Rapid Growth of Uropathogenic Escherichia coli during Human Urinary Tract Infection

ABSTRACT Uropathogenic Escherichia coli (UPEC) strains cause most uncomplicated urinary tract infections (UTIs). These strains are a subgroup of extraintestinal pathogenic E. coli (ExPEC) strains that infect extraintestinal sites, including urinary tract, meninges, bloodstream, lungs, and surgical sites. Here, we hypothesize that UPEC isolates adapt to and grow more rapidly within the urinary tract than other E. coli isolates and survive in that niche. To date, there has not been a reliable method available to measure their growth rate in vivo. Here we used two methods: segregation of nonreplicating plasmid pGTR902, and peak-to-trough ratio (PTR), a sequencing-based method that enumerates bacterial chromosomal replication forks present during cell division. In the murine model of UTI, UPEC strain growth was robust in vivo, matching or exceeding in vitro growth rates and only slowing after reaching high CFU counts at 24 and 30 h postinoculation (hpi). In contrast, asymptomatic bacteriuria (ABU) strains tended to maintain high growth rates in vivo at 6, 24, and 30 hpi, and population densities did not increase, suggesting that host responses or elimination limited population growth. Fecal strains displayed moderate growth rates at 6 hpi but did not survive to later times. By PTR, E. coli in urine of human patients with UTIs displayed extraordinarily rapid growth during active infection, with a mean doubling time of 22.4 min. Thus, in addition to traditional virulence determinants, including adhesins, toxins, iron acquisition, and motility, very high growth rates in vivo and resistance to the innate immune response appear to be critical phenotypes of UPEC strains.

T he urinary tract is the most common site of bacterial infection in humans (1). It has been estimated that at least 40 to 50% of women will experience a minimum of one symptomatic urinary tract infection (UTI) during their lifetime, with roughly 27 to 48% 7.6 ϫ 10 5 to 2.0 ϫ 10 6 CFU/g bladder. Thus, the fecal, UPEC, and ABU strains examined in this study can all survive in the urinary tract for at least 6 hpi.
Since it has been demonstrated that UPEC strains carry more virulence determinants than fecal strains (19,20) (see Fig. S1 in the supplemental material) and cause persistent infections, whereas fecal strains do not, we transurethrally inoculated mice with either fecal or UPEC strains with inocula of 10 8 , 10 9 , or 10 10 CFU. After 48 h, bladder bacterial burden was enumerated. All three inoculum concentrations resulted in bladder colonization by the UPEC isolates up to~10 8 CFU/g bladder; however, fecal strains survived poorly, with median values below the limit of detection of 100 CFU/g (Fig. 1B). Kidney colonization (Fig. 1C) was reflective of bladder colonization. These results indicated that fecal strains are unable to persist out to 48 hpi in the murine urinary tract following transurethral challenge, regardless of the inoculating dose. While resistance to the innate immune response, particularly neutrophil infiltration, may explain UPEC's ability to colonize the urinary tract more successfully than fecal strains (21,22), it is also possible that persistence may relate to the bacterial growth rate within urine and the host environment. For this reason, we sought to measure the bacterial growth rate within the urinary tract.
In vivo doubling time for ABU strains is statistically longer than those for UPEC and fecal strains via plasmid partitioning. We utilized pGTR902, which replicates in the presence of L-arabinose (23), for estimation of bacterial growth rate in a collection of UPEC, fecal, and ABU isolates in the murine model of ascending UTI. To assess the growth rate, pGTR902 was introduced into UPEC isolate CFT073 by electroporation and cultured in LB medium with and without 1% L-arabinose. The total number of bacteria was enumerated by plating on LB agar, and the number harboring pGTR902 was determined by plating on LB agar containing kanamycin and L-arabinose. The presence of pGTR902 did not affect growth of CFT073 in LB broth with or without L-arabinose, and a decrease in the number of CFT073 cells containing pGTR902 was only observed  (F3, F15, F11, F24, F54,  CFT073, CFT269, CFT189, CFT204, and CFT325) isolates. Bladder and kidneys were harvested at 48 hpi. In panels A, B, and C, symbols represent individual mice and bars represent the median. n ϭ 4 to 60. The limit of detection is 100 CFU/g. As an index of type 1 fimbrial expression, strain CFT073 agglutinated a suspension of yeast (Sacchoromyces cerevisiae) at a bacterial titer of 1:32. Strains ABU83972 and EFC7 failed to agglutinate yeast. Colonization and virulence gene data for each strain may be found in Fig. S1.
Measuring E. coli Growth Rate In Vivo ® in LB medium lacking L-arabinose (see Fig. S2 in the supplemental material). Comparisons between strains require precise determination of pGTR902 copy number in each strain. We therefore introduced pGTR902 into two additional UPEC isolates (UTI89 and 536), two fecal isolates (EFC7 and EFC2), and two ABU isolates (PUTS37 and ABU83972) and conducted in vitro growth experiments in LB medium without L-arabinose to calculate plasmid copy number based on plasmid segregation (Fig. 2). Plasmid segregation was observed in all seven strains ( Fig. 2A to G), and average copy number was determined from 2 to 8 independent experiments per strain by dividing the number of pGTR902-containing bacteria at stationary phase by the number of pGTR902containing bacteria present in the inoculum (Fig. 2H). Copy numbers differed between E. coli isolates, ranging from approximately 4 plasmids per bacterial cell in ABU83972 to approximately 47 plasmids per bacterial cell in UTI89. Plasmid segregation and copy number for CFT073, EFC7, and ABU83972 cultured in human urine were similar to that attained in LB (compare Fig. S3A to C in the supplemental material to Fig. 2A, E, G), suggesting that pGTR902 may segregate similarly within the urinary tract.
To estimate growth rate in vivo, C57BL/6 mice were transurethrally inoculated with 10 8 CFU/mouse of each E. coli isolate harboring pGTR902. Six hours postinfection was chosen as the ideal time point to harvest bladders for CFU as all seven isolates colonized to similar levels at this time (Table 1; see Fig. S4 in the supplemental material). Growth proportion, number of generations, and in vivo doubling time were estimated for each E. coli strain using total CFU recovered from the bladder of each mouse, the CFU of pGTR902-containing bacteria, and the experimentally determined plasmid copy number (Table 1). The in vitro doubling time of each strain during logarithmic growth in LB medium is shown for comparison. Overall, there were no statistically significant differences in bladder colonization, growth proportion, number of generations, or doubling time between each of the seven isolates. However, the FIG 2 pGTR902 copy number is variable among EXPEC isolates. (A to G) Representative growth curves in LB diluted 1:100 (ABU83972) or 1:1,000 (CFT073, UTI89, 536, EFC2, EFC7, and PUTS37) from overnight cultures grown in LB supplemented with 1% L-arabinose and kanamycin (25 g/ml). Cultures were grown at 37°C with aeration, and CFU per milliliter of pGTR902-containing bacteria and total bacteria (bacteria containing and those not containing pGTR902) were determined at 30-min or 1-h intervals by plating on LB agar containing 1% L-arabinose and kanamycin (25 g/ml) (open symbols) and LB agar containing no antibiotic (closed symbols), respectively. (H) The copy number of pGTR902 in each isolate was calculated using the following equation: CFU/ml of pGTR902-containing bacteria at stationary phase/CFU/ml of pGTR902-containing bacteria in the inoculum. Values are mean Ϯ standard deviation. average in vivo doubling time for the ABU group was statistically longer than the average for UPEC or fecal strains (P Ͻ 0.0001). Indeed, CFT073 and ABU83972 are the fastest-and slowest-growing strains, respectively, of all those tested using the plasmid segregation method. No significant differences in doubling time were observed in vitro. We conclude from this method that UPEC and fecal strains are capable of similarly high rates of growth within the mouse urinary tract, while ABU isolates exhibit longer doubling times in vivo than in vitro.
Growth rate of CFT073 in human urine correlates with PTR. As a second method to validate measurement of in vivo growth rate, we determined PTR. The assay is based on the principal that a rapidly growing bacterium will initiate multiple forks of replication about the origin of replication compared to the terminus of replication to keep pace with the rate of bacterial division. Thus, a faster-growing bacterium will have a higher PTR than a slower-growing bacterium (24). To test the technique, UPEC strain CFT073 was cultured with aeration in media that support different growth rates: M9 salts supplemented with 0.4% glucose, LB medium, and Terrific broth. Bacteria were collected during the mid-exponential phase of growth for each medium. Genomic DNA was isolated and subjected to Illumina sequencing. Sequence reads were aligned to the genomic sequence of CFT073 (Fig. 3A) (25). PTR was calculated using the following equation: % of reads at the origin/% of reads at the terminus. High PTR values were calculated for bacteria growing in rich media (LB medium and Terrific broth) (PTR ϭ 1.62, 40.5-min doubling time, and PTR ϭ 1.68, 40.0-min doubling time, respectively), and a low PTR value was calculated for bacteria growing in minimal medium (M9 salts supplemented with 0.4% glucose) (PTR ϭ 1.27, 54.2-min doubling time) (Fig. 3A). As expected, bacteria growing rapidly in rich media display higher PTRs than bacteria growing slower in minimal medium.
Before determining the PTR of CFT073 in vivo, we first calculated the PTR of CFT073 cultured in pooled human urine. As expected, PTR values were high during the exponential phase and decreased in a stepwise fashion as the culture approached and entered the stationary phase ( Fig. 3B and C). These data were used to construct a standard curve that correlated growth rate as measured by optical density at 600 nm (OD 600 ) with PTR calculated at 3, 4, 5, 6, 7, and 8 h (Fig. 3D). A near perfect correlation b Growth proportion was calculated as follows: (log 10 CFU/g of plasmid-containing bacteria/plasmid copy no.) Ϫ log 10 total CFU/g. of growth rate with PTR was observed (R 2 ϭ 0.98), allowing extrapolation of doubling time from a given PTR value.
E. coli CFT073, ABU83972, and EFC7 have differential growth rates over time in the mouse model of UTI, as measured using PTR. Mice were transurethrally inoculated with 10 8 CFU E. coli CFT073. The urine and bladder samples of each mouse were collected at 6, 24, and 30 hpi. Bacteria were harvested from pooled urine (n ϭ 4) by centrifugation, and genomic DNA was extracted. Bladder samples from individual mice were homogenized and enriched for bacterial cells using differential lysis (see Fig. S5 in the supplemental material), and DNA was isolated. DNA from urine and bladder preparations was subjected to Illumina sequencing. Enrichment of bladder homogenates for bacterial genomic DNA was critical to ensure that the threshold for accurate PTR determination, Ͼ0.2ϫ genome coverage (see Fig. S6 in the supplemental material), was met. Treatment of bladder homogenates with mammalian cell lysis buffer had only a modest effect on bacterial viability compared to phosphate-buffered saline (PBS)treated controls (see Fig. S7 in the supplemental material). PTR values determined for murine bladders (n ϭ 3 to 4) predicted a high growth rate (PTR ϭ 1.78, 36.9 ϩ 3.8-min doubling time) at 6 hpi, slowing at 24 hpi (PTR ϭ 1.21, 160 ϩ 27-min doubling time) and 30 hpi (PTR ϭ 1.22, 167 ϩ 70-min doubling time) (Fig. 4A and C). Similarly, urine PTR values indicated a high growth rate (PTR ϭ 1.75, 34.9-min doubling time) at 6 hpi, which slowed dramatically at 24 hpi (PTR ϭ 1.09; 797-min doubling time) and 30 hpi (PTR ϭ 1.10; 618-min doubling time) (Fig. 4B and C). The increasing fraction of sequence reads homologous to CFT073 at 24 and 30 hpi suggests that the bacteria reach a saturating population size within the nutrient-limited urinary tract, leading to decreased growth (Fig. 4D). This hypothesis is supported by an increase in bacterial load from~10 6 CFU/g bladder at 6 hpi to~10 8 CFU/g bladder at 24 hpi (see Fig. S8 in the supplemental material). An additional observation is the marked increase in the doubling time of the inoculum compared to the doubling time in the bladder at 6 hpi (119 min versus 36.9 min, respectively), indicating rapid adaptation of UPEC strain CFT073 to conditions within the murine host.
PTR values were also calculated for ABU strain ABU83972 and fecal strain EFC7 during murine UTI. However, because the ABU and fecal strain bacterial burden was not sufficient (Ͻ10 4 CFU/g bladder [ Fig. S8]) to obtain the genome coverage level required for the most accurate PTR determination, we can only define trends in growth rate. In contrast to CFT073, ABU83972 tended to maintain a relatively high growth rate in the bladder at 6, 24, and 30 hpi, with doubling times of 37, 31, and 47 min, respectively. Fecal strain EFC7 was found to have a moderately longer doubling time in the urine (72 min) at 6 hpi compared to ABU83972 and CFT073. Similarly, a modest increase in doubling time in the bladder was observed at 24 and 30 hpi (84 and 67 min, respectively) compared to 6 hpi (46 min). These results indicate that neither EFC7 nor ABU83972 reaches saturation in the bladder, yet ABU83972 maintains a high growth rate at 24 and 30 hpi, while EFC7 replicates more slowly at 24 and 30 hpi.
PTR indicates rapid growth of UPEC isolates during human UTI. By standardizing PTR measurement in vitro in human urine and in vivo using the mouse model, it was possible to estimate the growth rate of E. coli strains during uncomplicated UTI in women. Genomic DNA was extracted from urine that had been previously collected PTR values ranged from 1.78 to 2.55, averaging 2.23 ϩ 0.28 (Fig. 5A), which corresponds to very rapid doubling time estimates ranging from 16.6 to 34.4 min, averaging 22.4 ϩ 6.4 min (Fig. 5B). While it did not escape our notice that these rates approached maximal growth rates for any E. coli strain under optimal in vitro culture conditions, the findings nevertheless indicate that E. coli strains are growing at surprisingly high rates during human infection of the urinary tract and faster than that for UPEC strain E. coli CFT073 during in vitro growth in human urine (36.0 Ϯ 2.6 min) or within the mouse urinary tract (36.9 Ϯ 3.8 min at 6 hpi). Isolates from women with a primary UTI (stippled bars) tended to have a shorter doubling time than isolates from women who suffered from recurrent UTI (plain bars) (Fig. 5B). However, this difference was not statistically significant (P ϭ 0.19).

DISCUSSION
Uropathogenic E. coli cells divide rapidly in both the murine and human urinary tracts, as measured by both a plasmid segregation method (mice only) and PTR (both mice and humans). At early time points postinoculation, ABU strains may be reliant on growth rate alone to successfully colonize as they often lack adhesins. Fecal strains, in general, do not carry a significant number of virulence determinants (19,20) and did not colonize well at time points beyond 6 hpi. UPEC strains are relatively resistant to killing by the innate immune response at 48 h, as indicated by high bacterial loads recovered from the bladder over the entire time course of the infection studies. For the human studies, while we do not know precisely when E. coli entered the bladder, and thus the "hours postinoculation" are unknown, growth is uniformly rapid. These results suggest that UPEC strains are metabolically suited to grow rapidly in the urine within the bladder, even in the presence of a robust innate immune response. Indeed, damage to the host during infection, inflicted both by the bacterium and the host inflammatory response, likely releases additional nutrients to fuel this rapid growth.
We employed two techniques to measure growth rate in vivo. The first method followed plasmid segregation of a nonreplicating plasmid, pGTR902, developed for estimation of the growth rate of Vibrio vulnificus in skin lesions (23). One advantage of this technique is the ease of use in all E. coli strains tested. E. coli isolates transformed with plasmid pGTR902 were cultured in vitro in the presence of arabinose to drive replication of the plasmid and then washed and transurethrally inoculated into the murine urinary tract in which arabinose is absent. Rapid results were obtained within 24 h by plating on agar with and without L-arabinose and kanamycin to allow for differential enumeration of plasmid-containing bacteria. Limitations of the plasmid segregation method include the fact that it cannot be used in naturally occurring infections as the method requires experimental parameters to be defined. In addition, there is potential skewing of growth rate estimation if the plasmid does not properly segregate in vivo or if plasmid-containing bacteria are eliminated during infection.
The second method for determination of growth rate in vivo measured the peakto-trough ratios for genomic DNA from E. coli in vitro or in vivo. This is a powerful tool for estimation of growth rate in an open system like the urinary tract and will not be skewed by loss of bacteria due to urination or ascension to the kidneys. It allows for estimation of bacterial growth rate in naturally occurring infections, not just experimental models. The limitations of the technique are that it is expensive and timeconsuming compared to CFU determination. It requires sufficient read depth to accurately assess growth rate and requires assembly of reads on a genome scaffold (i.e., the strain's genome must have been sequenced and assembled). Also, PTR estimates the average value for all bacteria in urine whether they are truly planktonic, adherent to exfoliated epithelial cells, or dead with genomic DNA undamaged.
Determinations for growth rates by the plasmid segregation method and PTR at 6 h postinoculation in the murine model agree well, with one notable exception. The doubling times for pyelonephritis strain E. coli CFT073 in the bladder were measured at 36.3 and 36.9 min by plasmid segregation and PTR, respectively. Indeed, the values were nearly identical. Values for UPEC type strains CFT073, UTI89, and 536 were also similar at 6 h in the bladder by plasmid segregation at 36.3, 40.5, and 38.4 min doubling times (average of 37.5 min), respectively. The latter strains (UTI89 and 536) were not tested by PTR. The growth rate of fecal strain E. coli EFC7 was slower than that of CFT073, with similar doubling times in the bladder of 60.8 and 46 min, as assessed by plasmid segregation and PTR, respectively. On the contrary, asymptomatic bacteriuria strain E. coli ABU83972 had estimated doubling times of 113 and 37 min by plasmid segregation and PTR, respectively. The reason for this dramatic difference is unclear but could be due to increased plasmid stability in the ABU strain versus those in the UPEC and fecal strains tested.
Using the plasmid segregation and PTR techniques, we were able to answer fundamental questions about UPEC biology in the urinary tract. For example, can we understand the temporal dynamics of the infectious cycle? That is, how fast does UPEC adapt to nutrient availability within its host? In a mouse inoculated with an overnight culture grown in LB, the doubling time was 119 min at the time of inoculation. Six hours after inoculation, the growth rate in the bladder increased to 36.9 min as it adapted to growth in vivo. This doubling time was consistent with estimates of 30 to 35 min in the first 8 h within intracellular bacterial communities (IBCs) within the bladders of mice experimentally infected with UPEC strain UTI89 (26). Surprisingly, despite the fact that fresh urine is constantly synthesized in the kidney and delivered to the bladder, thus refreshing the growth medium, E. coli appears to enter a stationary growth-like phase with slow doubling times by 24 hpi (161 min) and 30 hpi (167 min). During human infection, E. coli cells grow in the gastrointestinal tract and then contaminate the periurethral area and ascend the urethra to the bladder, and symptoms of cystitis are elicited between 24 h and 3 days postinoculation of the bladder (27,28). It is at this point in the infection cycle that urine was collected from human patients, and PTR values were consistent with extraordinarily rapid growth (mean doubling time of 22.4 min). Thus, the human bladder appears to act more like a chemostat in which medium (urine) is refreshed constantly and outflow is accomplished by frequent urination. In the mouse bladder, urine may not be refreshed as rapidly as necessary to maintain exponential growth. This may reflect a fundamental difference between the murine model and human infection.
Differences in growth dynamics in the murine and human urinary tracts may also help explain differences in the expression of phase-variable type 1 fimbriae during infections in mice and humans. Selection for expression of these fimbriae is observed both under conditions of reduced oxygen and in the stationary phase of growth (29). Indeed, type 1 fimbriae are expressed during experimental murine UTI (30,31), especially late in infection. This would be consistent with UPEC growth entering stationary phase at high CFU (~10 8 CFU/g), likely due to limiting oxygen availability. On the other hand, several studies have found the orientation of the fim promoter is more often in the "off" position when examined directly from the urine of infected women (17,18,32). That E. coli cells examined in the urine collected and stabilized immediately have growth rates consistent with exponential growth may explain why a substantial percentage of the isolates are not expressing type 1 fimbrial genes during the human infections. Higher oxygen tension in the human bladder compared to the mouse bladder may also help to explain this, but has not been measured.
We can also ask whether there is growth variation occurring in different anatomical sites. When PTR values during murine infection were compared between the urine and the bladder for UPEC strain CFT073, similar doubling times were observed at 6 hpi (34.9 and 36.9 min, respectively); however, at 24 and 30 hpi, doubling times were dramatically longer in the urine than in the bladder. Given that PTR measures the average of all bacteria whether planktonic, adherent, or intracellular, bacteria in the bladder are replicating faster beyond 6 hpi. It is possible that concentrations of nutrients are higher in the bladder at later time points due to damage to the epithelium by the action of bacterial cytotoxicity and the process of inflammation.
Further, we may ask if meaningful comparisons can be made between strains with potentially different growth dynamics. We know well that UPEC strains display tremendous heterogeneity with respect to genes present in pathogenicity islands beyond the conserved base genome found in commensal strains (19,20,33). Indeed, this heterogeneity is displayed by the strains from women with symptoms of cystitis (17). Sequencing of these strains (34) revealed differences in genome size and the presence of a wide variety of accessory genes necessary to colonize the urinary tract (i.e., different combinations of adhesins, iron acquisition systems, and toxins). Consequently, the human UPEC strains displayed variation in growth in human urinary tract infection (Fig. 5). Indeed, expression of type 1 fimbriae may provide an alternative explanation for the differences observed in bladder colonization between strain types. Fecal strain EFC7 and asymptomatic strain ABU83972 do not agglutinate yeast (data not shown), indicative of a lack of expression of functional type 1 fimbriae.
E. coli transcriptome profiles from the urine of patients with urinary tract infection, described in three reports using either microarray technology (18) or RNA sequencing (17,35), are consistent with the ability to achieve rapid bacterial growth in vivo compared to in vitro culture (Fig. 5). In one study from urology clinic patients with E. coli bacteriuria (18), the most highly expressed bacterial genes in urine were those encoding ribosomal protein subunits. Ribosomal genes represented between 24 and 54% of the top 50 upregulated genes for the eight E. coli isolates compared to gene expression by the same isolates in LB cultures. Selected nonribosomal genes, upregulated during UTI, were also consistent with rapid growth in vivo, including those required for translation, the F o F 1 ATPase, fatty acid biosynthesis, and protein folding and secretion. E. coli from the urine of elderly patients with UTI (35) upregulated 202 genes compared to in vitro culture in rich medium. Twenty percent of these upregulated genes were involved in translation (ribosomal protein genes) or ATP synthesis, all indicative of rapid growth. Finally, in a transcriptome study of E. coli in the urine of patients with uncomplicated cystitis (17), similar genes, including those encoding ribosomal proteins, were highly upregulated during infection compared to in vivo growth in human urine or LB medium.
We have shown that the mean PTR value from E. coli strains collected during active uncomplicated UTIs is 2.0 Ϯ 0.5 (Fig. 5A). Although this value reflects a very high growth rate and exceeds the range of PTR values obtained in vitro used to establish a standard curve (Fig. 3 Certainly, future questions remain. For example, do UPEC strains persist by employing immune evasion or altering metabolic strategy at later time-points? What prevents UPEC strains in humans from progressing to a systemic infection with higher frequency than is observed clinically given the high growth rate calculated here? In all, however, this study highlights the use of PTR to estimate relative rates of growth in murine and human urinary tract infections and reveals that uropathogenic E. coli strains are replicating at a surprisingly high rate during human infection. This technique should be widely applicable for measurement of bacterial growth rates during infection.

MATERIALS AND METHODS
Extended materials and methods can be found in Text S1 in the supplemental material. Ethics statement. Urine collection was approved by the University of Michigan Institutional Review Board (HUM00004949). Informed consent for collection of urine specimens from women attending the University Health Services was approved by the Michigan Institutional Review Board (HUM00029910) (17). All animal protocols were approved by the Institutional Animal Care and Use Committee (IACUC) at the University of Michigan Medical School (PRO00005052).
Segregation of plasmid pGTR902 in vitro. Plasmid pGTR902, transformed into E. coli strains, was used to estimate growth rate as previously described (23).
Segregation of plasmid pGTR902 in vivo. Bacteria transformed with plasmid pGTR902 were cultured overnight with 1% arabinose and kanamycin (25 g/ml), harvested, and washed. A total of 10 8 CFU were transurethrally inoculated into the mouse bladder. At 6 hpi, bladders and kidneys were removed, homogenized, and plated onto LB agar with and without 1% L-arabinose and kanamycin (25 g/ml). The calculations used to determine the in vivo number of generations have been described previously (23). We have made the logical extension of these calculations to determine the doubling time.
Estimation of in vitro and in vivo growth rates via PTRs. Genomic DNA, isolated from bacteria harvested from culture medium, bladders, and urine of infected mice and bacteria in the urine of women with E. coli bacteriuria, was subjected to Illumina sequencing. PTRs were calculated using the peak and trough location with maximum and minimum values from the resulting smoothed coverage (36,(46)(47)(48).