples from each patient against a common reference point (preoperative samples from the same patient) (Fig. 3C). All patients' samples diverged considerably from their preoperative clinic visit sample, and in abdomen skin samples this distance peaked by POD 2 indicating that this is where the abdominal microbiome became most perturbed. Chest skin samples did not show the same degree of variation over time and were on average more similar to the preoperative samples.
To assess wound contamination during the surgery, we investigated swabs of subcutaneous fat at the beginning and end of the operation. Indeed, alpha diversity of the subcutaneous fat increased between the beginning and end of these operations, indicating an increase in bacterial load (eFigure 2, p = 0.03524). In fact, many subcutaneous fat samples from the beginning of surgery did not pass quality control ltering for sequencing data, presumably because so few bacterial cells were present. By contrast, swabs collected at closure tended to sequence well and possessed microbial communities relatively similar to intestinal samples. Beta diversity (community composition) of the subcutaneous swabs collected at the beginning and the end of the operation differed signi cantly (PERMANOVA p-value < 0.05) (not shown).
We also compared weighted UniFrac distances of the closure swabs to colonic contents of the same patient and from other patients. The subcutaneous fat microbiome of the extraction incision immediately after opening was highly dissimilar from luminal contents sampled intraoperatively (eFigure 2), regardless of whether the lumen contents were from the same patient or from other subjects. The same was not true for the subcutaneous fat swabs collected at the time of wound closure of the incision.
Rather, we found that the subcutaneous fat microbiome at closure was surprisingly similar to luminal contents sampled intraoperatively from the same patient but not contents from other patients (eFigure 2; P = 1.776e-15). This result provides evidence that microbes from the lumen of the resected colon colonize the wound during the surgery, although these microbes are not necessarily those involved in SSI pathogenesis several days later.

Analysis of Taxonomic Composition
We found multiple taxa within abdominal skin swabs that differed across time points, some of which were present in high abundance in some subjects. In the rst two days following surgery, the abundance of Enterobacteriaceae increased, with a concomitant decrease in Corynebacteriaceae ( Fig. 4A and eTable 2). It is known that ampli cation and sequencing of the V4 region of bacterial 16S rRNA genes is not accurate in measuring the abundance of the common skin genus Propionibacterium. For this reason, we utilized qPCR with Propionibacterium speci c primers to estimate its abundance (Fig. 4B). Like Corynebacterium, the abundance of Propionibacterium also decreased considerably after surgery. We found few changes in comparisons between preoperative clinic visit and postoperative chest swabs.
In agreement with beta diversity analyses of subcutaneous swabs, taxonomic analysis demonstrated a remarkably increased abundance of colonic anaerobes Lachnospiraceae and Bacteroidaceae in the extraction incision at closure but not at opening (Fig. 4C and eTable 2). We evaluated whether patterns of microbial diversity were associated with speci c clinical variables. However, we did not identify signi cant differences in the skin microbiota based upon gender, age, antibiotic exposure, mode of preoperative bowel regimen, presence of cancer, or the side (right or left) of the colorectal resection (data not shown).

Postoperative infection
Postoperatively, only one patient developed a wound infection. No other patients required their wounds to be opened for any reason. The study subject with an SSI (Patient 8) grew Enterococcus faecalis on a swab of wound drainage sent to the microbiology laboratory on POD 11. Interestingly, Enterococcus was not present in skin swabs at the preoperative clinic visit, before or after skin preparation on the day of surgery (eFigure 3), subcutaneous fat swabs of the extraction incision at the beginning or end of surgery, or incision swabs on the rst two days after surgery. However, by POD 4, Enterococcus was seen at high abundance in ileostomy e uent and also at low abundance in abdominal skin swabs. The abundance on the abdominal skin swabs then increased progressively and ultimately Enterococcus was identi ed in drainage from the wound both by culture and by 16S rRNA gene sequencing.
The postoperative appearance of Enterococcus on the wound also corresponded with a decrease in the abundance of skin commensals Corynebacterium, Staphylococcus, and Propionibacterium. These common skin commensals were each seen preoperatively, then disappeared in the postoperative period before partially reappearing on abdominal skin swabs later in the hospital stay and at the postoperative clinic visit. Enterococcus was still detectable on the skin at the postoperative visit when the wound infection was resolved, but at that time it was no longer detectable in the ileostomy e uent.

Discussion
Most abundant skin organisms are likely unknown to most clinicians and operating room staff. Indeed, preoperative skin preparation and antibiotic administration should be expected to obliterate the skin microbiota but the clinical consequences of these interventions are understudied. We postulate that improved knowledge of the skin microbiome and the associated immune response, in parallel with improved knowledge of the gut microbiome and gastrointestinal surgery, could provide actionable ndings about SSI pathogenesis.
The purpose of this study was to characterize changes in the cutaneous microbiome in patients undergoing major elective colorectal procedures for both benign and malignant disease. Swabs of the chest wall served as internal controls for postoperative swabs of the incision. To our knowledge, this is the largest study to date of the microbiota of surgical incisions, and the samples collected were paired with intestinal contents, stool samples, and intraoperative swabs of the subcutaneous fat in the wounds. Overall, as expected, we observed a skin microbiota marked by an abundance of Corynebacteriaceae, Propionibacterium, and Staphylococcus.
The key nding was a transient postoperative loss of the skin commensals Corynebacterium and Propionibacterium and a transient colonization by potential pathogens as well as anaerobes from the gut. This appears to be unique to the surgical site, as the observed chest microbiota was far more stable. During the period of microbial instability after surgery, skin swabs contained unusually high levels of Enterobacteriaceae and intestinal anaerobes such as Lachnospiraceae and Bacteroidaceae. Interestingly, the swabs of the subcutaneous fat from the opening incisions did not demonstrate complex microbial communities, whereas swabs at wound closure contained an intestine-like microbiota.
Because the SSI rate in this cohort was low, it was not possible to formally test the hypothesis that pathogens responsible for SSIs would be present on postoperative swabs prior to onset of infection. This lack of SSIs may re ect the large number of minimally invasive cases, which generally enjoy lower complication rates 34 . In the single patient with an Enterococcus wound infection, we observed a dramatic increase in abundance of Enterococcus in ileostomy e uent and concomitantly on swabs of the abdominal skin. This coincided almost perfectly with a loss of the commensals. The fact that the pathogen responsible for wound infection was not seen in the luminal contents or subcutaneous fat swabs supports the concept that wound infection may not necessarily re ect intraoperative contamination. It may be the case that Enterococcus bloomed in the setting of antibiotic exposure 35 , and subsequently spread to cause infection.
Although we did not observe a signal linking skin microbiome and SSI, the available literature would suggest that our ndings are clinically relevant. The loss of skin commensals near the wound likely creates an opportunity for growth of pathogens and other organisms not typically found on the skin, as seen also in models of skin wounds 27,31 . The observed dysbiosis, like dysbiosis at other body locations, brings about two related problems. First, affected individuals acutely suffer from the loss of commensals that may be important for health and immune function. Second, with commensals absent, potential pathogens may grow in a compensatory fashion in a manner akin to loss of colonization resistance, which has been better studied in the intestine 36,37 . Clearly, since most subjects in this study did not suffer complications despite displaying postoperative skin dysbiosis, additional variables must be responsible for the pathogenesis of SSIs. Identifying these variables and integrating them with knowledge of postoperative dysbiosis could be useful in preventing SSI.
If larger studies in the future allow for identi cation of colonization patterns associated with onset of infection, then there may be actionable opportunities for improved perioperative care. First, it may be possible to identify surgical incisions at high risk of infection, and then monitor the microbiome of these incisions over time. Early identi cation of postoperative skin dysbiosis may then allow for preventive measures to restore a normal skin microbiota. For example, topical probiotic therapy has now been embraced as a method to optimize the skin microbiome 38,39 . Other available interventions to modulate the skin microbiota include topical application of honey-based ointments or the use of enteral probiotics.
Proof-of-principle for perioperative manipulation of the microbiome was provided with the demonstration that administration of mupirocin for patients found to be colonized preoperatively with methicillinresistant S. aureus improves outcomes 40 .
Finally, this study raises questions about the unfocused approach to obliterate the perioperative skin microbiome in the hopes of preventing SSI. Minimizing intraoperative contamination will always be important, but could be integrated with microbiome science. Ultimately, when a wound becomes infected and a causative bacterial species is isolated from culture, one might attempt to determine the origin of   Table S1.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. Onlinetext.docx