Relevance of pepsinogen, gastrin, and endoscopic atrophy in the diagnosis of autoimmune gastritis

Simple objective modalities are required for evaluating suspected autoimmune gastritis (AIG). This cross-sectional study aimed to examine whether pepsinogen, gastrin, and endoscopic findings can predict AIG. The diagnostic performance of endoscopic findings and serology in distinguishing AIG was evaluated. AIG was diagnosed in patients (N = 31) with anti-parietal cell antibody and/or intrinsic factor antibody positivity and histological findings consistent with AIG. Non-AIG patients (N = 301) were seronegative for anti-parietal cell antibodies. Receiver operating characteristic curve analysis of the entire cohort (N = 332) identified an endoscopic atrophic grade cutoff point of O3 on the Kimura–Takemoto classification (area under the curve [AUC]: 0.909), while those of pepsinogen-I, I/II ratio, and gastrin were 20.1 ng/mL (AUC: 0.932), 1.8 (AUC: 0.913), and 355 pg/mL (AUC: 0.912), respectively. In severe atrophy cases (≥ O3, N = 58, AIG/control; 27/31), the cutoff values of pepsinogen-I, I/II ratio, and gastrin were 9.8 ng/mL (AUC: 0.895), 1.8 (AUC: 0.86), and 355 pg/mL (AUC: 0.897), respectively. In conclusion, endoscopic atrophy is a predictor of AIG. High serum gastrin and low pepsinogen-I and I/II ratio are predictors even in the case of severe atrophy, suggesting their usefulness when the diagnosis of AIG is difficult or as serological screening tests.


Results
Baseline characteristics of the whole study population. As shown in Fig. 1 13 initially presented to our hospital, and the remaining 18 were already being followed up at our hospital. The reasons for performing endoscopy were as follows: positive results in the ABC method (n = 14, 45.2%), abnormalities on barium X-ray (n = 5, 16.1%), annual endoscopic follow-up (n = 3, 9.7%), screening in asymptomatic individuals (n = 6, 19.3%), and others (n = 3, 9.7%).
Age, sex, serum biomarker levels (PG and gastrin), Helicobacter pylori (H. pylori) serology, and autoantibodies were evaluated for all patients, as shown in Table 1. Supplementary Table S1 shows the detailed histological findings in patients with AIG (according to our definition). Oxyntic mucosal atrophy and lymphocyte and plasma cell infiltration were observed in all cases. Supplementary Table S2 presents the observed Kappa values among the three endoscopists. As a Kappa value of ≥ 0.60 is usually considered reliable and applicable for evaluation, the "endoscopic atrophic border" (mean, 0.654) and disappearance of the gastric fold (mean, 0.622) were adapted as parameters to be analyzed in this study. However, three other endoscopic findings, including corpus-predominant advanced atrophy compared to the antral part (mean, 0.335), sticky mucosa (mean, 0.352), and hyperplastic polyp (mean, 0.516), were excluded from the analysis as endoscopic findings generally are not highly reliable.

Discussion
In this study, we evaluated the relevance of several clinical findings in AIG diagnosis among (1) patients diagnosed by autoantibody positivity and characteristic histology and (2) non-AIG patients diagnosed based on autoantibody negativity whose H. pylori infection status was precisely evaluated. The ROC curve analysis results of the entire study population (N = 332) indicated that the diagnostic performance of advanced endoscopic atrophy (≥ O3 in the Kimura-Takemoto classification) and serology, including PG (PG I, ≤ 20.1 ng/mL; PG I/II ratio, ≤ 1.8) and gastrin (≥ 355 pg/mL), was high; therefore, they have sufficient diagnostic performance for use in daily clinical practice.  www.nature.com/scientificreports/ As our results showed that advanced endoscopic atrophy itself may be a marker of AIG, it may be useful if highly suspected cases of AIG among patients with advanced atrophy can be further evaluated by serology (i.e., two-stage identification). Based on this idea, an ROC curve analysis of patients with severe gastric atrophy only (≥ O3, N = 58) was performed. Our results indicated that the diagnostic performance of PG (PG I, ≤ 9.8 ng/ mL; PG I/II ratio, ≤ 1.8) and gastrin (≥ 355 pg/mL) was also high (AUROC, > 0.85) and that they may be valid predictors. These findings are especially useful when severe atrophy is found during endoscopy, which may be the commonest way of identifying AIG in clinical practice.
In our study, gastrin (355 pg/mL) and PG I/II ratio (1.8) exhibited the same cutoff values for the prediction of AIG in both the overall study cohort and the subset of patients with advanced atrophy. These results highlight how they may be especially useful as serum markers in clinical practice. Considering the high AUROC of these serological markers, we advocate that PG and gastrin should be additionally checked in cases where the diagnosis is challenging, for example, when pathological examination of biopsy specimens does not show findings typical of AIG despite a positive APCA result. Re-biopsy and histological evaluation may be recommended if PG or gastrin levels are strongly suggestive of AIG. Moreover, in serological screening tests, these cutoff values are useful in detecting AIG; in this study; endoscopy was indicated in approximately half (45.2%) of the AIG cases based on a positive serology test using the ABC method.
The endoscopic atrophic grade (Kimura-Takemoto classification) is considered highly objective by several investigators 24,25 , and this atrophy grade was imposed for all upper gastrointestinal endoscopic cases in the Japan Endoscopy Database Project 26 . Although O4 is useful in detecting AIG of high prevalence compared with non-AIG (64.5% vs. 3.0%), new endoscopic criteria are not necessary for AIG diagnosis as it is included as part of grade O3 in the Kimura-Takemoto classification. In AIG diagnosis, endoscopy is often performed merely to take biopsies for histopathological diagnosis with low specificity and objectivity 3,10,27 . However, herein, we demonstrated that the "endoscopic atrophic border" had a Kappa value of 0.654. This is consistent with the report by Terao et al. 17 , wherein high interobserver variability (0.705) was shown in the "degree of atrophy;" however, in the case of "sticky mucosa, " our Kappa value seems to be lower than their value (0.352 vs. 0.604). Although corpus-predominant advanced atrophy compared with the antral region, which we referred to as the "corpuspredominant atrophic pattern," has been used as a diagnostic criterion 15,16 , it did not have a reliable Kappa value in our study. Regarding the disappearance of the gastric fold in the gastric corpus, it cannot be used as a diagnostic modality because it has a low AUROC value (0.773), even though its Kappa value was > 0.6. Hence, other objective endoscopic findings should be evaluated.
In this study, the cutoff values of gastrin for the diagnosis of AIG obtained from the ROC analysis (355 pg/ mL) were higher than those previously reported by Checchi et al. 28 . The cutoff value of gastrin for diagnosing AIG with severe atrophy was reported as 395 pg/mL 11 , and Terao et al. 17 set hypergastrinemia > 350 pg/mL as one of the criteria for AIG in a multicenter study, which is similar to the 355 pg/mL cutoff value for gastrin in this study. In Checchi et al. 's study, the cutoff value of gastrin (43 pg/mL) was close to normal (< 39.3 pg/mL), suggesting the possible diagnosis of individuals without AIG as having AIG 28 . Simple comparisons with other reports are difficult because of differences in the gastrin unit of measurement (pg/mL and pmol/L) 10,29 .
Regarding PGs, various cutoff values have been reported by several investigators; however, few reports involved an ROC analysis using a histology-based definition of AIG 10,28,30 . The cutoff values for PG I (20.1 ng/ mL) and PG I/II ratio (1.8) in our study were similar to those reported by Koc et al. 30 , but significantly lower than those in the studies by Venerito et al. 10  their cutoff values may be set higher than the actual physiological values. The evaluations of the cutoff value of PGs to predict AIG in other studies were inherently limited by various issues. A report evaluated the ROC curve of PGs in appropriately defined patients with AIG, but did not calculate the cutoff value 31 . Three more studies did not analyze the cutoff value in a strictly defined AIG cohort, including participants with multifocal atrophy 32 , pangastritis 29 , or severe atrophy of Operative Link on Gastric Intestinal Metaplasia Assessment > stage 2 and autoantibody positivity 15 . Because Koc et al. only analyzed a limited number of patients (N = 16) with AIG, this is the first report to demonstrate the practical cutoff value of PG in the diagnosis of AIG using a sufficient cohort of strictly defined patients with AIG.
In this study, individuals on acid inhibitors were strictly excluded; however, in clinical practice, the possible effects of acid inhibitory drugs should be considered when using these cutoff values. Hypergastrinemia characterized by gastrin levels over 500 pg/mL, which is induced by PPIs or potassium competitive acid blockers (PCABs) 33,34 , can be a major problem in serological screening for AIG. However, it is unclear whether eradication influences gastrin in patients with AIG because severe hypochlorhydria is induced by AIG in many cases, and eradication may not affect it. As for PG and gastrin, these would theoretically normalize after eradication; however, because of the extremely low PG level and extremely high gastrin level in AIG, eradication treatment does not seem to affect the results of serological screening.
In our study, the median value (IQR) of gastrin was 1,310 (448-2,490) pg/mL, and the PG I level and PG I/II ratio were 6.4 (3.8-15.9) ng/mL and 1.0 (0.6-1.5), respectively, suggesting that the clinical characteristics of AIG in our study were generally comparable to those in previous studies 10,17,27,35 . However, 45.2% of patients with AIG in our study had H. pylori infection (19.4% and 25.8% of present and previous infection cases, respectively), an incidence higher than those in previous reports of seropositivity rates (Terao et al. 17 , 7.8%; Venerito et al. 10 , 18.2% [4/22]). In our study, 29/31 patients with AIG (94%) were accurately evaluated by more than two modalities to establish H. pylori infection, and eradication was diagnosed by normalization of the urea breath test, which may explain the high infection rate in our study.
The prevalence of AIG (5.8%, 31/536) in this study is higher than the reported overall prevalence in Japan (0.49%) 16 . Eighteen patients with AIG, who were included in the analysis, were already being followed-up at our www.nature.com/scientificreports/ hospital with regular endoscopy. Thus, the prevalence of first-visit cases was 13/536 (2.4%), which is still higher than the national prevalence rate. However, we believe that this is reasonable given how rare cases tend to cluster at major referral hospitals, such as ours. This study has some limitations. First, the identification of patients with AIG was based only on autoantibodies and histological findings. In this study, 2/31 histologically defined AIG cases (6.4%) had normal gastrin levels, suggesting that false positive cases were inevitably included (although there is a possibility that these were very early stage AIG cases), and that additional criteria using serological markers, such as gastrin and/or PG, may be required to increase the accuracy of the diagnosis of AIG, as described earlier. Second, other endoscopic findings, including remnant oxyntic mucosa, scattered minute whitish protrusions, and intestinal metaplasia were not evaluated. They may be potential predictive factors of AIG and should, therefore, be evaluated. Furthermore, the small sample size of patients from a single institution may limit the generalizability of the obtained results.
In conclusion, in this study, we demonstrated that endoscopic findings of severe atrophy (≥ O3 in the Kimura-Takemoto classification) are useful predictors of AIG. High serum gastrin levels and low PG I and I/ II ratio are also predictors of AIG. The cutoff values of serum gastrin levels (≥ 355 pg/mL) and low PG I/II ratio (≤ 1.8) can be used in both the general population and those with endoscopic atrophy (≥ O3). Thus, patients with advanced atrophy with a strong suspicion of AIG can be further evaluated using these cutoff values. Clinicians should also note that AIG can be suspected by serology, especially when false negative or positive results are suspected or during serological screening for gastritis involving PG and/or gastrin.

Study design and participants.
In this cross-sectional study, 536 consecutive patients who underwent upper gastrointestinal endoscopy at Tokyo Dental College Ichikawa General Hospital between January 2017 and March 2020 were enrolled after obtaining written informed consent. The exclusion criteria for this study were as follows: (1) use of histamine-2 receptor antagonists, PPIs, or PCABs within the preceding 2 months; (2) presence of viral diseases; (3) pregnancy or lactation; (4) presence of renal and/or liver dysfunction; or (5) a past history of gastric cancer or any type of esophageal or gastric surgery 36 . Most of the excluded individuals were excluded by interview before enrollment.
A flowchart of the target group selection is shown in Fig. 1. "AIG" was defined based on both (1) seropositivity for APCA and/or AIFA, and (2) histological findings consistent with AIG. Patients positive for either APCA or AIFA were classified into the following two subgroups: patients in whom biopsy of the gastric body was performed, and those without biopsy. The latter were excluded from the analysis. For AIG diagnosis, in the case of APCA positivity only, at least three of the five histological findings compatible with AIG, as described in the "Histology" section, were necessary. In the case of AIFA positivity, irrespective of APCA, at least two of the five histological findings were necessary because the sensitivity of AIFA in diagnosing AIG is considerably higher than that of APCA 8 . Non-AIG controls were defined as individuals who tested negative for APCA, and 57 were excluded due to inadequate APCA or gastrin data, 13 due to PPI treatment, and 3 due to history of gastric cancer.
H. pylori infection status in the control participants was defined as follows: Patients who were seronegative for APCA and with a clear history of successful eradication, confirmed by a negative result on either a urea breath or stool antigen test, were defined as "H. pylori-eradicated patients. " In contrast, those with a result < 2.5% in the urea breath test were classified as negative according to the manufacturer's recommended cutoff value. Patients negative for APCA and with a history of eradication were further classified into two subgroups: patients positive (≥ 10 U/mL) and negative (< 10 U/ml) for H. pylori antibody titer. The former was defined as "present H. pylori infection cases. " The latter individuals with an atrophy grade from C0 to C1, which is regarded as an atrophy score of 0 (negative) in the Kyoto classification 37 , were defined as "H. pylori-uninfected cases. " The patient group negative for H. pylori antibody (< 10 U/mL) and endoscopic atrophy (≥ C-2) may include those who are H. pyloriinfected and those with previous infection-induced atrophic gastritis 36,38,39 . Among these patients, those with at least one positive urea breath, stool antigen, or culture test were classified as "present H. pylori infection cases, " whereas previous infection-induced atrophic gastritis cases were defined when negative results were obtained in urea breath or stool antigen tests, as we reported previously 40 . If a urea breath or stool antigen test was not performed, individuals with an eradication history and those with endoscopic atrophy (≥ C2) and seronegative results of H. pylori were excluded.
H. pylori infection status was defined differently in patients with AIG because of the extreme case bias. Those with at least one positive urea breath, stool antigen, H. pylori antibody titer, or culture test were classified as "present H. pylori infection cases". The cutoff value for a positive diagnosis in the urea breath test was set at 5% instead of 2.5% 41 because false-positive results tend to be obtained in AIG. Patients with an apparent history of successful eradication, confirmed by a negative result in either a urea breath or stool antigen test, were defined as "H. pylori-eradicated patients." The cutoff value for a negative result in these urea breath tests was strictly 2.5%. It is difficult to differentiate between H. pylori-uninfected individuals and those with a previous infection among patients with AIG; thus, for simplicity, patients other than those with a present infection or history of eradication were defined as H. pylori-uninfected. www.nature.com/scientificreports/ To evaluate the diagnostic validity and determine the threshold values of the serological tests and endoscopic findings for predicting AIG, an ROC curve analysis was performed.

Measurement of serum biomarkers.
Fasting serum samples (after an overnight fast) were collected before endoscopy. PG, gastrin, H. pylori antibody titers, and APCA were examined in the enrolled patients. Briefly, serum PG I and PG II levels were measured using a commercial chemiluminescence enzyme immunoassay kit (Architect Pepsinogen I, Pepsinogen II, Abbott Japan Co. Ltd., Tokyo, Japan); H. pylori IgG antibodies were measured using an enzyme immunoassay kit (E Plate "Eiken" H. pylori antibody, Eiken Chemical Co. Ltd., Tokyo, Japan); and gastrin level was determined with radioimmunoassay (Gastrin RIA Kit II, Fujirebio Diagnostics Co., Ltd., Tokyo, Japan), as previously reported 36 . In patients clinically suspected of having AIG (e.g., severe atrophy of ≥ O3 according to the Kimura-Takemoto classification or APCA positivity), thyroglobulin antibodies, thyroid microsomal antibodies, serum vitamin B12, and AIFA were additionally measured. APCA and AIFA measurements were outsourced to SRL Co. Ltd. (Tokyo, Japan), and thyroglobulin and thyroid microsomal antibody measurements were performed by LSI Medience Co., Ltd. (Tokyo, Japan). Anemia was defined as hemoglobin < 13 g/dL (men) or 11.4 g/dL (women); vitamin B12 level < 233 pg/mL was considered low; and pernicious anemia was defined as vitamin B12 < 233 pg/mL, mean corpuscular volume (MCV) > 80 fL, and hemoglobin < 13 g/dL (men) or 11.4 g/dL (women) 17 . Hashimoto's disease was defined as positivity for thyroglobulin antibodies or thyroid microsomal antibodies. APCA and AIFA were analyzed using a fluorescent antibody test and chemiluminescence enzyme immunoassay, respectively. APCA was considered positive when fluorescence was produced at a dilution of tenfold or more according to the manufacturer's instruction. Although there is limited information on the sensitivity of APCA, a study reported that 33.59% of APCA-positive cases were diagnosed as AIG 7 .
Histology. Patients positive for either APCA or AIFA with available gastric corpus biopsies were histologically evaluated by an experienced pathologist (AS) who was blinded to the patients' endoscopic and serological information and medical charts. Biopsy specimens were fixed with buffered formalin and stained using hematoxylin and eosin. The following five findings were evaluated to diagnose AIG based on previous reports 3, 9-14 : (1) oxyntic mucosal atrophy, (2) pseudo-pyloric and intestinal metaplasia, (3) diffuse lymphocytic cell infiltration, (4) proportion of gastric pits/duct in the gastric mucosa (> 1.0 or none), and (5) enterochromaffin-like cell hyperplasia. Typical histological findings are shown in Fig. 3a-f. Endoscopic examination. Upper gastrointestinal endoscopy was performed using an Olympus Elite system and Olympus electronic panendoscopes, including the GIF-HQ290 or GIF-HQ290Z (Olympus, Optical Co. Ltd, Tokyo, Japan). Based on previous reports 13,14,16,17 , the following five endoscopic findings considered specific for AIG were evaluated individually: (1) endoscopic atrophic grade, (2) characteristic corpus-predominant atrophic pattern (corpus-predominant advanced atrophy compared with the antral part), (3) disappearance of the fold in the greater curvature of the gastric body in the sufflation state, (4) presence of sticky adherent dense mucous, and (5) hyperplastic polyps. Sticky adherent dense mucous has a denser, creamy white-yellowish color and firmly adheres to the mucosa, as defined by Terao et al. 17 . The endoscopic atrophic grade was defined according to the Kimura-Takemoto classification 42 , which categorized gastric mucosal atrophy into closed (C1-3) and open types (O1-3). We added "O4" for severe endoscopic atrophic corpus gastritis, which was defined as marked vascular visibility observed not only on the lesser curvature but also on the whole greater curvature of the corpus. This is included in the O3 category of the Kimura-Takemoto classification and is similar to the definition provided by Terao et al. 17 ; however, it was not associated with the disappearance of the fold according to our definition. Figures 4a-d depict typical endoscopic findings of O4 (Fig. 4a), O3 (Fig. 4b), O2 (Fig. 4c), and O1 (Fig. 4d). Figure 4e shows the endoscopic findings of the corpus-predominant atrophic and non-atrophic patterns in the antrum (Fig. 4e-1), despite severe corpus atrophy ( Fig. 4e-2). Figures 4a, 4b, and 4e-2 show the disappearance of the fold. Figure 4f shows sticky adherent dense mucous, and Fig. 4g shows multiple hyperplastic polyps.
Three experienced endoscopists (KN, YH, and HH), blinded to the clinical, serological, and histological status of the patients, independently evaluated the five endoscopic images and determined the final endoscopic findings by consensus.
Statistical analyses. Serum gastrin and PG values are presented as medians (IQR), and statistical analyses were performed using nonparametric tests. The Mann-Whitney U test was used to determine the significance of the differences between patients with and without AIG. The χ 2 test or Fisher's exact test was performed to analyze categorical variables. Interobserver agreement between the diagnosis of endoscopic findings was analyzed using Kappa values. The strength of agreement was determined based on the guidelines by Landis and Koch 43 , and ≥ 0.6 was considered suitable. AUROC values were used to identify the optimal cutoff value. Comparison of AUROC values was performed using the Delong test. Sensitivities, specificities, positive predictive values, negative predictive values and overall accuracy were also calculated. All statistical analyses were performed using SPSS (version 25; SPSS, Chicago, IL, USA) and EZR (Saitama Medical Center, Jichi Medical University, Saitama, Japan) 44 , and a two-sided P < 0.05 was considered statistically significant.   body (a, b, e2). Typical endoscopic findings of sticky adherent dense mucous (f) and multiple hyperplastic polyps in the proximal stomach (g).