Migration of lymphocytes from blood vessels into tissues is a complex process with multiple adhesion and activation steps [1, 32]. Briefly, lymphocytes transiently adhere to the endothelial surface of venules in the tissue (step 1), are activated by chemokines (step 2), and adhere firmly to the endothelia (step 3), before migrating through the vessel wall into the tissue [1, 32]. Thus, the adhesion molecules that are expressed on the HEVs in a secondary lymphoid tissue are important in determining which subsets of lymphocytes migrate into the tissue. Here we evaluated BALT from lobectomy specimens of 17 human adults with lung carcinoma for the expression of HEV and lymphocyte adhesion molecules that might recruit lymphocytes from blood vessels into BALT. We found that BALT HEVs expressed PNAd, ICAM-1 and ICAM-2, with or without VCAM-1. MAdCAM-1 was not expressed on HEVs in any of the BALT samples. The adhesion molecule profile of HEVs in human BALT is similar to that of HEVs in mouse BALT, differing only in the extent of VCAM-1 expression (66% of HEVs in human and 91% of HEVs in mouse BALT) . Thus, BALT in humans, as in mice , has HEV adhesion molecules that are capable of recruiting distinct subsets of lymphocytes from the bloodstream into the tissue.
As in mice, the combination of adhesion molecules on HEVs in human BALT differs from that on HEVs in other secondary lymphoid tissues . Specifically, VCAM-1, which was expressed by endothelia of 66% of BALT HEVs in our study, is not significantly expressed on HEV endothelia in PP (Fig. 2D), adenoids and peripheral LNs [31, 33]. PNAd, which was strongly expressed on all HEVs in BALT, is weakly expressed on HEVs in appendix and PP (Fig. 2B) . MAdCAM-1, which was not expressed on BALT HEVs, is strongly expressed on appendix and PP HEVs .
We found that 86% of BALT CD4+ T cells had a memory phenotype (CD4+ CD45RO+). Since BALT does not have afferent lymphatic vessels , these memory T cells may have migrated through blood vessel HEVs into BALT and/or may have arisen in situ from the maturation of naive T cells. As in other secondary lymphoid tissues, BALT T cells differ from B cells in adhesion molecule expression. L-selectin, which binds to PNAd, was expressed on 20% of T cells and almost all B cells in BALT. α4 integrin, which is a subunit of α4β1 integrin that binds to VCAM-1, was expressed on 43% of T cells and almost all B cells in BALT. LFA-1, which binds to ICAM-1 and ICAM-2, was expressed on almost all lymphocytes in BALT.
The Stamper-Woodruff in vitro binding assay can be used to determine which adhesion molecules mediate the binding of viable lymphocytes to HEVs on frozen sections of tissues [12, 35]. Due to the scarcity of frozen samples of human BALT, we were unable to perform these assays. However, our immunohistology studies on human and mouse BALT and our functional studies on mouse BALT  suggest that PNAd and VCAM-1 are involved in organ-selective recruitment of specific subsets of lymphocytes to human BALT. Specifically, L-selectin+ naive T cells could be recruited from the bloodstream into human BALT by binding to PNAd+ HEVs . Since specialized cells, such as M cells and/or dendritic cells, transport airway luminal antigens into BALT, the naive T cells could meet antigen-bearing dendritic cells in BALT, resulting in the generation of lung-specific α4β1 integrin+ memory T cells. Following release into the bloodstream, the α4β1 integrin+ memory T cells could be recruited back to BALT by binding to VCAM-1+ HEVs. Additionally, the α4β1 integrin+ memory T cells could migrate into inflamed lungs by binding to VCAM-1, which is highly expressed on vessels in human inflamed lung [37–39]. Thus, in humans as in mice, the VCAM-1/α4β1 integrin adhesion system may unify the migration pathways of T cells into BALT and inflamed lung.
The frequency of BALT in normal lungs of healthy adults varies considerably between species. For example, BALT is found in lungs of most normal adult rats, rabbits and guinea pigs [5, 6, 10, 12, 13, 40]. In contrast, the existence of BALT in lungs of healthy adult humans is controversial. In published reports, BALT was found in 0% to 74% of normal human lungs, as compared to 35% of lungs in our study [14–16, 18, 20, 41]. The marked differences in the frequency of human BALT between studies may be due to several factors, including the criteria used to determine if the lung is "normal" and/or the person is "healthy", differences in human subject populations, the source and size of the specimen (i.e., bronchial biopsy, surgical wedge resection, or surgical resection of a lobe or lung), the number of tissue samples and sections examined, and the histologic criteria used to identify BALT.
As with other lymphoid tissues, BALT can undergo marked changes in size, cellular composition, and function during local immune responses. In animal models, BALT can be induced and/or activated by infection or immune stimulation of the lower respiratory tract [23, 42–45]. In humans, hyperplastic BALT can be seen in patents with chronic pulmonary inflammatory disorders, such as hypersensitivity pneumonitis, and in patients with autoimmune diseases, including rheumatoid arthritis and Sjogren's syndrome [15, 46–50]. Exposure to irritants, such as those in cigarette smoke, may influence the development of BALT, as BALT was found in 82% of smoking and 14% of non-smoking human adults in one study . In our study, BALT was slightly more common in current smokers (42%) than in non-smokers (29%).
It is clear from this study and previous studies that human BALT shares many features with secondary lymphoid tissues, including distinct T cell and B cell zones, follicular dendritic cells, PNAd+ HEVs, and expression of lymphoid chemokines CXCL13 and CCL21 (Fig. 2) . Some unique features of BALT in humans and animals, however, lead to the debate regarding whether BALT is a secondary lymphoid tissue controlled by a precise developmental program or a tertiary lymphoid tissue controlled by lymphoid neogenesis. BALT varies by species, strain, age, and antigen stimulation: it is more common in rabbits and rats than in mice [5, 6, 10, 13, 40]; autoimmune-prone nonobese diabetic mice and old mice have more prominent BALT than nonautoimmune-prone and young mice, respectively ; and antigen stimulation and microbe infection can induce BALT formation in lungs of mice [23, 42, 43, 51]. Additionally, HEVs in mouse and human BALT express VCAM-1 (Fig. 2) , which is frequently seen on HEVs of tertiary lymphoid tissues but not of LNs and PPs . Thus, BALT is not identical to "conventional" secondary lymphoid tissues such as LNs and PPs.
A limitation of our study and of several other studies of human bronchopulmonary immunology [35, 53] is that the tissues were obtained from carcinoma lobectomy specimens. The factors, such as cigarette smoke, that led to the development of the carcinoma could also lead to the development of chronic inflammation, such as chronic bronchitis. In addition, the carcinoma could initiate an inflammatory response (for example, by obstructing a bronchus, leading to post-obstructive pneumonia). We made every effort to minimize these potential confounding factors. Specifically, we excluded patients with 1) known inflammatory pulmonary diseases; 2) macroscopic evidence of inflammation or infection in the lobectomy specimen; or 3) histologic features of inflammation or infection . Moreover, we used strict histopathologic criteria to identify "classic" BALT as described by Bienenstock and colleagues and in the pathology literature [5, 29].