Role of the microbiota in ileitis of a mouse model of inflammatory bowel disease—Glutathione peroxide isoenzymes 1 and 2‐double knockout mice on a C57BL background

Abstract C57Bl6 (B6) mice devoid of glutathione peroxidases 1 and 2 (Gpx1/2‐DKO) develop ileitis after weaning. We previously showed germ‐free Gpx1/2‐DKO mice of mixed B6.129 background did not develop ileocolitis. Here, we examine the composition of the ileitis provoking microbiota in B6 Gpx1/2‐DKO mice. DNA was isolated from the ileum fecal stream and subjected to high‐throughput sequencing of the V3 and V4 regions of the 16S rRNA gene to determine the abundance of operational taxonomic units (OTUs). We analyzed the role of bacteria by comparing the microbiomes of the DKO and pathology‐free non‐DKO mice. Mice were treated with metronidazole, streptomycin, and vancomycin to alter pathology and correlate the OTU abundances with pathology levels. Principal component analysis based on Jaccard distance of abundance showed 3 distinct outcomes relative to the source Gpx1/2‐DKO microbiome. Association analyses of pathology and abundance of OTUs served to rule out 7–11 of 24 OTUs for involvement in the ileitis. Collections of OTUs were identified that appeared to be linked to ileitis in this animal model and would be classified as commensals. In Gpx1/2‐DKO mice, host oxidant generation from NOX1 and DUOX2 in response to commensals may compromise the ileum epithelial barrier, a role generally ascribed to oxidants generated from mitochondria, NOX2 and endoplasmic reticulum stress in response to presumptive pathogens in IBD. Elevated oxidant levels may contribute to epithelial cell shedding, which is strongly associated with progress toward inflammation in Gpx1/2‐DKO mice and predictive of relapse in IBD by allowing leakage of microbial components into the submucosa.

microbiota differ and will probably continue to differ model to model (Gkouskou, Deligianni, Tsatsanis, & Eliopoulos, 2014). Suspected pathogens and ordinary commensals might be acting as conditional pathogens depending on the genetic and environmental conditions contributing to each rodent model. Since the genetics of IBD is complex, as are the environments of sufferers, this might be the expectation for humans as well (Liu et al., 2015).
However, typical oxidant-producing Enterococcus faecalis caused more DNA damage in colon epithelial cells than a mutant strain with attenuated production after oral gavage of a bolus of the bacteria into antibiotic-pretreated rats (Huycke, Abrams, & Moore, 2002). Oxidant generation in IBD is usually ascribed to ER stress in Paneth cells, NOX2 in macrophages and mitochondria, although DUOX2 mRNA levels are elevated up to 20 times in active Crohn's ileitis and ulcerative colitis (Hamm et al., 2010;Haberman et al., 2014;Li et al., 2010;Malhotra et al., 2008;MacFie et al., 2014;Yanai et al., 2015).
Studies were performed that link pathology in DKO ilea to DUOX2, NOX1, and microbiota. A germ-free B6.129 Gpx1/2-DKO colony did not exhibit pathology (Chu et al., 2004;Esworthy, Binder, Doroshow, & Chu, 2003). Ileocolitis in B6 DKO mice is driven by DUOX2 and NOX1 (Chu et al., 2017;Esworthy et al., 2014). This was shown in triple knockout lines (TKO), one in which the Nox1 gene was knocked out and the second in which the Duoxa locus (Duoxa1 and Duoxa2 maturation subunit genes) was modified to eliminate cell surface expression of DUOX1(barely expressed in the intestine) and DUOX2 and the ability to generate H 2 O 2 (Grasberger et al., 2012). Lack of functional DUOX2 (Gpx1−/−Gpx2−/−Duoxa−/− TKO) eliminated crypt anoikis (exfoliation) and crypt abscesses. Excessive apoptosis remained resulting in partial loss of Paneth cells and crypts and slightly above background levels of macrophage numbers and monocyte infiltration (Chu et al., 2017) Our hypothesis for linking microbiota to ileitis was based on the idea that Lactobacillus-induced NOX1 oxidant generation produces damage on its own, which DUOX2 augments. We posited that antibiotic-induced differences in Lactobacillus abundance in the ilea of B6 Gpx1/2-DKO mice would be reflected in levels of pathology and E. coli abundance might not show a positive association with pathology since E. coli did not elicit NOX1 oxidant generation. To examine this, antibiotics were orally administered at the onset of ileum pathology to alter the evolution of the microbiota and possibly influence pathology levels. Three antibiotics were selected to be separately administered and produce distinct outcomes, impacting many operational taxonomic units (OTU; largely genus level). The primary analysis would be based on the association of abundance and pathology marker levels among the antibiotic-treated and control sets. We were uncertain about the similarity of microbiotas between Gpx1/2-DKO and non-DKO mice (Gpx1+/−Gpx2−/− and Gpx1−/−Gpx2+/−). Since Gpx1/2-DKO and non-DKO mice were sibs or half-sibs (cage mates), the differences might have been minor. Pathology in Gpx1/2-DKO mice might alter the abundance of some OTUs from that of non-DKO mice. This perspective was used to assess the antibiotic effects.
Another viewpoint is that ileitis provoking OTUs might have some growth advantage in the pathological milieu over commensals as defined by abundances in wild-type or other relatively normal mice (non-DKO cage mates) and suppress the growth of more benign commensals. Thus, comparisons of OTUs with increased abundance in mice with ileitis compared to non-DKO controls might indicate provoking candidates (Peloquin & Nguyen, 2013;Pircalabioru et al., 2016;Sartor & Wu, 2017). We also analyzed the results from this latter viewpoint.

| Antibiotics
Metronidazole, streptomycin, and vancomycin were administered in drinking water; metronidazole concentration was 750 mg/L, streptomycin-450 mg/L, and vancomycin-50 mg/L (Ferreira et al., 2011;Sekirov et al., 2008;Wlodarska et al., 2011). Mice shun the metallic taste of metronidazole (Chu, Esworthy, Doroshow, & Shen, 2016). To overcome this, the water used to prepare antibiotics and the water for the control mice contained 0.4% of the sweetener, Splenda®. Mice were allowed free consumption from 22-35 days of age. The treated water was refreshed on the 7th day. At 35 days of age, the mice were euthanized by CO 2 inhalation. Eighteen to 20 mice were in each group. Mice from multiple litters were used for each set (Splenda-4 litters; Gpx1/2-DKO and non-DKO; metronidazole-5 litters; vancomycin-5 litters; streptomycin-5 litters). The numbers of mice used were adequate to statistically distinguish sets of mice with intermediate pathology scores and marker values from DKO and non-DKO or wild-type mice in our previous studies (Chu et al., 2017;Esworthy et al., 2014). Eight Gpx1/2-DKO mice with overall marker scores representing the average or median and 6 non-DKO mice were selected for microbiome analysis. The number was based on examining comparable studies in B6 mice where the group sizes were generally 5-9 mice (Garidou et al., 2015;Gu et al., 2013;Jakobsson et al., 2015;Kar et al., 2017;Robertson et al., 2013;Tourret et al., 2017;Walk, Blum, Ewing, Weinstock, & Young, 2010).

| Tissue sampling
The length of the small intestine (pylorus to ileocecal junction) was measured and the fecal stream of the ilea, 1-5 cm above the cecum, was expressed into sterile tubes and immediately frozen (−80°C).
The length of the colon was measured. Ileum sections were fixed for histology (10% phosphate-buffered formalin). Samples were processed for sectioning and stained with hematoxylin and eosin (H&E).

| Histopathology
H&E histopathology was evaluated by enumerating crypt apoptosis, crypt exfoliation (anoikis), depletion of crypts and Paneth cells, and crypt abscesses, details in Figure 1 (Chu et al., 2017). The individual scoring was blinded to the identity of the slides.
2.5 | Isolation of DNA from ileum fecal stream, processing, and analysis of microbiome DNA was isolated from the ileum fecal stream following the procedure described (Elson, Cong, Qi, Hershberg, & Targan, 2006;Esworthy et al., 2010). DNA was diluted to 0.1 μg/μl in TE (10 mM Up to 15 ng of PCR products were carried forward to library preparation using second-round PCR. The Illumina primer PCR PE1.0 and index primers were used to allow the multiplexing of samples.

| Statistical analysis
GraphPad Prism 7.01 was used for statistical analysis of pathology.
Each data set was a check for a parametric distribution. Parametric

| Moderately different microbiome in the non-DKO cage mates
Pathology scores and small intestine lengths in the Splenda control Gpx1/2-DKO and non-DKO mice were consistent with prior observations (Figures 1 and 2). Splenda control Gpx1/2-DKO ilea harbored microbiomes with some differences from non-DKO cage mates ( Figure 3). In DKO mice, significant decreases in abundance were found for Barnesiella, Desulfovibrio, and Porphyromonadaceae A core microbiota (less than 2-fold difference; p > 0.09) was identi-

| Antibiotic treatment effect on ileitis
A filter for assessing OTUs would be abundances in responsive and unresponsive Gpx1/2-DKO mice. However, the penetrance of ileum pathology is 95%+; pathology scores were no less than 3 among 53 Gpx1/2-DKO mice analyzed for this study and in prior work ( Figure 1A) (Chu et al., 2019;Esworthy et al., 2014). Mice with scores of 3 exhibit elevated crypt apoptosis with reduction of Paneth cell incidence by half and/or crypt density less than the normal range. Non-DKO mice had scores from 0 to 1 ( Figure 1A).
Antibiotic treatment expanded the range of pathology scores in the DKO mice from 0.5 to 8 ( Figure 1A). Streptomycin and vancomycin had strong effects on the ileum pathology, while metronidazole had a moderate impact. Even though streptomycin treatment produces almost complete turnover of the microbiome, some pathology remained. Mice representing the extremes from DKO Splenda controls and antibiotic-treated mice were identified and the OTU abundances compared. Six mice were deemed cured, having pathology scores of 0.5-1 (mean score 0.83 ± 0.26), comparable to non-DKO mice.
Ten mice were selected to represent the sick extreme, having pathology scores of 5 and above (mean score 5.9 ± 0.88). Seventeen OTUs were identified as possibly contributing to pathology by finding values less than 1 for-abundance in cured animals÷abundance in sick animals ( Figure 4c). Second, we evaluated statistical sig- Inspection of the correlation graphs reveals full variation in pathology markers values (i.e., pathology scores of 0.5-7) occur in mice with nearly zero abundance of Helicobacter, Pasteurellaceae, Sporacetigenium, and Turicibacter ( Figure 5 and Figures S2-S8).
The low abundance of Lactobacillus and Clostridaceae-1 is more consistently associated with lower pathology for markers, while Streptococcus shows a mixed pattern. The Lactobacillus correlation with pathology score had a modest R-squared value of 0.35.

| DISCUSS ION
We examined nine publications to determine whether the micro- The reported microbiomes differed extensively, in some cases between replicate experiments within studies. One paper did not test

F I G U R E 4 Results of 3 analyses for candidate provocative and beneficial OTUs. Principal component analysis of OTUs showing
clustering of mice from Splenda DKO, non-DKO, and antibiotic-treated groups. This is based on Jaccard distances using abundance data. Panel a shows PC1 and PC2. Colored arrows indicate diverse microbiome groupings created by antibiotics relative to the Splenda DKO control; dark blue-vancomycin; red-streptomycin. Panel b has results for PC2 and PC3 and shows that non-DKO mice remained grouped with the Splenda DKO control while the metronidazole group is slightly distanced (light blue arrow). Panel c represents the assessment provocative of OTUs (abundance in low pathology mice ÷sick mice <1). Statistical significance in pairwise t tests was found for an abundance of 3 OTUs (* and **). None passed adjustment for multiple samples. The best 2 OTU candidates are indicated by **. Panel d groups candidate OTUs based on contraction in DKO vs non-DKO (↓) or expansion (↑) (statistically significant differences indicated by bold type), moderate to strong correlation with pathology markers (DKO purple oval stronger candidates in white letters; bold type for best candidates) or both (overlapping pink and purple ovals). The best candidates based on correlation are in bold white. The best consensus candidates are in whites letters and indicated by ↑ in the overlapping region. 3 OTUs found in Table 1   Note: Mean abundances for the non-DKO mice and antibiotic-treated mice (metron-metronidazole; vanco-vancomycin; and strep-streptomycin) are listed in adjacent columns. The last set of values (correlations) represent the average of the correlation coefficients for the pathology parameters listed in (b) for each mouse's pathology parameters versus the abundance of the OTUs for each mouse; all coefficients converted to positive values when larger OTU abundance is associated with greater pathology and converted to all negative values when larger OTU abundance is associated with less pathology. In (b), the median pathology score, the median level of crypt exfoliation, the median of the fraction of crypts with Paneth cells, the median value for crypt abscesses, and mean value for crypt apoptosis are listed.
A second question is whether the antibiotic treatments produced predictable effects on the microbiome. Our goal in using these 3 antibiotics was to achieve 3 distinct outcomes, which appeared likely based on results from prior studies using wild-type B6 mice (Ferreira et al., 2011;Ju et al., 2017;Sekirov et al., 2008;Wlodarska et al., 2011). Once the composition of control DKO/ non-DKO microbiomes was determined, we were in a position to retrospectively examine whether the impact of the antibiotics F I G U R E 5 Correlation of OTU abundances with pathology scores using data from antibiotics and Splenda DKO sets. L plus C represents adding the abundances of Clostridiaceae-1 and Lactobacillus and performing the correlation versus pathology score. This result was generated for comparison with the Clostridiaceae-1 and Lactobacillus correlations. The remaining panels show some of the top candidates based on correlations across all markers (see Table 1 correlations pathology column) would be expected. Metronidazole is supposed to target anaerobes. Thus, Clostridiaceae-1 abundance should be down. In B6 colon, treatment with metronidazole at the dose used here for

days depleted Clostridium coccoides (cluster XIVa) and suppressed
Bacteroidales allowing for increased abundance of Lactobacilli, Bifidobacteriaceae, and Enteroccaceae (Ju et al., 2017;Wlodarska et al., 2011). This is similar to what we found, although the impact on Clostridia was not nearly as strong in our samples. Vancomycin depleted Gram-positive Clostridiaceae-1, as expected. It did not affect Gram-positive Lactobacillus, which could be anticipated due to similar findings in B6 colon at this concentration (Sekirov et al., 2008). The increase in abundance of the Gram-negative Escherichia/Shigella OTU might have been expected from the loss of Clostridiaceae-1 or other OTUs, although the overgrowth far exceeded that found in B6 colon at this concentration (Sekirov et al., 2008). The effect of streptomycin in the reference papers was to deplete C. coccoides and suppress Lactobacilli while leaving Bacteroidales unaffected. This is what we found in our samples.
The only possible anomalies were the partial impact of metronidazole on Clostridiaceae-1 abundance and the level of overgrowth by the Escherichia/Shigella OTU with a low concentration of vancomycin. The latter finding might be related to the ileum having higher oxygen content than colon favoring more overgrowth by facultative anaerobes like the Escherichia/Shigella OTU (Sommer & Backhed, 2016).
The Gpx1/2-DKO and non-DKO comparison can be used to examine the issue of whether pathology alters the microbiome. Ileum pathology was mild in the newly weaned B6 Gpx1/2-DKO mice, confined to sporadic crypt exfoliation and sporadic crypt apoptosis that, in the majority of cases, was not over normal levels (Chu et al., 2017(Chu et al., , 2019. Nox1 and TNF-α mRNA levels were not elevated at this time. With that context, sibship of Gpx1/2-DKO and non-DKO mice and fostering by non-DKO dams, we would expect they shared very similar microbiotas at weaning. The 2 sets of microbiomes are comparable when pathology reaches its high point (35 days and thereafter), although the non-DKO microbiome is only marginally less different from the Splenda Gpx1/2-DKO set than the metronidazole set based on PCA (non-DKO separates in PC2; metronidazole in PC3). Four

OTUs show up as having significantly altered abundances between
Gpx1/2-DKO and non-DKO while 14 are altered between Gpx1/2-DKO mice on Splenda and metronidazole-treated mice . The possible impact of pathology in Gpx1/2-DKO mice on the microbiota can be used as a reference for evaluating the effects of vancomycin and streptomycin.
Our experiment was designed so that antibiotics would have an opportunity to modify the microbiota before pathology significantly altered the architecture of the ileum with ensuing immune cell infiltration. Based on prior studies, there was a window from 22 to 26 days of age for the antibiotics to operate. Without antibiotics, pathology becomes more aggressive on and after 27 days of age (Chu et al., 2017(Chu et al., , 2019. In our view, the early modification of the evolving microbiota would then influence subsequent pathology. The comparison of Gpx1/2-DKO and non-DKO microbiomes shows how pathology impacts the microbiota. The general effect of the antibiotics, vancomycin, and streptomycin was to significantly reduce pathology. Both produced a huge difference in the resulting microbiomes from the non-DKO mice (Figure 4a,b). The antibiotic effect appears to dominate over any differences produced by differences in pathology levels. We seem to be largely observing the impact of microbiota on pathology.
The stratification of mice from the Splenda control and antibiotic Gpx1/2-DKO sets into cured and sick divides the OTUs into provisional provocative and beneficial or opportunistic categories with statistical analysis for significance in either category. Bacteroides and Ureaplasma from the beneficial set pass statistical testing for differences in abundance in a pairwise t test and fail in after adjustment for multiple comparisons. The correlations with pathology markers for the beneficial set are generally weaker than in the provocative set (Table 1a). Lactobacillus, Clostridiaceae-1, and Streptococcus were identified as candidate provocative OTUs in this analysis in pairwise testing. All fail to pass after adjustments for multiple samples with Lactobacillus and Clostridiaceae-1 remaining as provisional candidates. Streptococcus is regarded as a probiotic (Koretz, 2018;Shiina et al., 2015). The literature on Clostridiaceae-1 (C. perfringens excluded) is generally unfavorable to candidacy as a pathogen (Kanai, Mikami, & Hayashi, 2015;Lawson, Citron, Tyrrell, & Finegold, 2016;Peloquin & Nguyen, 2013;Peyrin-Biroulet et al., 2012). While some species of Lactobacillus are used as probiotics, members of the genera exhibit properties that fit into our notion of a pathogen for Gpx1/2-DKO mice by eliciting oxidant generation and/or generating oxidants (Jones et al., 2013;Knaus et al., 2017). Escherichia coli did not elicit ileum oxidant generation in earlier studies and this analysis, the associated OTU appeared to be eliminated as a provocative candidate. Some species of Streptococcus have a pyruvate oxidase activity and under aerobic conditions can generate H 2 O 2 (Redanz et al., 2018). There is no information on whether Streptococcus elicits NOX1 oxidant generation by the host. The abundance is so low that its possible impact by these mechanisms would be minor compared to Lactobacillus.
Lactobacillus does not show a singular dominance in the correlation analysis with pathology markers and no other compelling candidates emerge. The reason for this may be that we are looking at genera and families rather than species. Alternatively, a superior correlation is found by using the combined abundance of Lactobacillus and Clostridiaceae-1 and pathology markers. This improves the average of the correlation coefficients to 0.62 and more consistently associates high pathology maker values with high abundance. We examined this to investigate the idea that the response of Gpx1/2-DKO mice may be to components of several pathology provoking OTUs. The residual pathology in the streptomycin set indicates at least a low-level reaction to a different set of bacteria than found in the Gpx1/2-DKO Splenda control (pathology scores of 3, 2, 2 and 2 for 4 of the last 5 mice in the streptomycin set, Figure 3e).
A second notion is that the displacement of Clostridiaceae-1 by the Escherichia/Shigella OTU in the metronidazole and vancomycin sets might produce direct competition with Lactobacillus for microaerobic niches (Espey, 2013;Sommer & Backhed, 2016).
Studies suggest that this would normally be an unlikely circumstance made possible by sustained antibiotic treatment. H 2 O 2 and lactic acid production by Lactobacillus can provide an edge over E. coli for the population of mucosal sites (Gupta et al., 1998). The impact of high abundances of E. coli could be a reduction in oxidants both from bacteria and host (Jones et al., 2013). Adjusting for the presence of Escherichia/Shigella OTU as a factor diluting Lactobacillus improves the average of the correlation coefficients to 0.58 (EF supporting file Lactobacillus adjusted for E. coli dilution and Figure S10) . Catalase, present in many strains of E. coli, can be a significant sink for H 2 O 2 (Rodriguez, Peiroten, Landete, Medina, & Arques, 2015). This would be another mechanism by which the Escherichia/Shigella OTU could counter Lactobacillus and host oxidant generation . While 2 attempts to factor in the interaction between OTUs produce a better fit in correlations with pathology markers, neither yield a spectacularly enhanced outcome and employ opposed models (pathology enhancing interaction vs. counter pathological interaction) to yield similar improvements.  (Cominelli, Arseneau, Rodriguez-Palacios, & Pizarro, 2017). In common with most of these models is distress in or loss of Paneth cells due to aberrant autophagy and ER stress (Adolph et al., 2013). We found that the population of apoptosed and exfoliated cells while including Paneth cells was predominantly other cell types based on the general absence of lysozyme (Chu et al., 2017). Some of these ileitis models (Samp/YitFc, Caspase-8 ∆iec, and FADD ∆iec ) exhibit sterile inflammation which can be augmented by the presence of microbes. In models with sterile inflammation, the Paneth cells may undergo necrosis or necroptosis, which are inherently inflammatory (Stolzer et al., 2020). Inflammatory signals may also derive from danger-associated molecular patterns (DAMPs) generated from defective autophagy and unfolded proteins (unfolded protein response; UPR) in Paneth cells with sub-lethal distress (Cadwell et al., 2008). Findings from genome-wide association studies link genes in the UPR and autophagy pathways to Crohn's ileitis and Paneth cell defects (Cadwell, Stappenbeck, & Virgin, 2009). We see an outright loss of Paneth cells in Gpx1/2-DKO mice, which has been proposed by others to be a possible consequence of ER stress (Kaser et al., 2008). ER stress increases the production of H 2 O 2 as a by-product of augmented protein folding activity (Delaunay-Moisan & Appenzeller-Herzog, 2015). Indications of ER stress were detected in tissues of B6 and 129 Gpx1/2-DKO mice at very high levels of pathology and not at moderate levels of pathology (Esworthy et al., 2011;Gao et al., 2010). Since pathology in B6 Gpx1/2-DKO mice shows a dependence on NOX1, we doubt that ER stress generates the ileitis, although when pathology is underway ER stress may contribute to the Paneth cell loss.
Pathology in the Xbp-1 ∆iec and TNF ∆ARE models is dependent on microbiota (Adolph et al., 2013;Schaubeck et al, 2016). In the Xbp-1 ∆iec model (ER stress-driven ileitis; B6; 129; FVB mixed background), the demonstration was limited to evaluation of germ-free mice. As in the Xbp-1 ∆iec model, we found nearly complete loss of Paneth cells and only partial loss of goblet cells (Chu et al., 2017;Kaser et al., 2008). In the TNF ∆ARE model (TNFα over-expression driven ileitis; We are also comparing cecum to the ileum and the abundance of unknown Clostriadales was in the range of 1%-4% of total reads as opposed to 25% for Clostridiaceae in our samples. TNF ∆ARE mice with low ileitis scores had microbiota compositions resembling wild-type mice; the Splenda control Gpx1/2-DKO mice closely resembled the non-DKO controls. Loss of up to one-half of the Paneth cells in B6 Gpx1/2-DKO mice, largely through apoptosis, was associated with very mild, sporadic inflammation in Gpx1/2-Duoxa-TKO mice at 35 days at age and Gpx1/2-DKO mice at 29 days at age (Chu et al., 2017). Macrophage numbers were elevated with few infiltrating monocytes and rare crypt abscesses. Further loss of Paneth cells was driven by anoikis/exfoliation in Gpx1/2-DKO mice. This appeared at a later time than apoptosis and never attained high levels in Gpx1/2-Duoxa-TKO mice. While Paneth cell loss B6 Gpx1/2-DKO mice is nearly complete by 32 days of age, we viewed this loss as a marker for the underlying processes of apoptosis and exfoliation and not a singular cause of inflammation. Lack of Paneth cells in B6 Gpx1/2-DKO mice was not associated with a marked dysbiosis as gaged by the non-DKO microbiome. However, increased susceptibility to the core microbiome could be due to the failure of anti-bacterial defenses normally provided by Paneth cells. Exfoliation of any epithelial cell type can be inflammatory due to leakage of microbial components into the submucosa during this protracted process (Williams et al, 2015). After the treatment of IBD, high levels of exfoliation are prognostic of the potential for relapse (Kiesslich et al, 2012;Turcotte et al, 2012). In this way, we can link a process observed in Gpx1/2-DKO mice to relapse in IBD and dependence on the microbiota.

CO N FLI C T O F I NTE R E S T S
None declared. The content is solely the responsibility of the authors and does not necessarily represent the official views of NIH.

E TH I C S S TATEM ENT
Studies were approved by the City of Hope IACUC (protocol #92008), which conforms to NIH and the American Association for Accreditation of Laboratory Animal Care (AAALAC) rules and standards.

DATA AVA I L A B I L I T Y S TAT E M E N T
Supporting data sets generated and/or analyzed during the current study are available at figshare :