The Microbial Etiology of Community-Acquired Pneumonia in Adults: from Classical Bacteriology to Host Transcriptional Signatures

SUMMARY

All modern advances notwithstanding, pneumonia remains a common infection with substantial morbidity and mortality. Understanding of the etiology of pneumonia continues to evolve as new techniques enable identification of already known organisms and as new organisms emerge. We now review the etiology of pneumonia (at present often called “community-acquired pneumonia”) beginning with classic bacteriologic techniques, which identified Streptococcus pneumoniae as the overwhelmingly common cause, to more modern bacteriologic studies, which emphasize Haemophilus influenzae, Staphylococcus aureus, Moraxella catarrhalis, Enterobacteriaceae, Pseudomonas, and normal respiratory flora. Urine antigen detection is useful in identifying Legionella and pneumococcus. The low yield of bacteria in recent studies is due to the failure to obtain valid sputum samples before antibiotics are administered. The use of high-quality sputum specimens enables identification of recognized (“typical”) bacterial pathogens as well as a role for commensal bacteria (“normal respiratory flora”). Nucleic acid amplification technology for viruses has revolutionized diagnosis, showing the importance of viral pneumonia leading to hospitalization with or without coinfecting bacterial organisms. Quantitative PCR study of sputum is in its early stages of application, but regular detection of high counts of bacterial DNA from organisms that are not seen on Gram stain or grown in quantitative culture presents a therapeutic dilemma. This finding may reflect the host microbiome of the respiratory tract, in which case treatment may not need to be given for them. Finally, host transcriptional signatures might enable clinicians to distinguish between viral and bacterial pneumonia, an important practical consideration.

INTRODUCTION

Until the second half of the 20th century, nearly all patients who had pneumonia came from their homes to seek medical attention: thus, by definition, pneumonia was “community acquired.” In fact, until the 1970s, with rare exceptions (1), terms such as hospital-acquired or health care-associated pneumonia simply did not exist (2–4). The present review focuses exclusively on what has come to be called community-acquired pneumonia (CAP) in adults, but which, following our earlier precedent (5), we shall simply call pneumonia.

From the 19th century through the 1950s, clinical, pathological, and, eventually, radiologic findings were used to distinguish “lobar pneumonia,” generally a disease of sudden onset causing dense consolidation involving a segment or an entire lobe of the lung, from “bronchopneumonia.” Bronchopneumonia in adults was regarded as less acute in onset, occurring as a complication of another (often viral) pulmonary infection and causing an infiltrate that was less dense, did not occupy an entire segment or lobe, and was actually “patchy” at autopsy (Osler [6], p 101–104; MacCallum [7]; Heffron [8], p 68–71). There was, however, some circularity to the reasoning, since the diagnosis of lobar pneumonia tended to be made when pneumococci were implicated as the cause. About 80% of patients hospitalized for pneumonia were thought to have lobar pneumonia, with the remaining cases being regarded as bronchopneumonia (9, 10). Already in the 1930s, Sutliff and Finland (11) questioned the reliability of this distinction, and, by the 1950s, it was largely abandoned. At the present time, in adults, with few exceptions (12), both forms are simply called pneumonia. During the time that the distinction was used, an etiologic agent was identified in nearly all cases of lobar pneumonia but only in about 80% of bronchopneumonia, as we shall discuss below.

METHODOLOGIC CONSIDERATIONS

Any study of the etiology of pneumonia must be understood in light of methodologic considerations, specifically, (i) tests that, at the time of the study, were available to identify a causative organism; (ii) in the case of sputum culture, the quality of the specimens as determined by definition (purulent) or microscopic examination; (iii) whether prior antimicrobial treatment had been given that would alter the results; (iv) the proportion of patients in which each of these tests was used and the proportion of these that were positive, and (v) whether data are presented as a percentage of all cases or only of those in which an etiologic agent was identified. Early data reported as a proportion of all those in whom a diagnosis was established are problematic because, with usually used techniques, atypical or nonbacterial causes (viral) simply could not have been identified even if they were present. Modern data reported in the same fashion remain problematic because a good sputum sample is often not available or is only collected after antibiotics have been given. If the denominator used is all patients with pneumonia, the proportion in whom a bacterial cause is identified may become vanishingly small (13).

Some reports from the preantibiotic era are meticulous in stating their methods. For example, Sutliff and Finland (11) and Bullowa and Wilcox (14) cultured blood by pour plates as well as by direct culture on agar, cultured other sites that might be secondarily infected, and plated sputum on blood agar only. These investigators also routinely injected sputum intraperitoneally into mice. They also stated the percentage of patients that had received prior treatment with antiserum to S. pneumoniae types 1 and 2. Some modern studies have cultured blood and sputum, measured serum procalcitonin, obtained urine for pneumococcal or Legionella pneumophila antigen, undertaken PCR for respiratory viruses in nearly all cases, and stated the proportion in which sputum was cultured before or within a stated time of administration of antibiotics (13, 15, 16). Others have used large databases in which none of this information was available (17), raising substantial questions regarding their reliability.

Because a causative bacterium is isolated from blood or another normally sterile site in only about 25% of patients with pneumococcal pneumonia (14), 20% of those with pneumonia due to Staphylococcus aureus (18), and in a much smaller proportion of patients with pneumonia due to other usually recognized bacterial pathogens such as Haemophilus influenzae (19) or Moraxella catarrhalis (20), diagnosis in the rest has traditionally been dependent on sputum Gram stain and culture. The results of these studies depend, in turn, on (i) the ability of a patient to provide a high-quality specimen (16, 21, 22), (ii) the assiduity with which health care personnel attempt to obtain such a specimen, and (iii) the timing of the specimen relative to the prior administration of antibiotics, data that are often not provided in published reports. The importance of a valid sputum sample was noted as long ago as 1919, in a comparison of sputum cultures with postmortem culture results (23). In 1971, Fekety et al. (24) predicted that many of the 40% of pneumonia cases in which no bacterial etiology was established might well be due to nonbacterial organisms. Therefore, it becomes of the greatest importance to consider whether data are presented as a percentage of all patients with pneumonia or only as a percentage of all those in which an etiologic agent is established (25).

ETIOLOGY OF PNEUMONIA, PREANTIBIOTIC ERA

If one includes both lobar and bronchopneumonia as defined above, the etiology of pneumonia in the preantibiotic era was more nuanced than we have previously reported (26). Unlike modern studies, in patients hospitalized for pneumonia, an etiologic agent was identified in nearly every case (27). Patients were kept under observation and were, therefore, able to provide a sputum sample at some point after, if not immediately upon admission, without eradication by an effective antimicrobial intervention. In the absence of treatment, complications gave a further opportunity to isolate a causative organism. When death resulted, autopsies were regularly performed (28), and lung contents were routinely cultured postmortem.

Large case series in 1917 (29), 1927 (9), and 1933 (11) implicated pneumococcus in more than 95% of cases of lobar pneumonia (Table 1). In fact, textbooks of medicine from Osler’s First Edition (1892) (30) through Cecil-Loeb’s Eleventh Edition in 1963 (31) discussed pneumonia and pneumococcus in the same chapter. Lower-numbered pneumococcal serotypes appeared to be the most virulent and were most frequently responsible for lobar pneumonia.

TABLE 1.

Etiology of lobar pneumonia and bronchopneumonia in the preantibiotic era^a

Etiologic agent	Lobar pneumonia (n = 3,319)	Bronchopneumonia (n = 815)
Pneumococcus	96.1	18–81
Streptococcus	2.8	7–50
Haemophilus	0.2	0–27b
Staphylococcus	0.2	1–12
Klebsiella	0.5	0
Mixed infection	0.2	12
No etiology identified	0	16–32

^aComposite data from Heffron et al. (8) are presented as percentage of total number of cases. Numbers for lobar pneumonia were very similar in all studies cited and are, therefore, averaged. Numbers for bronchopneumonia showed great variation and, therefore, are presented as a range.

^bIn one study, sputum was cultured only on blood agar, which generally does not allow growth of H. influenzae.

The etiology of bronchopneumonia in adults was more complex, with higher-numbered pneumococcal serotypes, other streptococci, Staphylococcus aureus, Haemophilus influenzae (presumably type b), or Gram-negative rods (Klebsiella and E. coli) being implicated in 20% to 40% of cases (9, 32). A causative organism was usually found; in one large series, a presumptive bacterial etiology was determined in 98% of cases (10).

However, by the late 1930s other causes of pneumonia were clearly identified. Influenza pneumonia had been well recognized during outbreaks, although we now know that it was frequently complicated by secondary bacterial pneumonia (33). An infection that was first called acute respiratory disease of adults occurred commonly in outbreaks, especially in military settings; this was first thought to be “viral” and later shown to be due to adenovirus (34). Reimann (35, 36) called attention to a pneumonia that behaved differently from pneumococcal pneumonia and that he first attributed to a “filterable virus.” Eaton et al. (37) solidified the term “primary atypical pneumonia,” (37), and Chanock et al. (38, 39) later showed conclusively that the causative organism was a Mycoplasma. Toward the end of World War II, a commission appointed by the U.S. Armed Forces investigated what had been recognized as frequent outbreaks of pneumonia in troops and differentiated pneumonia into four etiologic groups: pneumococcal, influenzal, “acute respiratory disease” (as defined above), and mycoplasmal (40).

ETIOLOGY OF PNEUMONIA, ANTIBIOTIC ERA

Bacteriologic Studies and Identification of “Atypical” Organisms

Of the original techniques for determining an etiologic agent in pneumonia, Gram stain, culture of sputum on sheep blood and chocolate agar plates, and blood cultures have persisted to the present. Bacteria that cause pneumonia grow readily on chocolate agar; however, through the 1960s, some laboratories used only blood agar, relying on satelliting to detect H. influenzae. Microscopic examination of Gram-stained sputum has been reported to be unreliable (41, 42) or very reliable (16, 43–45). Clearly, such discrepancies reflect differences in the quality of the specimen, the care with which the microscopic slide is prepared, and the skill of the examiner. Since what is coughed up is the inflammatory exudate that has collected in the alveoli, examination of good-quality sputum should be diagnostic. Attention needs to be paid to how purulent material is selected from a sputum cup for streaking slides and culture plates.

In contrast to the preantibiotic or early antibiotic era, persons who now develop pneumonia tend to be much older and have many more comorbid conditions; these factors are likely to increase oral colonization by Gram-negative rods (46, 47). Furthermore, many laboratories now routinely use MacConkey’s agar plates, which select for these organisms. As a result, sputum cultures often yield Gram-negative rods, which are present in small numbers but are simply contaminants. Our experience is that clinicians need to be taught to pay more attention to Gram stain results; the absence of Gram-negative rods on microscopic examination presumably excludes a diagnosis of Gram-negative pneumonia no matter what is grown on culture. Blood cultures are positive in about 5 to 8% of patients with pneumonia (48); urine antigens will be discussed below.

The discovery of effective antimicrobial treatment brought about a reduction of interest in the etiology of pneumonia. A careful search of the English-language medical literature (26) disclosed only eight studies of causative organisms in pneumonia between 1949 and 1980; in these reports, pneumococcus was identified as causing just under 50% of cases (26). This decreased prominence of S. pneumoniae as the cause of pneumonia was partly artifactual; hospitalized patients were now treated with an effective antibiotic (penicillin), so if they failed to provide a good sputum sample at admission, a diagnosis could often not be made. However, other etiologies were now being considered for community-acquired pneumonia, including influenza and adenoviruses, so-called “atypical” organisms, at first Mycoplasma and Coxiella burnetii (49), but then Legionella (50), and in the early 1980s, Chlamydia pneumoniae (51, 52). It is important to note that Mycoplasma infections were almost always described in children and young adults; older adults were much less frequently affected (53), and Grayston, who first identified Chlamydia pneumonia did his work on young adults seen at a university health clinic. The use of the term atypical to describe these infections is most inopportune. What they have in common is that examination of Gram-stained sputum will be negative and they will not grow on conventional media. However, the epidemiology, clinical features, and complications of these infections are quite different, and as we have pointed out previously (5), the term would best be avoided. Nonetheless, in this present article, we are forced to use it, because that is how it has been used to report results in earlier studies.

The etiology of CAP in Asia is quite different from that seen in the Western Hemisphere Gram-negative bacilli, especially Pseudomonas, Klebsiella, and Acinetobacter, play prominent roles (54–57). One explanation has been the dramatically increased rate of nasopharyngeal carriage of these organisms in southeast Asia (58), but a reason for that carriage has not been determined. Furthermore, in studies of CAP in Australia and New Zealand in particular, Legionella longbeachae is a notable cause (59); this may be due to unique ecological factors leading to environmental exposure in this region. In this review, we will focus on data from the Western Hemisphere and Europe. Numerous other agents from anthrax bacillus to Rickettsia typhi (60) have been identified in cases of CAP, but this review will confine itself to the generally seen common causes.

Detection of Pneumococcal and Legionella Antigen in Urine

As early as 1917, pneumococcal capsular polysaccharide was detected by precipitation when concentrated urine from patients with pneumococcal pneumonia was incubated with type-specific antiserum (8). This test was reported as positive in 62% (29) to 81% (61) of cases. In 2003, the FDA licensed an immunochromatographic technique to detect the cell wall (C-) polysaccharide of S. pneumoniae (62). This test has been applied widely in clinical practice. Two meta-analyses with surprisingly little overlap reported an overall sensitivity of 69% to 81% and a specificity of 84% to 98% (63, 64). In a recent study of CAP, detection of pneumococcal cell wall antigen in urine was the sole basis for the diagnosis in 71% of cases that were attributed to S. pneumoniae (65). Individual reports (63, 64) show very broad ranges in positivity and specificity, largely depending on patient selection and the basis for the final diagnosis.

Laboratory identification of Legionella by culture requires a good-quality sputum sample and growth on selective buffered charcoal yeast extract agar (66). Detection of soluble Legionella antigen in the urine (67) yields results that are increasingly positive in proportion to disease severity and detecting the antigen in up to 90% of cases (66). This test is only useful to detect Legionella serogroup 1, but this serogroup is responsible for about 90% of Legionella infections (68). To our knowledge, no urinary detection tests have been developed for diagnosing other causative agents of pneumonia.

More Complete Microbiologic Studies

In 1990, Fang et al. (69) reported the results of a seminal, prospective observational study carried out from 1985 to 1988 that examined the etiology of pneumonia using all traditional microbiologic techniques to which the investigators added serologic studies for so-called “atypical organisms” (Table 2). A bacterial cause was found in 41% of cases, but only 15% were attributed to S. pneumoniae and 11% to H. influenzae. An additional 15% of cases of so-called “atypical bacteria” (Legionella, Chlamydia, and Mycoplasma) were identified serologically. However, without knowing that acute and convalescent-phase sera were studied simultaneously and in what number an IgM result was used, it is difficult to evaluate these results; for reasons that remain unclear, IgM studies can be exceedingly difficult to replicate day to day (70).

TABLE 2.

Etiologic agents and studies used to detect them in 359 cases of community-acquired pneumonia, 1985–1988 (69)^a

Etiologic agent	Blood culture	Sputum culture	Serology	Total
Pneumococcus	4	11	–	15
Haemophilus	<1	10	–	11
Mycoplasma/Chlamydiae	–	–	8	8
Legionella	–	Not stated	“Most”	7
Gram-negative rodsb	<1	5b	–	6
S. aureus	1	2	–	3
Other bacteria	2	4	–	6
Pneumocystis	–	3	–	3
Postobstructive				5
Aspiration				3
Unknown				33

^aData are presented as percentage of total number of cases; total exceeds 100% because coinfections were detected in some cases. Viruses were not sought. –, not reported.

^bDefined as moderate or heavy growth without Gram stain confirmation or light growth with Gram stain confirmation.

Between 1993 and 2010, five studies from the United Kingdom and Europe used classical microbiologic, serologic studies, and PCR for atypical organisms to which they added a variety of other tests to detect pneumococci (Table 3). They reported pneumococcus as the cause of pneumonia in 30% to 64% of cases, but each had unusual features, a potential problem that underscores the importance of methodologic considerations. MacFarlane et al. (71) sought pneumococcal antigen in urine by latex agglutination or counterimmunoelectrophoresis, but they also cultured nasal swabs and saliva and did not state the positive study on which they relied for diagnosis. Huijskens et al. (72) used similar techniques but relied on PCR of nasal swabs rather than nasal cultures. Diagnosing pneumococcal pneumonia by detecting, in upper respiratory tract (URT) samples, a bacterium that colonizes the nasopharynx is obviously problematic. Snijders et al. (73) excluded patients who had chronic obstructive lung disease, thereby altering the denominator and greatly reducing the likelihood of identifying H. influenzae. Lim et al. (74) used unspecified serologic studies and found Chlamydia in 13% of cases. Johansson et al. (75) used PCR directly on sputum, a method that has yet to be validated, to implicate pneumococci in 64% of cases.

TABLE 3.

Intensive study of etiology of pneumonia, including conventional microbiology, urine antigen, serologic studies, and limited PCR^a

Etiologic agent	MacFarlane et al. (71)	Snijders et al. (73)	Lim et al. (74)	Johansson et al. (75)	Huijskens et al. (72)
Pneumococcus	30	37	48	64	34
Haemophilus	8	0	7	11	5
S. aureus	1	1	1	1	2
Gram-negative rods	1	4	1	1	4
Legionella	0	4	3	1	4
Viruses	8	5	19	29	29
Mycoplasma/Chlamydia	1	7	18	8	2
No pathogen detected	56	44	25	11	36

^aData are presented as percentage of total cases studied; totals exceed 100% because coinfections were frequently identified. See text for methodology.

From 2013 to 2015, the results of three important prospective clinical investigations into the etiology of CAP are presented (Table 4). In nearly identical studies, utilizing all validated microbiologic studies cited above, Musher et al. (15) at a single center and Jain et al. (13) in eight centers included consecutive adults who were hospitalized for pneumonia. In a study of the efficacy of pneumococcal conjugate vaccine in inpatients and outpatients, Bonten et al. (76) used highly sensitive urine antigen detection tests for 13 common serotypes of S. pneumoniae. In all three studies, the rate of detection of bacteria was low (15 to 30%) and of pneumococci even lower (5 to 16%). Other pathogens were detected in roughly similar proportions of cases, but no etiologic agent was found in 46 to 62% of cases. The highest yield was in the single-center study (15), presumably because a tighter degree of control over obtaining sputum samples was possible. Respiratory viruses were frequently detected (see below for a full discussion).

TABLE 4.

Etiology of infectious pneumonia in adults 2011–2014: all studies as above, with molecular studies for viruses and “atypical bacteria”^a

Etiologic agent	Musher et al. (15)	Jain et al. (13)	Bonten et al. (76)
Bacteria	28	15	30
Pneumococcus	9	5	16
Haemophilus	6	<1	7
S. aureus	5	2	3
Pseudomonas	3	<1	2
Legionella	1	1	1
Mycoplasma/Chlamydia	–	<3	1
Other bacteria	4	3	3
Mycobacteria	2	1	<1
Nocardia	1	0	0
Fungi (PCP)	3	1	2
Viruses	25	27	3
Rhinovirus	12	9	–
Coronavirus	3	2	–
Human metapneumovirus	2	4	–
Influenza	1	6	3
Parainfluenza	2	3	–
RSV	2	3	–
No cause identified	46	62	66

^aData are presented as percentage of total cases studied; totals exceed 100% because coinfections were frequently identified. PCP, pneumocystis pneumonia; RSV, respiratory syncytial virus; –, not done.

Taken together, these bacteriologic results have led some investigators (26) to conclude that (i) there had been a steady decline in the role of pneumococcus as a cause of pneumonia since the beginning of the antibiotic era, and (ii) pneumococcal infection remained more common in Europe than in the United States, which the authors attributed to a substantially higher uptake of pneumococcal vaccine and a decline in cigarette smoking among adults in the United States. A recent systematic review of the literature through 2020 (25) showed that the frequency with which pneumococcus has been implicated ranges from 5% (13) to 64% (75), averaging about 25% to 40% of those cases in which an etiology was demonstrated, with the important caveat that in many or most cases no etiologic agent was identified.

The Low Yield in Prospective Studies

The low yield of an etiologic agent of pneumonia even in prospective studies results from the important fact that the only way to determine a bacterial cause in most cases of pneumonia is to have valid sputum specimens to study. Cultures of blood are positive in only 5% to 10% of all-cause CAP (48). Urine antigen studies detect a variable proportion of pneumococcal (see above) and Legionella (64, 77) cases of pneumonia but, to date, no others. Because the current standard of care (48) mandates prompt administration of an empirical antibiotic after pneumonia is diagnosed, if a patient is unable to cough up good-quality sputum shortly after admission, no etiology is likely to be found. In patients with proven (bacteremic) pneumococcal pneumonia, Musher et al. (78) related the positivity of sputum cultures to the time after antibiotics were first administered in patients with bacteremic pneumococcal pneumonia. Within 6 h, pneumococci were identified in sputum by Gram stain and/or culture in >75% of cases; this number fell to 27% after 24 h.

In a thorough meta-analysis of studies of microscopic examination of high-quality, Gram-stained sputum, Ogawa et al. (21) showed a specificity of 91% and a sensitivity of 61%. The lower sensitivity likely reflected prior antibiotic therapy; the meta-analysis did not address the duration of antibiotic therapy before specimens were obtained, probably because such information is rarely available. Studies included in the meta-analysis defined a high-quality specimen as one that had >25 white blood cells (WBCs) per 10 epithelial cells; this definition was stated in the conclusion of the classic paper by Murray and Washington (79), although the figures in that article show a much higher ratio of inflammatory to epithelial cells.

Quantitative Bacteriologic Studies of Valid Sputum Specimens

If one were to confine results to patients who could provide valid sputum specimens promptly upon hospitalization, results might look different from those reported above. Musher et al. (16) studied patients who were able at the time of, or shortly after admission, to provide purulent sputum that had >20 WBCs per epithelial cell on microscopic examination. The investigators quantitated bacteria in solubilized specimens and accepted ≥10⁵ bacteria per mL of usually recognized bacterial pathogens, such as S. pneumoniae, H. influenzae, or S. aureus as diagnostic of an etiologic agent (19, 80, 81). Using more stringent criteria, namely, the presence of >10⁶ bacteria per mL, these investigators implicated commensal organisms such as Streptococcus mitis, Streptococcus salivarius, or Corynebacterium spp. that are traditionally reported as normal respiratory flora and disregarded as etiologic agents. Gram stain and quantitative culture results were in full accord in nearly every case. However, in this kind of study, selection bias may have resulted from the possibility that certain infections are more likely than others to cause purulent sputum in patients who are sick enough to be hospitalized for pneumonia, a question that, to our knowledge, has not been answered.

Utilizing these classical quantitative microbiologic techniques, together with PCR on nasopharyngeal specimens to detect respiratory viruses or atypical bacteria, Musher et al. (16) identified ≥1 etiologic agent(s) in 95% of patients hospitalized for CAP (Table 5). The majority (57%) were due to recognized bacterial pathogens with S. pneumoniae and H. influenzae each being found in about one-quarter of cases. Another 26% of cases were caused by commensal flora, including viridans, streptococci (of which Streptococcus mitis was clearly predominant) (82), and others. These organisms are readily detected by microscopic examination but are documented convincingly by only quantitative culture techniques that are not practicable in diagnostic laboratories. S. mitis is closely related genetically to S. pneumoniae (83), and some isolates even express pneumococcal capsule. In 20% of patients with a recognized bacterial cause and 32% of those with pneumonia attributed to commensal bacteria, a respiratory viral coinfection was implicated. A respiratory virus alone was the cause of pneumonia in 13% of cases. Had attention not been paid to commensal bacteria, a total of 45% of cases of pneumonia might have been said to be viral. This study showed that if appropriate sputum specimens are available and there is an examination of sputum together with a nasal PCR for viruses, Mycoplasma, and Chlamydophila, then an etiologic diagnosis can be determined in nearly all cases of pneumonia.

TABLE 5.

Intensive study of etiology of pneumonia in 120 patients who provided a high-quality sputum specimen at or shortly after admission^a

Etiologic group	Number (%)
Recognized bacterial pathogen	54 (45.0%)b
Recognized bacterial pathogen + viral coinfection	14 (11.7%)
Respiratory virus alone	16 (13.3%)
Commensal bacteria alone	21 (17.5%)
Commensal bacteria alone + viral coinfection	10 (8.3%)
Undetermined	5 (4.2%)
Total	120 (100%)

^aAdapted from Musher et al. (16).

^bIncluded in this category are 13 cases in which a recognized bacterial pathogen was isolated but, based on Gram stain and quantitative culture results, an etiologic role for coinfection with commensal flora was thought to be likely.

The importance of small-scale studies in which all clinical information is available and the quality of specimens can be assured is best illustrated by a comparison of the above findings with those of Haessler et al. (17). These investigators used database studies of 139,000 patients hospitalized for pneumonia, many of whom met previously used criteria for health care-associated pneumonia; they found that only 14% of respiratory cultures grew pneumococci, and 4% grew H. influenzae, while 17% and 18% yielded Pseudomonas aeruginosa and other Gram-negative rods, respectively. It has long been documented that Gram-negative flora colonize the mouth in patients who are hospitalized and/or have previously been treated with antibiotics (47). Sputum is routinely cultured on media that select for Gram-negative bacilli. When such organisms are reported, clinicians make decisions based on those findings often without regard to Gram stain results, thereby leading to excessive diagnoses of, and treatment for Gram-negative pneumonia.

NUCLEIC ACID AMPLIFICATION TESTS

General

Molecular techniques have the advantage of detecting bacteria whether “living” (capable of replication in vitro) or “dead.” An important underlying assumption of molecular techniques in diagnosing pneumonia is that bacterial genomic fragments will persist for some length of time after appropriate antibiotic therapy has begun; thus, results will not immediately be altered by prior antibiotic administration. Recent molecular bacterial quantitative studies are a logical progression from developments in the technology available to detect other respiratory pathogens over the last 3 decades. At this time, there has been a fundamental shift in the laboratory diagnosis of respiratory viral infection from cumbersome cell culture-based techniques, immunofluorescence, and antigen tests, to highly sensitive and specific rapid molecular methods of viral detection. The most commonly used molecular methods have employed PCR (for DNA viruses) and reverse transcription-PCR (RT-PCR) for RNA viruses. However, other methodologies such as nucleic acid sequence-based amplification (NASBA) and loop-mediated isothermal amplification (LAMP) have also been explored over time; these methods are collectively referred to here as nucleic acid amplification tests (NAATs) for simplicity. During this period, there was a general technological progression from the amplification of single viral targets to multiple viral targets in one reaction tube and from gel electrophoresis and hybridization-based detection to real-time detection using integrating fluorescent dyes, hydrolysis probes, or microarrays (84–94).

NAATs were also developed for other bacteria that cause pneumonia such as M. pneumoniae, C. pneumoniae, C. psittaci, L. pneumophila, and other Legionella spp. (37, 38, 95–97) that are difficult (and in the case of C. psittaci, hazardous) to culture in the routine diagnostic laboratory. Serological detection methods, which had been available for various periods of time, lacked sensitivity or specificity and, most importantly, were only useful retrospectively through the demonstration of seroconversion between acute and convalescent-phase sera.

Detection of Bacteria That Are Not Readily Cultivatable and Viruses

Initially, multiple multiplex NAATs were used to detect bacteria that were difficult to isolate using conventional techniques (atypical bacteria) and viral targets, while classically recognized bacterial pathogens were still sought by traditional methods. Templeton et al. (98) used such a strategy in a prospective study of 105 adults with CAP in both outpatient and inpatient settings; an etiologic agent was detected in 50% of patients by conventional methods and by real-time PCR in 76% of patients (Table 6). The difference was due to increased detection of respiratory viruses and atypical bacteria using real-time PCR compared to viral culture and serology. Rhinovirus was detected in 17%, coronavirus in 13%, and influenza in 11% of cases; atypical pathogens were detected in 19%, with M. pneumoniae most common and Chlamydia spp. least common. By comparison, the most commonly cultured bacterium from sputum was S. pneumoniae in 21% of cases. Mixed infections, including mixed bacterial and viral infections, were found in only 3% of cases by conventional methods, and 27% of cases by NAAT. A potential limitation of this kind of study is that NAATs are carried out on URT specimens whereas sputum is used for bacterial culture; the problem of being uncertain of the role of URT samples in lower respiratory infection diagnosis persists unresolved to the present day.

TABLE 6.

Viral and atypical bacterial detection in adults with pneumonia using molecular testing approaches^a

Etiologic agent	Templeton et al. (98)	Johansson et al. (75)	Huijskens et al. (72)	Gadsby et al. (22)	Gilbert et al. (120)
Viruses
Rhinovirus	17	8	8	13	15
Influenza	11	3	8	7	22
Coronavirus	13	3	6	3	10
Parainfluenza	8	<1	6	3	3
Adenovirus	8	ND	<1	2	1
RSV	3	<1	2	1	8
HMPV	0	2	<1	<1	8
“Atypical” organisms
M. pneumoniae	10	8	<1	2	ND
Legionella spp.	6	<1	4	2	ND
Chlamydophila	3	0	2	<1	ND

^aData are presented as percentage of total cases studied; totals exceed 100% because coinfections were frequently identified. RSV, respiratory syncytial virus; HMPV, human metapneumovirus; ND, not done.

In a study of 156 Swedish adults hospitalized with CAP (75), in which URT specimens were tested by a similarly broad panel of real-time viral PCR assays, only 17% of cases were positive for a virus by NAAT (Table 6), a figure that was comparable overall to conventional viral culture and serology. The most common viral findings by NAAT were rhinovirus in 8%, coronavirus in 3%, and influenza in 3% of cases. M. pneumoniae was detected in 8% of cases tested, and 0% with C. pneumoniae. About one-third of patients had bacterial/viral coinfections.

In a later and larger prospective study of 408 adults presenting to the hospital with CAP in the Netherlands, Huijskens et al. (72) deployed an extensive range of real-time PCR assays for viruses and atypical bacteria on both sputum and throat swab specimens, in addition to conventional culture, antigen testing, and serological studies. Viruses were detected in 29% of cases and bacterial pathogens in 52%, with 16% of patients having viral and bacterial coinfection. Rhinovirus and influenza virus were each detected in 8% and coronaviruses and parainfluenza virus each in 6% of cases; atypical bacteria were detected in 15% of cases, with C. burnetii (detected serologically) being the most common, due to an outbreak in the area at the time, and M. pneumoniae and C. pneumoniae least common (Table 6).

The increasingly unwieldy series of multiple multiplex NAATs required for diagnosis in patients with pneumonia has progressed into molecular panel assays and automated platforms capable of detecting 20 or more different targets in a single reaction (93, 99–102). One such panel (BioFire FilmArray Respiratory Panel) was used by Musher et al. (15) on URT specimens from 215 consecutive patients admitted with CAP; a respiratory virus was identified in 25% of patients, one-half of whom had bacterial coinfection (Table 4). Rhinovirus was identified in two-thirds of the viral detections, with the remainder being coronavirus, parainfluenza virus, respiratory syncytial virus, human metapneumovirus, and only a single case of influenza, likely to due a high level of vaccination and a nonoutbreak year.

Jain et al. 2015 (13) in their similarly designed study of patients hospitalized for CAP used CDC-developed methods on URT specimens (with the exception of L. pneumophila testing on sputum) and found viruses in 27% of cases; the majority were diagnosed by NAAT with a small contribution from serological tests (Table 4). The most commonly detected virus by NAAT was rhinovirus in 9% of cases, influenza in 5% of cases, with lower incidences of respiratory syncytial virus, parainfluenza virus, coronaviruses, and adenovirus. Atypical bacteria, most commonly M. pneumoniae, were detected in fewer than 5% of cases. Thus, PCR results on URT samples in Musher et al. and Jain et al. were remarkably similar. Using a range of bacterial detection methods, Jain et al. found that only 3% of cases had bacterial/viral coinfection but few patients provided a good quality sputum specimen for culture in a timely fashion, and overall rates of bacterial detection (15%) were very low. To examine the possibility that NAATs were detecting postinfection shedding or infection not relating to the lower respiratory tract (LRT), the same viral NAATs in controls matched by time of year and location found an overall low rate of detection (2%).

Gadsby et al. (22) used an in-house developed NAAT panel for a broad range of 26 viral and bacterial respiratory pathogens on good-quality sputa or endotracheal aspirates adults hospitalized for CAP in the United Kingdom. Viruses were detected in 30% of 323 patients; rhinovirus was most common (13% of cases) followed by influenza virus (7%), with lower incidences of parainfluenza virus, coronavirus, adenovirus, respiratory syncytial virus, and human metapneumovirus (Table 6). Over two-thirds of viral cases had codetection of bacteria by either culture or NAAT methods; atypical bacteria were only detected in 4% of LRT specimens by NAAT.

In conclusion, despite differences in geographical location, time, type of respiratory specimen (URT versus LRT), and NAAT technology, the results of viral detection by NAAT in patients with CAP have been strikingly similar, with rhinovirus being most common, followed by influenza virus. Detection of bacteria and so-called atypical agents appears to be more variable, the former dependent on efforts to obtain a sputum sample in timely fashion and the latter likely influenced by the occurrence of cyclical M. pneumoniae epidemics and local outbreaks of other agents such as C. burnettii. Worthy of emphasis is that, in recent studies using standardized detection methods, these so-called atypical organisms have been implicated in ≤5% of cases of CAP.

Quantitation of Bacterial Pathogens

In recent years, targets for readily cultivatable bacteria such as S. pneumoniae, H. influenzae, M. catarrhalis, and S. aureus that are typically recognized as causing pneumonia have been added to respiratory NAAT panels, both those designed in-house and those commercially available (75, 103–109). Some also deliver either semi- or fully quantitative bacterial load outputs (75, 103, 109). The newest panels contain additional Gram-negative bacterial targets and resistance genes most relevant for hospital-acquired or ventilator-acquired pneumonia. Head-to-head comparator studies of NAATs versus culture methods in patients with pneumonia usually show more frequent detection of respiratory bacteria by PCR than by culture, regardless of molecular assay or platform used, despite substantial differences among the molecular methods themselves (110). Many of these studies have focused on specific patient populations, such as those with hospital-acquired or ventilator-acquired pneumonia, HIV infection (111), other immunocompromising conditions, or severe pneumonia or those admitted to critical care (108, 110, 112–117). Conversely, others have used very broad inclusion criteria such as all types of lower respiratory tract infection (RTI) or have included less clinically well-defined CAP (104–106, 118, 119). Relatively few studies have used these broad viral and bacterial NAAT panels on LRT specimens specifically to study the etiology of patients admitted from the community with pneumonia (13, 15, 120, 121).

Johansson et al. (75) used quantitative real-time PCR for S. pneumoniae, H. influenzae, and M. catarrhalis on good-quality sputa taken before antibiotics. They attributed a probable etiologic diagnosis with a threshold for culture or PCR of ≥10⁵ CFU/mL for S. pneumoniae and ≥10⁶ CFU/mL for other bacteria. Although limited to a small number of bacterial pathogens, NAATs contributed to a modest increase in CAP etiological diagnosis in this cohort, finding 33% of 126 patients positive by quantitative NAAT compared to 26% by sputum culture (Table 7).

TABLE 7.

Bacterial detection in pneumonia using molecular panel or quantitative molecular assays on sputum or other LRT samples^a

Etiologic agent	Johansson et al. (75)	Gadsby et al. (22)	Gilbert et al. (120)b	Serigstad et al. (121)
H. influenzae	4	40	34	35
S. pneumoniae	8	36	29	25
M. catarrhalis	0	14	14	11
S. aureus	–	11	26	6
E. coli	–	12	NS	7
K. pneumoniae	–	4	NS	4
Enterobacter spp.	–	–	12	1
Other Enterobacterales	–	–	NS	8
P. aeruginosa	–	3	NS	3
A. baumannii	–	<1	NS	1
“Nonfermenters”	–	–	3	–
S. agalactiae	–	–	8	7
S. pyogenes	–	–	3	0

^a–, not done.

^bNS, not stated.

Gadsby et al. (22) similarly used quantitative real-time PCR but extended this approach to eight additional respiratory bacteria with different bacterial gene targets from those cited above, alongside the viral and bacterial targets discussed earlier. In a series of patients who were able to provide purulent sputum or who were intubated, bacteria were detected in 81% of patients, very few of which were so-called atypical bacteria. The majority were positive for typically cultivatable respiratory bacteria, S. pneumoniae in 36% and H. influenzae in 40%; detection of more than one bacterial species occurred in 32% (Table 7). Taking the more stringent cutoff ≥10⁵ genome copies/mL for all targets reduced the total level of bacterial detection from 81% to 72%. It is important to note that these results were obtained despite 85% of patients having been already on treatment with antibiotics at the time of specimen collection, and only "high-quality" specimens were included for study.

Two recent studies have utilized the new BioFire Pneumonia FilmArray panel (BPFA) (bioMérieux). This multiplex panel detects 15 respiratory bacterial agents semiquantitatively (as either 10⁴, 10⁵, 10⁶, or ≥10⁷ genome copies/mL) along with three atypical bacterial and eight viral agents. In an important study, Gilbert et al. (120) utilized blood and LRT sample cultures, supplemented with either BPFA or a multitest bundle (comprising URT specimen multiplex viral NAAT plus S. pneumoniae and S. aureus PCR, and urinary antigen tests for S. pneumoniae and L. pneumophila) in 274 hospitalized CAP patients with purulent sputum or other LRT specimen available. The BPFA detected ≥1 bacterial pathogen(s) in 76% of patients and a viral pathogen in 61% of patients; 91% of patients had at least one pathogen detected, which was higher than the multitest bundle on each of the three measures (66%, 41%, and 81%, respectively). Findings by BPFA in this CAP cohort were broadly similar to those of Gadsby et al. (22) and Musher et al. (16) (Tables 5 and 6); the most commonly detected bacteria were S. pneumoniae (28%) and H. influenzae (34%), and viruses were influenza (22%) and rhinovirus (15%). The authors did not use the semiquantitative output of the BPFA to attempt to distinguish possible colonization from infection because concordance of bacterial culture quantification with the BPFA panel quantification over a range of bacterial loads is relatively low at 44% (115). Instead, they used procalcitonin serum levels of <0.25 ng/mL to distinguish likely bacterial colonization from infection, finding that 20% of cases with bacterial detection alone had low procalcitonin, as did 33% of cases with both viral and bacterial detection. Unfortunately, procalcitonin levels are only 70 to 80% sensitive or specific for recognizing bacterial infection (122).

In 72 patients hospitalized in Norway with suspected CAP, Serigstad et al. (121) collected induced sputa for testing by the Biofire Pneumonia FilmArray Plus (BPFA with the addition of a target for MERS-CoV) and compared results to blood culture, sputum or endotracheal aspirate culture, viral and atypical bacteria PCR on URT specimens, pneumococcal urinary antigen testing, and point-of-care testing for influenza, as part of a wider study of the utility of the panel in suspected RTI. In this study, however, 58% of specimens were not of good quality and both radiologically confirmed CAP and clinically diagnosed CAP without radiological confirmation were considered a single entity in the analysis. Like Gilbert et al. (120), the authors did not use the semiquantitative output as a potential proxy measure of colonization versus infection, but instead of procalcitonin levels, they used predetermined microbiological and clinical criteria to stratify organisms detected by each testing modality into different categories according to their potential clinical relevance. The majority of the BPFA plus panel typical bacterial targets were classed as “usually not pathogens” unless in certain circumstances such as chronic disease, repeated antibiotic treatment, and immunosuppression; H. influenzae, S. pneumoniae, and S. pyogenes were classed as “usually pathogens.” The ones deemed clinically relevant were predominantly S. pneumoniae in 25% or H. influenzae in 35% (Table 7). The viruses and atypical bacteria in the panel were deemed “always pathogens,” and the most commonly identified viruses by BPFA plus were influenza (22%) and human metapneumovirus (14%), likely influenced by the short winter recruitment period; atypical bacteria were detected in only 4% of cases.

Although many studies have utilized broad molecular panels such as BPFA to investigate bacterial coinfections, often hospital acquired, in critically ill patients with COVID-19 (123), we are unaware of comparable quantitative molecular virologic and bacteriologic studies in CAP in the current SARS-CoV-2 pandemic era.

Clinical Applications

An unresolved problem that may be of the greatest importance to clinicians and to hospital epidemiologists is how to interpret the lack of concordance in which NAATs identify bacteria in sputum that are not found by culture, a finding that has been reported in numerous previous studies (22, 109, 110, 115, 118, 124, 125). The crucial question is whether the finding of bacteria by NAATs indicates greater sensitivity of the molecular techniques and treatment should be directed against bacteria so-found or whether it should be interpreted as a “false positives” and be ignored by the clinician. With the increasing use of these tests, clinicians will increasingly be prompted to select antibiotic treatment depending on the interpretation.

A recent study (D.M. Musher, J.J. Dunn, and N.J. Gadsby, in preparation) has specifically addressed the concordance issue by comparing quantitative bacterial cultures with the results of two NAATs (BPFA and Gadsby’s in-house assay) in patients with CAP who provided frankly purulent sputum at or near admission and received minimal or no antibiotics before study. Results from the two NAATs were quite consistent with each other, both finding >10⁶ gene copies for the same organisms in most instances. However, in nearly one-half of the cases, these organisms were neither seen by Gram stain nor found at culture. Most prominent among bacteria detected by NAAT but not by bacterial culture were P. aeruginosa, Streptococcus agalactiae (an uncommon cause of CAP [126]), and S. aureus. Earlier reports of NAATs have found similar results. For example, Murphy et al. (109) identified P. aeruginosa, S. aureus, H. influenzae, or M. catarrhalis each in 5.6% and 9.7% of respiratory samples that did not yield these organisms on culture. These authors called these findings “false positives,” as did Buchan et al. (118). In official recommendations from the Infectious Diseases Society of America, Hanson et al. (127) attributed this finding to “inherent sensitivity of NAAT combined with potential detection of dead, fastidious, colonizing, or metabolically impaired organisms.”

In our small series of cases, treatment was not given to several patients in whom NAATs found P. aeruginosa or MRSA (about 20% of cases, respectively), and the patients recovered well on usually recommended empirical antibiotic therapy that is not effective against these organisms. If clinicians accept NAATs as definitive, antibiotics will be added to treat P. aeruginosa and/or methicillin-resistant S. aureus in some proportion of cases in which the treatment may not be indicated. Authors, including Buchan et al. (118) and Gadsby et al. (22), have suggested that the use of quantitative NAATs will enhance efforts at antimicrobial stewardship, primarily in populations where cultures are frequently negative due to prior antibiotic exposure. However, this may not be the case for every patient. In their current iteration, such NAATs involve a trade-off between the likelihood of overtreating some patients, while benefiting others by permitting deescalation; the exact proportion in each group will depend very much on the clinical setting.

There is, as yet, no explanation for the detection of high numbers of bacteria by molecular techniques that are not found by conventional bacterial culture in the absence of salivary contamination or prior antibiotic therapy. Shedding of molecular constituents from microbiome bacteria in a biofilm state may be possible. Host biomarkers of lung infection (having mainly been studied in ventilator-associated pneumonia) (128) and respiratory microbiome profiling (see below) may cast further light upon the utility and context of these molecular results. In the meantime, results are awaited from clinical trials under way in community-acquired, hospital-acquired, and ventilator-associated pneumonia, which aim to focus antibiotic prescribing based on the outputs of broad bacterial NAAT panels (some of which are at least semiquantitative). Conversely, a negative broad bacterial NAAT result on a good quality LRT specimen may have a useful negative predictive value in CAP, although neither NAAT nor standard microbiologic techniques will exclude a role for so-called normal respiratory flora.

HOST TRANSCRIPTIONAL SIGNATURES

As noted above, health care providers are unable to obtain a high-quality sputum at admission from the majority of patients hospitalized for pneumonia, even despite intense efforts to do so. If sputum specimens are to be sent for bacterial NAATs without microscopic examination, there might be no way to distinguish upper airway colonization from pulmonary infection. From the point of view of treatment, an important clinical need is to distinguish between bacterial and viral pneumonia; empirical antibiotics have a high likelihood of success for the former and are unnecessary and potentially harmful in the latter.

Microarray technology offers the possibility of making this distinction without examining or culturing sputum. Bacterial and viral pneumonia appear to upregulate different host genes that may be recognized by examining gene expression signatures using microarray, reverse transcription-quantitative PCR (qRT-PCR) and RNA sequencing (123, 129, 130), or reverse transcriptase (RT) or multiplexed RT-OCR technology (131). Furthermore, metagenomic next-generation RNA and DNA sequencing for the combined detection of pathogens, microbiome, and host transcriptional signatures may have utility in determining infectious versus noninfectious respiratory illness in critically ill patients (132).

HOST MICROBIOTA

The microbiome of the LRT greatly influences the proliferation of viruses and potentially pathogenic bacteria. A general balance among microorganisms protects the lower airway from infection by preventing proliferation of pathobionts in the upper airways, and it has been suggested that bacterial infection can be distinguished from viral based on analysis of the microbiota of the respiratory tract (133). Enrichment of one bacterial taxa at the expense of others may lead to an imbalance that facilitates the proliferation of potentially pathogenic organisms, leading to pneumonia. Haak et al. (134) have recently used this concept in a prospective study to distinguish bacterial from viral pneumonia with a high degree of reliability, but, to our knowledge, application to clinical practice has yet to take place. In children, a 10-day course of antibiotics for pneumonia caused greater changes to the respiratory resistomes and microbiota than a 5-day course (135), but this concept has yet to be related to the etiology of pneumonia.

CONCLUSIONS

In studies of the etiology of pneumonia, widely disparate findings are likely due to patient selection and other methodologic considerations. Despite the availability of a vaccine against S. pneumoniae and, in the absence of a vaccine for nontypeable H. influenzae, these two organisms continue to cause a total of about 50% of cases of pneumonia that lead to hospitalization. Many other recognized bacterial pathogens, so-called ‘atypical’ organisms or commensal organisms continue to be implicated. In patients who have not been treated with antibiotics, microscopic examination and culture of high-quality sputum specimens provide reliable information regarding the etiology. NAATs, which are both extremely sensitive and highly specific, detect recognized bacterial pathogens in sputum from patients for whom bacterial cultures are negative. An unanswered question is whether this finding indicates bacteria that are causing pneumonia or simply reflects bacterial presence in the microbiome and, therefore, how clinicians should treat such data remains uncertain. Our interpretation is that organisms that are detected only by NAATs and not by culture are probably not causing acute bacterial pneumonia and should not be treated.

Although M. pneumoniae and C. pneumoniae most commonly infect children and young adults, most adults hospitalized for pneumonia in developed countries are over the age of 65. There is probably too much consideration given to these so-called atypical agents in guidelines for empirical treatment of patients with community-acquired pneumonia. NAATs of nasopharyngeal swabs in patients hospitalized for pneumonia have identified Mycoplasma and Chlamydia each in about 1 to 2% of all cases, and these may well have been in young adults in whom the disease is much more common. Unfortunately, the same term, atypical, is applied to pneumonia caused by Legionella, which occurs at about the same frequency and may be life threatening. Whereas the guidelines from the United States (48) recommend routine addition of a macrolide to a beta-lactam to treat patients hospitalized for pneumonia, the Swedish (136), Dutch (137), and United Kingdom (138) guidelines recommend addition of a macrolide only if an atypical organism is suspected clinically or in cases of severe infection.

A respiratory virus is found in 25 to 60% of patients with pneumonia, in a substantial proportion of whom, viral and bacterial coinfection occurs. These findings support the initial selection of recommended empirical antibiotic therapy for persons who present with pneumonia even if a NAAT at admission reveals a respiratory virus. The use of available laboratory techniques, whether microbiologic, with emphasis on high-quality sputum specimens, or molecular, should facilitate selection of an appropriate narrow-spectrum antibiotic. However, NAATs do not detect commensal organisms, which are always present to some degree on culture plates, so withholding antibiotics without a negative microscopic examination of a high-quality sputum obtained before antibiotic administration is potentially problematic. Procalcitonin is neither sufficiently sensitive nor specific to allow reliance upon it for decision-making. Host transcriptional analysis or studies of the host microbiota are likely to enhance these efforts, reducing the overall risk of unneeded or unnecessarily broad-spectrum antibiotics in patients hospitalized for pneumonia, but much more study is needed.

Biographies

Naomi J. Gadsby, Ph.D., graduated from the University of Cambridge, UK in 2001 and received her PhD in Infection and Pathogen Genetics from the University of Edinburgh, UK in 2004. She completed her clinical scientist training in the National Health Service (NHS) in Edinburgh, UK and became a Fellow of the Royal College of Pathologists in Medical Microbiology in 2017. Since 2019 she has been working for NHS Lothian as a Consultant Clinical Scientist in Medical Microbiology at the Royal Infirmary of Edinburgh, UK and is an Honorary Fellow of the University of Edinburgh. Her interests include molecular diagnostics for bacterial pathogen diagnosis, hospital infection prevention and control, and medical and scientific training in clinical microbiology.

Daniel M. Musher, M.D., is a Distinguished Service Professor of Medicine and Professor of Molecular Virology and Microbiology at the Baylor College of Medicine and a member of the Infectious Diseases Section at the Michael E. DeBakey Veterans Administration Medical Center, Houston, Texas, USA. He has worked on Treponema pallidum, Staphylococcus aureus, Proteus mirabilis, and the milleri streptococci, as well as on syphilis, urinary tract infections, pneumonia, and acute myocardial events associated with infections. The present paper is his 600th publication. He is an avid player of string quartets and was the founding concertmaster of the Texas Medical Center Orchestra.