Comparative microbiomics analysis of anti-microbial antibody response between lung cancer patients and control subjects with benign pulmonary nodules

Abstract

Background:

Computer Tomography (CT) screening can detect lung cancer early but suffers a high false positive rate. There is a need for molecular biomarkers that can distinguish malignant and benign indeterminate pulmonary nodules (IPN) detected by CT scan.

Methods:

We profiled antibodies against 901 individual microbial antigens from 27 bacteria and 29 viruses in sera from 127 lung adenocarcinoma (ADC), 123 smoker controls (SMC), 170 benign nodule controls (BNC) individuals using protein microarrays to identify ADC and BNC specific anti-microbial antibodies.

Results:

Analyzing 4^th quartile odds ratios, we found more antibodies with higher prevalence in the 3 BNC subgroups than in ADC or SMC. We demonstrated that significantly more anti-Helicobacter pylori antibodies showed higher prevalence in ADC relative to SMC. We performed subgroup analysis and found that more antibodies with higher prevalence in light smokers (≤ 20 pack-years) compared with heavy smokers (> 20 pack-years), in BNC with nodule size > 1cm than in those with <= 1cm nodules, and in stage I ADC than in stage II and III ADC. We performed multivariate analysis and constructed antibody panels that can distinguish ADC vs. SMC and ADC vs. BNC with area under the receiver operating characteristics curve (AUC) of 0.88 and 0.80, respectively.

Conclusions:

Anti-microbial antibodies have the potential to reduce the false positive rate of CT screening and provide interesting insight in lung cancer development.

Impact:

Microbial infection plays an important role in lung cancer development and the formation of benign pulmonary nodules.

Introduction

Lung cancer is the leading cause of cancer death among men and women ¹. Although risk factors like smoking are well established, lung cancer among never-smokers is the 6^th most common cause of cancer death in the United States and most smokers never get lung cancer ^2,3. Clearly, other factors contribute to carcinogenesis ^3–6. There is growing interest in a microbial link ^7–11. We still have limited understanding on which and how certain microorganisms cause or contribute to lung cancer development. Previous studies suggest Chlamydia pneumoniae, human immunodeficiency virus (HIV), human papilloma virus (HPV), Streptococcus pneumoniae, and Mycobacterium tuberculosis may promote lung cancer development. This is an area of active research as previous studies were limited by small sample sizes, observational study designs and inconsistent results. Culture-free next-generation sequencing has improved our understanding of compositional changes in various microbiota related to lung cancer ^12–14. It is now clear that the lung has distinct microbiota that can promote or prevent disease. Besides lung microbiota, oral and gut microbiota have also been implicated in lung cancer development ^15–18.

Microbial infection elicits complex host innate and adaptive immune responses ^19,20. The humoral immune response represents an important aspect of the adaptive immune response. The detection of anti-microbial antibodies in blood, the products of the humoral immune response, has frequently informed disease research. Unlike sequencing analysis, which provides information on the presence or absence of microorganisms limited to the time of sampling, the elicitation of antibody against a microbial protein demonstrates a host response to the microorganism and allows us to infer an “infection history”. Studies of antibody response against microorganisms have relied on whole virus preparations or crude bacterial lysates of usual microbial suspects, one microorganism at a time. Challenges associated with these traditional methods include low throughput, inadequate quantification, and poor reproducibility. Moreover, they lack the ability to compare responses to multiple organisms in the same sample. Additionally, antibodies to specific microbial antigens are more informative than the overall infection status to detect cancer ^21–24. Furthermore, the etiology for many epithelial cancers including lung cancer is most likely polymicrobial, involving cooperation of microorganisms that are non-pathogenic or weakly-pathogenic when acting alone, but able to initiate and promote tumorigenesis when acting together as a community. A study at the human microbiota level to understand the association of patterns of multiple-microorganism coinfection with lung cancer is especially relevant.

Antibody response to bacteria or viruses, such as Chlamydia pneumoniae ^25,26, Helicobacter pylori, HPV, and polyomaviruses, have been studied, but their associations with lung cancer remains elusive. To the best of our knowledge, there are no systematic studies on the antibody responses against viral and bacterial proteins in lung cancer patients. CT screening can detect lung cancer early, when the cancer is most treatable, and reduce mortality. However, only approximately 4 percent of patients with a positive finding of indeterminate pulmonary nodules (IPN) by CT scan were found to have lung cancer ²⁷. Most inflammatory lung nodules are due to microbial infections that are difficult to distinguish from lung cancer. A history of previous lung disease such as chronic obstructive pulmonary disease (COPD), tuberculosis and pneumonia has been associated with an increased risk of developing lung cancer and / or the formation of IPNs. Common to all of these lung diseases is a strong connection with microbial infections ^28,29. We hypothesized that just as ADC, BNC, and SMC have distinct etiologies, they will have distinct anti-microbial antibody profiles. ADC patients generate antibodies to a subset of the thousands of antigens encoded by these microorganisms in a cancer-specific manner, whereas BNC generate antibodies to a different subset of the thousands of antigens encoded by these microorganisms in a control-specific manner, providing an opportunity to rule out cancer. Using the Nucleic Acid Programmable Protein Array (NAPPA) ^30–32, we have profiled antibodies against about one thousand antigens from tens of different commensal and pathogenic bacteria and viruses, some of which have been reported to possess various degrees of potential association with lung cancer or respiratory diseases, whereas many are not previously known to be associated with lung cancer. We studied hundreds of lung adenocarcinoma patients (ADC), benign nodule controls (BNC) from CT screening, and smoker controls (SMC) samples.

Materials and methods

Patients and samples

All plasma samples were acquired from New York University with 127 from lung adenocarcinoma patients, 123 from gender matched smoker controls and 170 from benign nodule controls, including 47 emphysema, 50 granuloma and 70 stable nodule samples ³³ (Table 1). The ADC patients were recruited from the clinics at the NYU Cancer Center, and all patients gave written informed consent and the study was approved by institutional review board. The recruitment of high-risk smokers and benign nodule controls were also institutional review board approved.

Table 1.

Demographic information of 127 lung adenocarcinoma patients, 123 smoker controls and 170 benign nodule controls.

Characteristics	Lung adenocarcinoma	Smoker controls	Benign nodule control
			Emphysema	Granuloma	Stable nodule
N	127	123	47	50	73
Mean age	70 ± 10	64 ± 11	68 ± 8	63 ± 12	62 ± 10
Sex
Male	54	52	21	25	28
Female	73	70	25	25	45
No data		1	1
Smoking history
Never	11	0	1	8	2
Former	82	91	30	28	44
Current	20	31	15	8	26
No data	14	1	1	6	1
Mean pack years	31 ± 27	33 ± 24	50 ± 32	28 ± 47	30 ± 19
Nodules Size, cm	2.3 ± 1.4	0	0.4 ± 0.3	1.9 ± 1.6	0.6 ± 0.3
Stage
I	84
II	29
III	14

Microbial antigen array production and anti-microbial antibody profiling

Microbial antigen arrays were produced displaying 579 proteins from 27 bacteria and 322 proteins from 29 viruses, including both commensal and pathogenic microorganisms with an emphasis on those associated with respiratory tract infection (Supplementary Table S1) using the Nucleic Acid Programmable Protein Array (NAPPA) technology to assay anti-microbial antibodies in plasma samples ^30,31. Microbial genes in pANT7_cGST, an expression vector that allows gene transcription from the T7 protomer and protein expression using HeLa lysates based in vitro expression system, were ordered from the plasmid repository DNASU (DNASU.org). Microbial antigen NAPPA was produced by spotting plasmid DNA on silicon nanowell substrates as described ³². Antibody profiling was also performed as previously reported ³⁴. In brief, on the day of antibody profiling, c-terminal GST tagged microbial antigens were expressed using HeLa cell lysates-based in vitro expression system (ThermoFisher Cat# 88881) and in situ captured by co-spotted anti-GST antibody (GE Healthcare Cat# 27–4577-01) for their display on microbial antigen arrays. Plasma samples were randomized and applied on the microbial antigen arrays to reduce potential bias. Antibodies in plasma samples against displayed antigens were detected by anti-Human IgG (Jackson ImmunoResearch Labs Cat# 109–605-008) and anti-Human IgA (Jackson ImmunoResearch Labs Cat# 109–165-011) antibodies labelled with different fluorophores. Arrays were scanned, and the images were analyzed by the Array-Pro image analysis software (Media Cybernetics). Median fluorescence intensities at each spot were calculated for downstream data analysis.

Data analysis

Spot intensities on each array were normalized by dividing by the median spot intensity of the corresponding array before statistical analysis to minimize the effects of the overall background differences among samples. Seropositivity was determined using the empirical median normalized intensity cutoff 2 as previously reported ^21,22. We computed descriptive statistics for demographic and clinical variables. The 4^th quantile odds ratio (OR) of 2 comparison groups was computed for each antibody based on the maximum of the empirical technical cutoff value 2 and 75th percentile of all samples, and p-values were computed using the chi-square test. Volcano plots of negative logarithm base 10 of p-values versus logarithm base 2 of OR were generated for each comparison. Autoantibody seropositivity were determined using an empirical cutoff 1.5 for the ELISA data previously reported ³³. The proportions of antibodies having P values less than 0.05 and OR less than and greater than 1 were compared using two-sample proportion test.

Antibody panels were built using logistic regression of antibodies selected using the Maximum Relevance — Minimum Redundance (MRMR) feature selection algorithm ³⁵, and the classification performance was evaluated using the ROC generated with pROC package available for R (v4.1.0).

Bioinformatics analysis

Correlation between antibody reactivity was measured by the Spearman rank correlation coefficient. Correlation heatmaps were generated with Pandas (v1.3.4) and Seaborn (v9.11.2) packages available in Python (v3.8.10).

Data availability

All data required to reproduce the results may be requested from the corresponding author.

Results

Anti-microbial antibody profiling on microbial antigen arrays

We profiled plasma samples to identify anti-microbial antibodies against antigens from a set of representative microorganisms in 420 subjects, including ADC, BNC, and SMC (Table 1). Because of the important role of IgA in lung immunity, we profiled both IgG and IgA anti-microbial antibodies. Most microorganisms we included in this study had one or more antigens showing antibody response in more than 10% of the study population. Streptococcus pyogenes, Streptococcus pneumoniae, and Haemophilus influenzae elicited the strongest IgG antibody response among bacteria, while respiratory viruses, such as rhinoviruses, coronaviruses, influenza viruses and adenoviruses elicited the strongest IgG antibody response among viruses (Supplementary Fig. S1).

Microorganisms with higher IgG antibody reactivity also had higher IgA antibody reactivity, although IgA reactivity was usually lower than IgG for the same antigen (Supplementary Fig. S1). The median numbers of anti-bacterial IgG and IgA antibodies were 77 and 34 out of 579 possibilities, respectively (Supplementary Fig. S1). The medians for anti-viral IgG and IgA antibodies were 103 and 19 out of 336 possibilities, respectively (Supplementary Fig. S1). We focused our analysis on anti-bacterial antibodies because anti-viral antibodies showed similar reactivity with no clear trend of differences among ADC, BNC, and SMC (Table 2, Supplementary Table S2 and S3). We analyzed the correlation of the most reactive antibody of each microorganism among all subjects (Supplementary Fig. S2). We observed positive correlations between closely related bacteria and viruses for both IgG and IgA antibodies, such as Shigella flexneri and Klebsiella pneumoniae, and among the 4 seasonal human coronaviruses OC43, HKU1, 229E, and NL63. Correlations among microorganisms of different families or between viruses and bacteria were weak.

Table 2.

Prevalence of anti-bacterial antibodies with odds ratio p-values <0.05 in two comparison groups and their significance of differences.

Groups		Number of antibodies with higher prevalence in A	Number of antibodies with higher prevalence in B	Two sample proportion test P value	Number of different species
A	B
ADC	BNC	8	23	<0.001	9
ADC	BNC-E	6	28	<0.001	12
ADC	BNC-G	0	36	<0.001	12
ADC	BNC-SN	2	29	<0.001	13
SMC	BNC	3	23	<0.001	7
SMC	BNC-E	4	39	<0.001	9
SMC	BNC-G	1	45	<0.001	10
SMC	BNC-SN	4	29	<0.001	8
Light smokers with <=20 pack-years smoking history	Heavy smokers with > 20 pack-years smoking history	51	2	<0.001	12
Benign nodule size <= 1 cm	Benign nodule size >1 cm	0	40	<0.001	14
Stage I ADC	Stage II and III ADC	26	9	<0.001	12
Male	Female	37	12	<0.001	17

ADC: Lung adenocarcinoma; BNC: Benign nodule controls; SMC: Smoker controls; BNC-E: Emphysema; BNC-G: Granuloma; BNC-SN: Stable nodule.

Comparison between lung adenocarcinoma patients and benign nodule controls

We analyzed anti-bacterial antibodies showing differential reactivity between ADC and BNC and found more of such antibodies with higher prevalence among BNC than among ADC (Table 2). The p-values for the 4^th quartile odds ratios (OR) for 31 anti-bacterial antibodies were less than 0.05; 23 had higher seroprevalence in BNC and only 8 were higher in ADC (two sample proportion test p-value < 0.001) (Fig. 1A). When ADC was compared with each of the 3 subgroups of BNC individually, i.e., emphysema (BNC-E), granuloma (BNC-G), and stable nodule (BNC-SN), more antibodies had higher seroprevalences in each of the 3 BNC subgroups (Fig. 1B–D). The BNC-G subgroup and ADC had the greatest difference, 36 with higher prevalence in BNC-G vs. 0 in ADC (two sample proportion test p-value < 0.001). The 3 subgroups of BNC showed heterogeneity in antibody profiles. However, we did not observe overall differences of number of antibodies with higher seroprevalence in one BNC subgroup vs. the other (Supplementary Table S2).

Figure 1.

Fourth quartile odds ratio of anti-bacterial antibodies vs. P values for comparisons between A) ADC vs. BNC; B) ADC vs. BNC-E; C) ADC vs. BNC-G; D) ADC vs. BNC-SN; E) SMC vs. BNC; F) SMC vs. BNC-E; G) SMC vs. BNC-G; and H) SMC vs. BNC-SN. Dotted lines indicate p-value = 0.05. Antibodies above the dotted lines had OR p-values < 0.05. Antibodies against Helicobacter pylori, Pseudomonas aeruginosa and Streptococcus spp. antigens are show in red triangle, blue square, and inverted green triangle, respectively. The annotation “X > Y Z Abs” on each plot indicates that Z anti-bacterial antibodies with OR p-values < 0.05 between X and Y groups had higher seroprevalence in X than in Y.

ADC: Lung adenocarcinoma; BNC: Benign nodule controls; SMC: Smoker controls; BNC-E: Emphysema; BNC-G: Granuloma; BNC-SN: Stable nodule.

Comparison between smoker controls and benign nodule controls

BNC had more anti-bacterial antibodies than SMC (Table 2). When SMC was compared with BNC as a group, 26 anti-bacterial antibodies had OR p-values less than 0.05 with 23 higher in BNC and 3 higher in SMC (two sample proportion test p-value < 0.001) (Fig. 1E). When SMC was compared with each of the 3 subgroups of the BNC individually, more antibodies had higher reactivity in all 3 subgroups (Fig. 1F–H). The BNC-G subgroup also had the greatest difference with SMC (45 vs. 1, p-value < 0.001).

Comparison between lung adenocarcinoma patients and smoker controls

ADC and SMC had similar numbers of anti-bacterial antibodies. When ADC was compared with SMC, 39 anti-bacterial antibodies had the 4^th quartile OR p-value smaller than 0.05 (Fig. 2A). Out of these 39 antibodies, the number of antibodies with higher seroprevalences in ADC was similar to that higher in SMC (21 vs. 18, two sample proportion test p-value = 0.650) (Supplementary Table S2). Among them, anti-H. pylori antibodies were significantly enriched in antibodies with significantly higher prevalence in ADC compared with SMC (11 higher vs. 0 lower, two sample proportion test p-value < 0.001) (Fig. 2A). This was not observed for anti-Streptococcus spp. antibodies, which showed similar numbers of antibodies with significantly higher or lower prevalence in ADC relative to SMC (8 higher vs. 6 lower, two sample proportion test p-value = 1).

Figure 2.

Anti-H. pylori antibodies with differential prevalence between lung cancer patients and high-risk controls. A. Fourth quartile odds ratio of anti-microbial antibodies between lung cancer patients and high-risk controls vs. P values. Antibodies are colored to indicate source bacteria of target antigens. Antibodies against H. pylori and Streptococcus spp. are shown in orange triangle and inverted blue triangle respectively. The dotted line indicates p-value = 0.05. B. Dot plots for selected anti-H. Pylori antibodies. Anti-HP1341_IgG and IgA had the highest reactivity among all anti-H. pylori antibodies. Anti-HP0596_IgA, Anti-HP0477_IgA, Anti-HP0923_IgA are 3 examples of the anti-H. pylori antibodies showing significantly higher prevalence in lung adenocarcinoma patients (ADC) vs. smoker controls (SMC). Solid lines indicated seropositivity cutoff 2 of median normalized intensity.

We studied antibodies against 233 H. pylori proteins. IgG and IgA antibodies to H. pylori protein HP1341 had the highest reactivity among both ADC and SMC; however, neither showed a differential prevalence between ADC and SMC (Fig. 2B). Anti-H. pylori antibodies showing differences, such as anti-HP0596, anti-HP0923 and anti-HP0477, had lower overall reactivity compared with that for anti-HP1341 (Fig. 2B). Anti-H. pylori antibodies were also enriched in antibodies showing higher prevalence in BNC-G compared with ADC (Fig. 1C) or SMC (Fig. 1G). Similar to what we observed for antibodies showing significant differences in prevalence between ADC and SMC, anti-HP1341 had the highest overall IgG and IgA reactivity but did not show differences between BNC-G and ADC or SMC.

Association of anti-bacterial antibodies with clinical parameters

We performed sub-group analysis based on clinical parameters other than diagnosis (Table 2 and Supplementary Table S2). We were interested in the effects of smoking on the anti-bacterial antibody profiles. Because most of our 420 study subjects were smokers, we compared light smokers, subjects with ≤ 20 pack-years smoking history, and heavy smokers, subjects with >20 pack-years smoking history. We observed more anti-bacterial antibodies with higher seroprevalence in light smokers relative to heavy smokers (51 vs. 2, two sample proportion test p-value < 0.001, Supplementary Fig. S3A). For BNC, we observed lower reactivity for subjects with ≤ 1 cm nodules than those with > 1 cm nodules (0 vs. 40, two sample proportion test p-value < 0.001, Supplementary Fig. S3B). However, we did not observe differences comparing ADC with small nodules (≤ 3 cm) and ADC with large nodules (> 3 cm) (7 vs. 10, two sample proportion test p-value = 0.492) (Supplementary Table S2). We used different cutoffs of the nodule size when analyzing the benign and the case groups because the case group had overall larger nodules than the benign group. Slightly higher anti-bacterial reactivity was observed for ADC of stage I than for stages II or III (26 vs. 9, two sample proportion test p-value < 0.001, Supplementary Fig. S3C).

Correlation of anti-microbial antibodies and autoantibodies

Target antigens for many cancer-specific antibodies play an important role in cancer development ^36–38. We previously assayed autoantibodies in the same set of samples by ELISA ³³. We did not observe significant correlations with any anti-bacterial or anti-viral antibodies. p53 plays an important role in the development of many cancers, including lung cancer. Cancers with anti-p53 antibodies may have different pathogenesis from cancers related to microbial pathogens. Patients with anti-p53 antibodies usually carry p53 mutations, though we do not have the p53 mutation status for our study population. Using anti-p53 seropositivity as a surrogate, we compared anti-bacterial antibodies between anti-p53 positive and anti-p53 negative ADC. Among the 9 antibodies showing significantly higher prevalence in anti-p53 negative ADC, 6 were against Streptococcus spp. and 2 were against Haemophilus influenzae (Fig. 3).

Figure 3.

Comparison of anti-bacterial antibodies between anti-p53 positive and negative ADC patients. A, Heatmap showing antibodies with significantly higher prevalence in 101 anti-p53 negative ADC patients relative to 26 anti-p53 positive ADC patients. B, Dot plot for anti-p53 positive and negative ADC patients based on anti-p53 ELISA. The solid red line indicates the ELISA normalized O.D.450 seropositivity cutoff. C, Dot plots for anti-microbial antibodies against Haemophilus influenzae and Streptococcus spp. antigens between anti-p53 positive and negative ADC patients. Solid red lines indicate seropositivity cutoff 2 of median normalized intensity.

Panel analysis using Maximum Relevance - Minimum Redundance (MRMR) logistic regression

We have constructed antibody panels to distinguish ADC, SMC and BNC using anti-microbial antibodies with significantly different prevalence among these groups. We used MRMR logistic regression to build 20-antibody panels with area under the receiver operating characteristics curve (AUC) of 0.80 (95% confidence interval: 0.75–0.85), 0.88 (95% confidence interval: 0.85–0.92), and 0.84 (95% confidence interval: 0.79–0.88) distinguishing ADC vs. BNC, ADC vs. SMC, and BNC vs. SMC respectively (Fig. 4). The set of twenty antibodies used for each comparison is listed in Supplementary Table S4. When ADC and SMC were compared with the 3 BNC subgroups separately, anti-microbial antibodies could distinguish both ADC and SMC from BNC-E (Supplementary Fig. S4A and S4D) and BNC-G (Supplementary Fig. S4B and S4E) better than BNC-BN (Supplementary Fig. S4C and S4F). The antibodies selected by MRMR algorithm is independent of the various comparisons performed in Table 2. Overall, anti-microbial antibodies could distinguish SMC from all 3 BNC subgroups better than ADC (Supplementary Fig. S4).

Figure 4.

Receiver Operating Characteristics (ROC) analysis for the antibody panels built using the logistic regression model with 20 antibodies selected using the MRMR method between A) ADC vs. BNC; B) ADC vs. SMC; and C) BNC vs. SMC.

Discussion

Cancers attributable to infections have a greater incidence than any individual type of cancer worldwide ³⁹. Given the profound exposure of the lungs to microbes in the environment, as well as a growing body of evidence for a unique lung microbiome, we sought to understand how a history of microbial infection, as evidenced by specific anti-microbial immune responses, relates to lung cancer. Moreover, many causes of IPN relate to prior or concurrent microbial infection. These infections may lead to specific immune response signatures that can distinguish between malignant and benign IPNs. Thus, we have performed the largest survey on anti-microbial antibodies in lung cancer. The microorganisms and their antigens were selected from our microbial protein collection (DNASU.org) based on preliminary studies in our laboratory and a review of the literature. For antibodies showing significant ORs among the 3 groups, we observed enrichment of source microorganisms for the target antigens with previously suspected or novel connections with lung diseases or lung cancer. Besides disease diagnosis, we also discovered that cancer genetics, smoking history and nodule sizes affected anti-microbial antibody prevalence. Using MRMR logistic regression, we built an antibody panel with an AUC of 0.80 distinguishing ADC and BNC with 43% sensitivity at 90% specificity and 46% specificity at 90% sensitivity.

Most microorganisms had one or more antibodies showing high reactivity among most samples against their antigens displayed on the arrays. Interestingly, highly reactive antibodies or those with the highest prevalence for a microorganism generally did not show differential reactivity among different subject groups. On the contrary, antibodies with moderate reactivity/prevalence showed more differential reactivity among different subject groups. We have observed similar results for infections in other diseases ^21–24. The ability of these specific antigen responses to distinguish subject groups speaks for the importance of assaying antibodies against individual proteins, especially proteins that are not the dominant antigens of the tested microbes. These results are consistent with our previous finding that antibodies against antigens of the same microorganism displayed different trends of reactivity in healthy controls, gastric cancer (GC), and subjects with intestinal metaplasia ^21,22. It is not immediately apparent why antigens from the same microorganism showed differential reactivity. The physiological conditions existing in the infected tissues of subjects in different health/disease states can modulate microbial antigen expression profiles, and the immunological microenvironment at the infection sites also affects humoral immune response to expressed antigens.

Besides exposure to smoke or other hazardous chemicals, microbial infection in the lung is the main cause of pulmonary diseases that can lead to the formation of nodules. Some of these nodules would remain benign, but some would progress into malignant nodules. Around one-fifth of the cancer are caused by microbes. Lungs are one of the organs which get exposed to the environmental microbes the most due to continuous breathing. Similar to smoking, microbial infection can cause chronic inflammation and mutation that can lead to cancer. Our results support the associations of bacterial, and to a lesser degree, viral infections with the development of benign pulmonary nodules. Our data showed overall higher prevalence of many anti-microbial antibodies in each of the 3 BNC subgroups compared with either ADC or SMC (Table 2). There can be several explanations why ADC patients had lower antibody prevalence relative to BNC. One possibility is that the microbial infections that caused benign nodules contribute less to cancer development. Alternatively, it is also possible that some of these microorganisms may contribute to benign nodules and cancer initiation but play a lesser role in the progression and maintenance of cancer. It is interesting to note the granuloma, whose etiology has the strongest known association with microbial infection, showed strongest anti-microbial antibody reactivity among 3 subgroups of BNC ^40,41. Future experiments are warranted to better understand these connections.

Antibodies against H. pylori, a known carcinogenic microbe, showed greatest overall trend differences among ADC, SMC, and BNC. H. pylori is one of the best studied bacteria because of its etiological role in many digestives and extra-digestive diseases, especially in GC ⁴². H. pylori has also been associated with many respiratory disorders, including COPD, bronchiectasis, asthma, tuberculosis, and lung cancer ^42–49. However, results from previous sero-epidemiological studies were inconsistent ^50–55. Our data support the hypothetical role of H. pylori infection in ADC. Two possible carcinogenic mechanisms have been proposed for how H. pylori infection might contribute to lung cancer development. First, the lungs originate embryologically from the same endoderm cells which form the epithelia lining of the digestive tract, where gastrin provides the major proliferative stimulus ^56–59. H. pylori infection in the stomach results in the enhanced and prolonged release of gastrin in circulation, which also stimulates the proliferation of bronchial epithelium. In addition, immune effector proteins such as cytokines produced from native or adaptive immune response to H. pylori infection in the stomach may also enter the circulation causing a stimulatory effect on lung cell proliferation. Second, H. pylori infection in the lung could cause direct damage and chronic airway inflammation that promote lung cancer development ⁶⁰. However, H. pylori has not yet been detected in human bronchial tissue or isolated from bronchoalveolar lavage fluid. Furthermore, even for GC, only an extremely small percentage of people infected with H. pylori develop GC. It is believed that co-infection with additional microorganisms, besides genetic and other environmental factors, contributes to cancer development. However, for the set of microbial antigens displayed on our arrays, none of their antibodies had significant positive or negative correlation with anti-H. pylori antibodies. Future studies with more antigens and organisms may shed light on the polymicrobial nature of the pathogenesis of H. pylori infection in lung cancer.

Many factors including disease stages, disease grades, genetic background, microbial antigen expression levels, local and global immune environment contribute to the heterogeneity in antibody response. Among them, smoking is a major contributing factor. Our results (Supplementary Fig. S3A) suggest that the role microbial infection plays and the level of its importance in lung disease pathology might be different in light and heavy smokers. It is plausible that microbial infection might play a more important role in disease etiology in light smokers than in heavy smokers. Alternatively, smoking may impair lung immunity and dampen antibody response to microbial infection to a greater extent in heavy smokers than in light smokers. Similar to smoking habits, early disease stage (stage I ADC) led to higher antibody prevalence compared to later disease stage (stage II, III ADC). This observation might be due to the reduction in immunity caused by the progression of disease.

We have used the same set of samples to study tumor associated autoantibodies (TAAb) and anti-microbial antibodies. We did not observe significant correlation between autoantibodies and anti-microbial antibodies. TAAb against p53 is probably the most studied autoantibody in cancer. One prerequisite for eliciting anti-p53 is the over-expression of p53 due to its mutation. The presence or abscence of a p53 mutation leads to different disease etiologies. For example, EBV-associated GC had lower prevalence of anti-p53 antibody relative to GC not associated with EBV ⁶¹. We do not have the p53 mutation status for our study population. Using anti-p53 seropositivity as a surrogate (i.e, there was a high likelihood of p53 mutation for anti-p53 positive individuals), we observed anti-Streptococcus spp. antibodies having strikingly higher prevalance in anti-p53 negative cases than anti-p53 positive cases. This suggests that Streptococcus spp. and p53 mutation may play mutually exclusive roles in lung cancer development.

The strength of this study lies in the large number of samples profiled against hundreds of bacterial and viral antigens. Limitations include the absence of some microorganisms with known association with lung cancer like Chlamydia pneumonia, human immunodeficiency virus and human papilloma virus. No antigens from fungi were available on our array, although fungal infection may contribute to benign nodule development, especially granulomas ⁶². Our study focused on ADC patients, which is the most common type of non-small cell lung cancer. This had the advantage of providing a more homogeneous group for a first study, but the disadvantage that we are missing important information on other common forms of lung cancer. In future studies, patients with squamous cell carcinoma and large cell lung carcinoma will provide us a holistic picture of lung cancer and its associated anti-microbial antibodies. Also, a limited number of antigens from each microorganism were profiled on serum collected from a particular geographical location. It will be interesting to study how regional microbes that can cause chronic infections change the anti-microbial antibody response. Lack of antibiotics usage information in patients is another limitation of this study. The microarray used in this study is a robust platform to discover non-invasive biomarkers. However, one disadvantage of microarray lies in the need to clone microbial genes of interest in an expression vector, thus limiting the searching space. Furthermore, some pathogens may not elicit a strong antibody response but a strong cellular response. Therefore, we believe that an integrated approach of imaging and molecular biomarkers will be the best way for early diagnosis of lung cancer.

The effect of the microbiome on the host immune response in lung or at distal sites is important for lung cancer initiation and progression. Anti-microbial antibodies reveal one important aspect of host adaptive immune response to microbial infection and provides a starting point for further investigation of the link between the microorganism and cancer. These antibodies can serve as biomarkers that can distinguish between benign and malignant nodules, thereby reducing false positive rate of CT imaging ^63,64. We believe our data-driven immunoproteomics study, together with (meta)genomics and other approaches, can be helpful to decipher the complex relationship between microorganisms and lung cancer development.