ñaco

[I wrote this commentary for the spring issue of Commonwealth Magazine.  I am watching the new crop of 17 Accountable Care Organizations — ACOs — with great interest.  This is a nationally important demonstration that also holds risks for the medical care of many MassHealth enrollees.]

ON MARCH 1, the state’s Medicaid program—known as MassHealth—entered a new era with the launch of 17 accountable care organizations, or ACOs, aiming to provide better coordinated care at lower costs to its low-income enrollees. It’s an ambitious effort with lots of risk and big potential rewards. Within this is another compelling effort to redefine how community health centers fit into the changing health care landscape of Massachusetts and the nation.

Christina Severin, CEO of C3, the new accountable care organization formed by community health centers.

It began with a serendipitous encounter at a grocery store. Sometime in the fall of 2014, Christina Severin bumped into Lori Berry at the seafood counter of the Brighton Whole Foods market. Severin, a long-time leader in the MassHealth scene, had been mulling the creation of a community health center-based non-profit to join the cohort of ACOs being planned for as many as two-thirds of the 1.9 million Massachusetts residents who rely on the program.

Severin had been a savvy player in Boston-area health care organizations for 20-plus years, at Codman Square Health Center in Dorchester, at the Medicaid managed care organization called Network Health, at Beth Israel Deaconess Hospital’s new ACO, and more. Berry, now retired, was the long-serving CEO of the Lynn Community Health Center. They agreed that, under the likely scenario, the state’s highly regarded community health center network would play second banana at best in the emerging, hospital-dominated MassHealth ACO sweepstakes. Severin floated an out-the-box idea: “Why don’t health centers start their own ACO?” Berry was intrigued.

Manny Lopes, the energetic CEO of the East Boston Neighborhood Health Center, the state’s largest community health center, had the same idea and organized four other health center leaders, including Berry, to promote it.  An early stop in 2015 took them to Marylou Sudders, Gov. Charlie Baker’s new secretary of health & human services and the key driver in the complex, high-stakes ACO deal with the US Centers for Medicare and Medicaid Services.

At her first meeting with the rebels, Sudders and her MassHealth director, Daniel Tsai, expressed skepticism, worried that the earnest health center directors didn’t sufficiently grasp the concept of risk. She was concerned that failure could bankrupt some of the state’s most important community health resources.  At their next meeting, the health center leaders brought Severin as their CEO for the new “Community Care Cooperative,” or C3. Berry saw in their faces the changing attitudes of Sudders and Tsai.

With the March 1 launch of 17 MassHealth ACOs, C3 stands as the second largest, with 113,000 enrollees as of early March and 15 participating health centers. A new player, led by Lopes as chairman and Severin as CEO, has emerged on the Massachusetts health scene.

What is an accountable care organization and why is this happening now to 830,000 low-income MassHealth enrollees? ACOs are networks of health providers such as hospitals, physicians, health centers, post-acute providers, home health organizations, and others that join together to provide coordinated care to a set of patients, assuming financial risk and clinical responsibility to improve enrollees’ health, quality of care, and costs.

ACOs were legitimized in the Affordable Care Act to move US health care away from fee-for-service medicine that rewards quantity and toward value-based care that rewards quality and efficiency. Since 2010, more than 1,000 ACOs have formed in Medicare, in commercial insurance, and increasingly in state Medicaid programs. The track record in Medicare shows improved quality and little—if any—progress on costs.

This value-based transformation was an overarching goal for the Obama administration. On November 4, 2016, days before the election that brought Donald Trump to the White House, the federal and Massachusetts state governments agreed on a five-year waiver allowing 1.3 million of MassHealth’s 1.9 million members to move into ACOs. The first wave includes 830,000. Importantly, the waiver lasts until 2022, beyond Trump’s first term. The deal authorizes $52 billion in federal and state spending, including a $1.8 billion investment fund to help providers build ACOs. In return, the state commits to quality improvements and 2.8 percent annual savings with financial penalties for failure. The 17 ACOs—with affiliated managed care, behavioral health, and community partners—are at risk at the sharp end of that promise.

For MassHealth enrollees whose plans changed on March 1, the promise is better coordinated and more effective care to keep them healthy and better treated when sick (see table for list of ACOs). New integrated networks of behavioral health and long-term service providers are mandatory for each ACO. East Boston’s Lopes hopes patients “will see better coordination and more resources in nurse managers and care coordinators, better hospital follow-up, and medication reconciliation.” Dr. Tim Ferris of Massachusetts General Hospital’s ACO says, “the most important and positive thing about the ACO is that it is asking people who deliver the care to manage the care.”

Most of the 830,000 MassHealth clients were auto-enrolled in an ACO based on the affiliation of their primary care physician. For many, their ACO network will not include specialists or other providers with whom they have prior relationships. Enrollees have until June 1 to change plans and/or work through new relationships. For many, notified of the change by letter sent in February, confusion reigns as MassHealth, ACOs, and advocacy groups scurry to navigate care transitions and negotiate exceptions.

While consumer advocates praise MassHealth’s efforts, Bill Henning, executive director of the Boston Center for Independent Living, a key disability advocate, worries about “lots of moving parts and people getting lost.” Vicky Pulos of the Massachusetts Law Reform Institute is “concerned that some people with complex needs who were assigned to the plan their primary care provider joined may lose access to specialists or not understand what steps to take to maintain access. MassHealth is working hard to get the word out but the changes are complex and confusing.”

All ACOs had to decide early whether to partner with a private managed care organization such as Tufts Health Public Plans or Boston Medical Center’s HealthNet for financial and administrative services (Model A) or to work directly with MassHealth (Model B).  Only Steward Medicaid Care Network, Partners HealthCare, and C3 chose Model B.

Severin appreciated that this model emphasizes keeping primary care practice as the focal point and sends money directly to them, not through an intermediary managed care organization.

A surprise was the non-inclusion among the final 17 ACOs of UMass Memorial Health Care in Worcester, the biggest institutional player in Central Massachusetts. UMass was an original pilot ACO in 2017 but couldn’t reach a financial agreement with Tufts Health Public Plans on an ACO for the full-fledged program. The development, a blow for UMass as a system that has embraced value-based programs, was a win for C3 which then added Worcester’s two major community health centers to its network.

In this new MassHealth ACO world, size is not an advantage. Across the nation, ACOs not tied to hospital systems tend to outperform their institutional counterparts. Three MassHealth ACOs fit that category: Atrius Health, Reliant Medicaid Group in Central Massachusetts, and C3. For C3, this is central to their identity. Berry, the former Lynn health center CEO, recalls: “We feared that unless primary care providers made the decisions, savings from reducing unnecessary hospital care would be appropriated by the hospitals. This is an opportunity to use savings to enhance primary care, behavioral health, and prevention relating to social needs such as housing, transportation, and food insecurity for our vulnerable clients.”

All in all, it’s a mighty complex undertaking, in many ways exceeding the massive 1997 Massachusetts Medicaid transformation into today’s MassHealth. That era anointed managed care organizations as system organizers for 20 years. But MCOs proved unable to align, integrate, and coordinate medical providers, who were always kept at arm’s length. This new era seeks to put providers in the driver’s seat, with or without an MCO. Will it happen, and will it matter enough in quality and costs to satisfy the federal government? We have to wait and see—though we will see winner and loser ACOs along the way.

We don’t have to wait for judgments on the key architects, Sudders, and her staff, led by Tsai, the MassHealth director. Praise for their skill, transparency, and collaboration is close to universal. “Nobody has done this in the country,” Severin said of the state putting so much of its Medicaid program into ACOs, with all the risks that carries.

Most ACOs across the country take “one-sided risk,” meaning “heads we win, tails we’re held harmless.” In Massachusetts, all 17 are accepting risk, even some who know they will lose financially because providers recognize this as a new reality, not a passing fad. Across the nation, ACO networks are expected to pay for all the sizable infrastructure costs. In Massachusetts, state officials negotiated a $1.8 billion pool from the federal government to subsidize those costs. This is an important and fascinating experiment, and an historic moment for MassHealth.

THE NEW MASSHEALTH ACOS

John E. McDonough teaches at the Harvard T. H. Chan School of Public Health.

ACO, MACO, and amphotericin B were purchased from Sigma-Aldrich (St. Louis, MO, USA). They were prepared as stock solutions by dissolving in dimethylsulfoxide (DMSO) and diluting to the final concentration with external or internal solution before the start of the experiment. The final concentration of DMSO was less than 0.01%, which does not affect peak I_Na, I_Na/K, or APs. Tetrodotoxin (TTX, Sigma-Aldrich) was also prepared as a stock solution by dissolving it in distilled water and was stored at −80 °C. OUA (Sigma-Aldrich) was simply added to the external solution before the experiments. Type II collagenase was purchased from Worthington Biochem, Freehold, NJ, USA. Nimodipine and RIPA lysis buffer were purchased from Thermo Fisher Scientific, Rockford, IL, USA. Fluo-4 AM was purchased from Invitrogen, Carlsbad, CA, USA.

Male guinea pigs (Hartly, 300–350 g) were obtained from the Experimental Animal Center of Hebei Medical University. The uses of experimental animals were approved by the Institutional Animal Care and Use Committee of Hebei Medical University. This investigation conformed to the Guidelines for the Care and Use of Laboratory Animals and followed the approval of the Bioethical Committee of Hebei Medical University.

The ECG recordings were performed on 25 male guinea pigs weighing 300–350 g. The animals were randomly assigned into three groups: the saline control group (n = 5, to reduce the number of animals used), the ACO group (25 μg/kg group, n = 10), and the MACO group (25 μg/kg group, n = 10). After weighing, the animals were anesthetized by intraperitoneal injection with 1.2% sodium pentobarbital at a dosage of 3 mL/kg [11]. After 10 min of stabilization, 25 μg/kg ACO or MACO was administered as a bolus injection through the femoral vein. The dosages of ACO and MACO were determined based on previous studies [4, 5] and our preliminary experiment. Normal saline solution (1.0 mL/kg) was given to the control animals. Next, the standard lead II ECGs were continuously recorded using a computerized Bipoc System (RM6240, Chengdu, China) for 120 min before and after the administration of ACO, MACO, or normal saline. Arrhythmias were evaluated according to the diagnostic criteria advocated by the Lambeth Convention [16]. After recordings, the time of the first occurrence (onset time) and the incidence (the rate of occurrence) of various arrhythmias in each group were analyzed. Finally, the time to death of each animal and mortality in each group were calculated.

Single ventricular myocytes were enzymatically dissociated from the hearts of healthy adult guinea pigs (300–350 g) as described previously [17] but with slight modification. Using a Langendorff retrograde apparatus, the isolated hearts were perfused retrogradely with Ca²⁺-free cold Tyrode’s solution composed of (in mM) NaCl, 140; KCl, 5.4; MgCl₂, 1.0; HEPES, 10; and glucose, 10 (pH 7.4, adjusted with NaOH) for 5 min, and then the solution was switched to one containing 0.4 mg/mL type II collagenase and continually perfused for 12–15 min. Next, the left ventricular free wall was cut into small pieces in a high-K⁺ solution composed of (in mM) KOH, 80; KCl, 40; KH₂PO₄, 25; MgSO₄, 3; glutamic acid, 50; taurine, 20; HEPES, 10; EGTA, 1; and glucose, 10 (adjusted to pH 7.2 with KOH). Cells were harvested, and only those exhibiting a rode-shaped morphology were used for electrophysiological recording within 6–8 h after isolation.

Patch pipettes were fabricated from glass capillaries (OD, 1.5 mm; ID, 0.9 mm, Warner Instrument Co.) using a Sutter P-97 microelectrode puller (Sutter Instrument, Novato, USA). The tips were then heat polished with an MF-900 microforge (Narishige, Tokyo, Japan). When filled with the standard pipette solution, the pipettes had a resistance of 2–4 MΩ. All experiments were performed at the room temperature (22–25 °C). After patching into a myocyte in a whole-cell configuration, the membrane potential was measured with an Axoclamp 700B amplifier (Axon Instrument Inc., Foster City, USA). The electrical signals were sampled at 2.5–10 kHz, filtered at 2 kHz using a low-pass filter and digitized with an A/D converter (Digidata 1440 A; Axon Instruments). pClamp software (Version 8.1; Axon Instrument) was used to generate voltage-pulse protocols and acquire and analyze data.

The cell suspension was placed on the microscopic groove. The amphotericin B (250 μg/mL) perforated patch-clamp technique was used to record the APs in the current-clamp mode [17, 18]. Pipette solution contained (in mM) potassium glutamate, 120; KCl, 25; MgCl₂, 1; CaCl₂, 1; and HEPES, 10 (pH 7.2 adjusted with KOH). The external solution contained (in mM) NaCl, 138; KCl, 4; MgCl_2, 1; CaCl₂, 2; NaH₂PO₄, 0.33; glucose, 10; HEPES, 10 (pH 7.4 adjusted with NaOH). APs were evoked with a supra-threshold (2000 pA) current pulse of ~4–6-ms duration at a rate of 1 Hz. Then, the parameters of APs, including RMP, APA, APD₃₀, APD₅₀, and APD₉₀, were analyzed before and after the treatment of the myocytes with ACO or MACO at different concentrations [17, 18]. The concentrations of ACO (0.3, 3 μM) were selected based on a previous study [5] and our preliminary experiments. For MACO, since it has similar yet more potent effects than ACO, relatively lower concentrations (0.1, 0.3 μM) were adopted.

Peak I_Na was recorded according to the method described previously with some modifications [19]. The patch pipettes were backfilled with a Cs⁺-rich internal solution containing (in mM) CsCl, 120; MgCl₂, 5; CaCl₂, 1; Na₂ATP, 5; EGTA, 11; and HEPES, 10 (pH 7.4 adjusted with CsOH). The external solution contained (in mM) choline chloride, 130; CsCl, 5.4; CaCl₂, 1; MgCl₂, 1; NaH₂PO₄, 0.33; HEPES, 10; NaCl, 20; and nimodipine, 0.01 (pH 7.3 adjusted with CsOH). As peak I_Na is fast and large, special conditions were necessary to record it quantitatively. Patch pipettes with low resistance (1.5–2.5 MΩ) were used to facilitate the dialysis of cells with the pipette solution and to minimize voltage errors. Cs⁺-based external and internal solutions containing similar levels of Na⁺ were used to reduce the gradient for Na⁺ entry and to block any contamination from the potassium current. In addition, 65%–75% of the series resistance could be compensated [20]. To determine the voltage dependence of steady-state activation, currents were elicited by a 50-ms pulse from a holding potential of −90 mV to test potentials between −100 and 0 mV in 5 mV increments. The Na⁺ conductance (G) was calculated by dividing the peak current for each voltage step by the driving force (V_m − V_rev) and then normalizing it to the peak conductance (G_max). Data were fitted with the Boltzmann equation, G/G_max = 1/{1+exp[(V_1/2 − V_m)/k]}, in which V_1/2 was the voltage at which half of the Na⁺ channels were activated, k was the slope factor, and V_m was the membrane potential. The steady-state inactivation curves were generated via standard two-pulse protocols: cells were stepped to 50-ms preconditioning potentials from the holding potential of −90 mV, varying between −120 and −25 mV (prepulse), followed by a 50-ms test pulse to −20 mV. Currents (I) were normalized to I_max and fit to a Boltzmann equation of the form I/I_max = 1/{1+exp[(V_m − V_1/2)/k]}, in which V_1/2 was the voltage at which half of the Na⁺ channels were inactivated, k was the slope factor, and V_m was the membrane potential. Recovery from inactivation was analyzed by fitting data with the two exponential equations I(t)/I_max = A_f × [1–exp(–t/τ_f))] + A_s ×  [1–exp(–t/τ_s)], where values for A and τ refer to amplitudes and time constants, respectively [19, 21, 22]. Curve fitting and data analysis were performed using Clampfit 10.2 software (Axon Instruments) and Origin 8 (OriginLab Corporation).

The external solution used for recording the I_Na/K contained (in mM) NaCl, 137.7; NaOH, 2.3; MgCl₂, 1; glucose, 10; HEPES, 5; KCl, 4.6; BaCl₂, 0.5; and CdCl₂, 0.2 (pH 7.4, adjusted with HCl or NaOH). The standard pipette solution contained (in mM) sodium aspartic acid, 50; potassium aspartic acid, 20; CsOH, 30; TEACl, 20; HEPES, 5; MgSO₄, 5; EGTA, 11; glucose, 10; and Na₂GTP, 5 (pH 7.2, adjusted with CsOH). Under the present experimental conditions with selected external and pipette solutions, the membrane current through Ca²⁺ channels, K⁺ channels, and Na⁺/Ca²⁺ exchangers (NCXs) was minimized. K⁺ channel currents were suppressed by the addition of CsOH and TEACl to the internal solution and BaCl₂ to the external solution. High-threshold (N-, L-, and P- type) and low-threshold (T- type) calcium channel currents were eliminated by the removal of Ca²⁺ from the external solution, and any residual influx of Ca²⁺ through these voltage-gated Ca²⁺ channels was blocked by the addition of CdCl₂ to the external solution. The Na⁺ channels and T-type Ca²⁺ channels were inactivated by setting the holding potential to 0 mV, and the NCX current was blocked by eliminating Ca²⁺ from the external and internal solutions [23]. Usually, the recording was started when the current reached steady-state level within 3–5 min after patch rupture. Under these conditions, membrane currents through Ca²⁺ channels, K⁺ channels, and NCXs were mostly blocked, and the I_Na/K was identified as the ouabain (OUA)-sensitive current [24, 25].

Fresh guinea pig ventricular myocytes were incubated with ACO (0.3, 3 μM) and MACO (0.1, 0.3 μM) at the room temperature for 30 min, and then 500 μM RIPA lysis buffer was added to lyse the cells at 4 °C for intracellular ion measurement. The ion extraction was centrifuged at 4 °C for 20 min at 12,000 rpm [20, 25]. [Na⁺]_i was measured by an ion-selective electrode with a Na/K/Cl analyzer (Easylate PLUS, Medica Corporation, USA).

[Ca²⁺]_i measurement was performed according to a previous report with slight modifications: ventricular myocytes were incubated with 2 μM Fluo-4-acetoxymethyl ester (Fluo-4 AM, Invitrogen, Carlsbad, CA, USA) in normal Tyrode’s solution containing 1.8 mM Ca²⁺ for 30 min at the room temperature. Cells were then washed with dye-free normal Tyrode’s solution for 15 min for de-esterification. The fluorescence intensity (IF) of ventricular myocytes was detected by confocal microscopy (SP5, Wetzlar, Germany) using a 488 nm laser for excitation and a 530 nm laser for emission during one experimental process; 60 scans with 0-s intervals were used [9, 11, 26]. Representative fluorescent images were captured at a frames per 8 s before (control) and after the addition of ACO (0.3 or 3 μM) or MACO (0.1 or 0.3 μM) by using 60 mM KCl as a background. The changes in [Ca²⁺]_i were expressed as F/F₀, in which F₀ and F indicated the IF before and after the administration of ACO or MACO, respectively, within 24 s for lower concentrations and within 64 s for higher concentrations.

All averaged values presented are means ± SEMs. Statistical analysis was performed by one-way analysis of variance (ANOVA), followed by Dunnett′s multiple comparison test where applicable (SPSS for Windows v11.0). The differences in the parameters before and after treatment were compared using a paired Student’s t test. The percentages of occurrence were compared by the χ² test or Fisher’s exact probability method. Differences were considered significant when P < 0.05.

The global burdens of asthma and chronic obstructive pulmonary disease (COPD) are increasing, each of which was estimated to affect respectively approximate 339 million and 251 million people worldwide in 2016 [1]. It has been widely accepted that asthma and COPD are strikingly different airway disorders [2,3]. Although the “Dutch hypothesis” suggested a common genetic background underlying airway obstruction with a spectrum of clinical entities from asthma to COPD, recent genetic research indicated that it was unlikely that genetic factors are shared by asthma and COPD [4].

Asthma is a heterogenous and inflammatory disease affecting large and small respiratory tracts but not the lung parenchyma, and contains clusters of demographical, clinical and pathophysiological characteristics underpinned by different pathophysiological processes [5]. This heterogeneity may be explained by the complexity of dysregulated innate and adaptive inflammatory responses to exogenous allergens and proteases leading to the spectrum of abnormal tissue remodeling, where type 2 cytokines such as interleukin (IL)-4, IL-13 and IL-5 primarily promote airway eosinophil infiltration, mucus hypersecretion, bronchial hyperresponsiveness and mast cell activation [6]. Major subpopulations of asthmatics have molecular signatures of T helper 2 (Th2)—inflammation and airway obstruction that markedly respond to inhaled corticosteroid (ICS) [7]. In line with this translational study, accumulated evidence from randomized control trials have revealed the importance of ICS usage from the early steps of asthma treatment because clinical studies have shown that ICS robustly reduced the risk of symptoms, exacerbations, hospitalization and mortality from asthma [8,9,10].

COPD is defined as a common, preventable and treatable disease that is characterized by persistent respiratory symptoms and airflow limitation that is due to airway and/or alveolar abnormalities usually caused by significant exposure to noxious particles or gases and influenced by host factors including abnormal lung development [11]. In addition to cigarette smoking, known as the most common COPD risk factor [12], the susceptibility could be influenced by genetic factors [13,14] and abnormal lung growth [15]. Unlike asthma, CD4⁺ T helper 1 (Th1) cells, CD8⁺ cytotoxic T (Tc) cells, neutrophils and macrophages predominantly affect the small airways and the lung parenchyma leading to mucus hypersecretion, alveolar wall destruction (emphysema) and small airway fibrosis in COPD [2,16]. These pro-inflammatory cell-types are functionally altered by oxidative stress and intracellular signaling pathways including activation of the proinflammatory transcription factor nuclear factor κB (NF-κB) [17], and alveolar macrophages are defective in bacterial phagocytosis, possibly via several phagocytic receptors and mitochondrial molecules related to oxidative stress [18,19,20,21]. The small airway narrowing induced by pro-inflammatory cell infiltration, luminal exudates, wall thickening, and the loss of small airways associated with emphysema increases airway obstruction [22,23]. In the wall thickening, hyperplasia of basal cells, known as airway epithelial stem cells, could be formed through several molecules such as Axl receptor tyrosine kinase [24] and Yap-Wnt7b [25]. The airflow limitation progressively leads to gas-trapping in peripheral lungs during expiration on exercise, resulting in dynamic hyperinflation which is postulated to be the main mechanism of exertional dyspnea [26,27]. Thus bronchodilators, long-acting muscarinic antagonists (LAMA) and long acting beta₂-agonists (LABA), are commonly used as the pharmacological therapy for COPD and are known to reduce lung hyperinflation, dyspnea and exercise endurance [28,29] leading to improvement of the quality of life and a reduction in the frequency of exacerbations [30]. Accumulated evidence indicates that LAMA significantly reduce the frequency of exacerbations and non-serious adverse events and increase the trough forced expiratory volume in one second (FEV₁) compared to LABA in patients with stable COPD [31].

Open AccessThis article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/

The energy metabolites, cholesterol and fatty acids correlated significantly with the immunological mediators, suggesting existence of a possible link between the inflammatory status of these patients and impaired metabolism. The present findings could be possibly extended to better define the ACO diagnostic criteria, management and tailoring therapies exclusively for the disease.

Eleven metabolites [serine, threonine, ethanolamine, glucose, cholesterol, 2-palmitoylglycerol, stearic acid, lactic acid, linoleic acid, D-mannose and succinic acid] were found to be significantly altered in ACO as compared with asthma and COPD. The levels and expression trends were successfully validated in a fresh cohort of subjects. Thirteen immunological mediators including TNFα, IL-1β, IL-17E, GM-CSF, IL-18, NGAL, IL-5, IL-10, MCP-1, YKL-40, IFN-γ, IL-6 and TGF-β showed distinct expression patterns in ACO. These markers and metabolites exhibited significant correlation with each other and also with lung function parameters.

Global metabolomic profiling using two different groups of patients [discovery (D) and validation (V)] were conducted. Serum samples obtained from moderate and severe asthma [n = 34(D); n = 32(V)], moderate and severe COPD [n = 30(D); 32(V)], ACO patients [n = 35(D); 40(V)] and healthy controls [n = 33(D)] were characterized using gas chromatography mass spectrometry (GC-MS). Multiplexed analysis of 25 immunological markers (IFN-γ (interferon gamma), TNF-α (tumor necrosis factor alpha), IL-12p70 (interleukin 12p70), IL-2, IL-4, IL-5, IL-13, IL-10, IL-1α, IL-1β, TGF-β (transforming growth factor), IL-6, IL-17E, IL-21, IL-23, eotaxin, GM-CSF (granulocyte macrophage-colony stimulating factor), IFN-α (interferon alpha), IL-18, NGAL (neutrophil gelatinase-associated lipocalin), periostin, TSLP (thymic stromal lymphopoietin), MCP-1 (monocyte chemoattractant protein- 1), YKL-40 (chitinase 3 like 1) and IL-8) was also performed in the discovery cohort.

Asthma-COPD overlap (ACO) refers to a group of poorly studied and characterised patients reporting with disease presentations of both asthma and COPD, thereby making both diagnosis and treatment challenging for the clinicians. They exhibit a higher burden in terms of both mortality and morbidity in comparison to patients with only asthma or COPD. The pathophysiology of the disease and its existence as a unique disease entity remains unclear. The present study aims to determine whether ACO has a distinct metabolic and immunological mediator profile in comparison to asthma and COPD.

Our earlier findings using NMR metabolomics indicate an enhanced energy and metabolic burden associated with ACO as compared to asthma and COPD [27]. This motivates us to gain a deeper insight into inflammation-related metabolism in ACO. The present study combines gas-chromatography-mass spectrometry (GC/MS) based metabolomics coupled with wide spectrum profiling of inflammatory mediators for this purpose.

Metabolic profiling has been successfully applied to obtain an in-depth understanding of the pathophysiology of obstructive lung diseases, such as asthma and COPD [21–24]. The precise involvement of metabolites in the pathobiology of ACO, however, is yet not well understood. Limited reports exist on metabolite studies in ACO. Eicosanoids, found to be in higher levels in ACO and metabolized through lipoxygenase, is suggested to discriminate well between ACO and COPD [25]. Another study has indicated significantly increased levels of L-histidine in urine of patients with ACO as compared with asthma or COPD [26].

Understanding the metabolic implications of chronic inflammatory processes is, therefore, an urgent need. A suitable tool for this purpose is metabolic profiling, as it allows the investigation of a broad range of small molecules (metabolites) in various body fluids. Metabolites are the intermediate and end products of cellular metabolic processes within an organism under any given physiological condition [18]. Metabolomics deals with the analysis of these metabolites present in human specimens in various states of health and disease [18]. The most commonly used samples are serum, urine, sputum, saliva and faeces, because they are obtained from patients involving minimally/non-invasive procedures [19, 20]. Advancement of analytical techniques such as nuclear magnetic resonance (NMR) and mass spectrometry (MS) has enabled quantitative identification of a wide range of metabolites using a small volume of sample.

The local and systemic responses are highly activated in both asthma and COPD [11, 12]. There are reports of markers such as NGAL, YKL-40, IL-6, periostin being studied in ACO in recent years [13, 14]. Multiplexed analysis of immunological markers allows for the quantitative measurement and comparison of a broad range of inflammatory mediators that aids in creating a better understanding of the immune response under any biological condition. Heightened immune response and inflammation are reported to be associated with shift in tissue metabolism [15, 16]. The altered metabolism is a result of the recruitment of inflammatory cell types, particularly myeloid cells such as neutrophils and monocytes. This leads to the generation of large quantities of reactive nitrogen and oxygen intermediates, depletion of nutrients and increased oxygen consumption. The migration of myeloid cells to the site of inflammation is an energy consuming process and demands a large amount of ATP. Further, at the site of inflammation there is an increased nutrient, energy and oxygen demand to accomplish the process of phagocytosis [15]. This can further alter cellular metabolism, including extracellular metabolic pathways which generates biologically active molecules capable of initiating and modulating inflammatory responses [17].

The universally accepted definition of ACO remains elusive till date. In fact, the definition of ACO is still evolving, and different clinical definitions are being provided in various studies [5–8]. The prevalence of ACO depends on how it is defined, but it is relatively common in clinical practice, affecting 15 to 20% of patients with asthma and COPD [9]. In general, patients with ACO are reported to have poorer quality of life, rapid decline in lung function, higher frequency of exacerbations, higher mortality and morbidity in comparison to patients with only asthma or COPD [10]. ACO patients have been largely excluded from basic research and pivotal therapeutic trials, as a result of which the pathogenesis of ACO, including underlying inflammation patterns, remains poorly understood [9].

Asthma and chronic obstructive pulmonary disease (COPD) are two heterogenous obstructive airway disorders that are associated with distinct pathological mechanisms. Asthma is broadly characterized by airway hyperresponsiveness which leads to reversible airflow obstruction based primarily on type 2 eosinophilic inflammation [1, 2]. COPD shows progressive and irreversible airflow obstruction typically caused by exposure to noxious gases and is majorly associated with neutrophilic inflammation involving CD8+lymphocytes and macrophages [1, 3]. Asthma-COPD overlap (ACO), frequently encountered in medical practice, refers to patients presenting with characteristics of both asthma and COPD, thereby making both diagnosis and treatment challenging for the clinicians [4].

All values are expressed as mean ± standard deviation (SD). One way ANOVA (Dunnett’s post hoc test) or Kruskal–Wallis test (Dunn’s post hoc test) was conducted for pairwise comparisons. Statistical analyses were performed using GraphPad Prism version 7.00 for Windows, GraphPad Software, San Diego, CA, USA. A p-value ≤0.05 was considered to be statistically significant. Immunological markers significantly altered in ACO as compared with asthma and COPD were identified and only those dysregulated mediators common to both ACO vs. asthma and ACO vs. COPD considered for further analysis. Pearson’s correlation analysis was performed between each of these mediators and the lung function parameters, i.e. FEV1 and FEV1/FVC of ACO subjects.

Data was read on the Luminex MAGPIX machine (Luminex Corporation) and analyzed using XPONENT 4.2 software (Luminex Corporation). The test procedure was adopted as per the manufacturer’s instructions. As per the manufacturer’s instructions, the lower detection limit of each analyte is shown in brackets: IFN-γ (0.40 pg/mL), TNF-α (1.2 pg/ml), IL-12p70 (20.2 pg/ml), IL-2 (1.8 pg/ml), IL-4 (9.3 pg/ml), IL-5 (0.5 pg/ml), IL-13 (36.6 pg/ml), IL-10 (1.6 pg/ml), IL-1α (0.9 pg/ml), IL-1β (11.1 pg/ml), IL-6 (1.7 pg/mL), IL-17E (27.7 pg/ml), IL-21 (0.869 pg/ml), IL-23 (11.4 pg/ml), eotaxin (14.6 pg/ml), GM-CSF (4.1 pg/ml), IFN-α (0.26 pg/ml), IL-18 (1.93 pg/ml), NGAL (29.2 pg/ml), periostin (95.7 pg/ml), TSLP (0.432 pg/ml), MCP-1 (9.9 pg/ml),YKL-40 (3.30 pg/ml), TGF-β (15.4 pg/ml) and IL-8 (1.8 pg/ml). Each sample was analysed in triplicate, and the mean of the three was calculated for every analyte. The standard curve was generated by a 5-parameters logistic fit.

The relationship between significantly altered metabolites (common to ACO vs. asthma and ACO vs. COPD) and lung function parameters of ACO subjects was explored using Pearson’s correlation analysis (GraphPad Prism version 7.00 for Windows, GraphPad Software, San Diego, CA, USA). This was done to investigate the extent to which the dysregulated metabolites were linearly related to FEV1 and FEV1/FVC of ACO subjects.

For MSEA, quantitative enrichment analysis (QEA) was performed on normalized data for comprehensive screening of affected pathways. QEA is based on the global test algorithm to perform enrichment analysis directly from raw concentration data (peak area in this case) and does not require a list of significantly changed compounds. The QEA algorithm uses a generalized linear model to estimate a ‘Q-stat’ for each metabolite set. In addition to the Q-stat values, the QEA module also provide p-values, Holm adjusted p-values, and estimates of false discovery rate (FDR).

Using Metaboanalyst 4.0 (www.metaboanalyst.ca), the peak areas of all the identified metabolites were subjected to pathway analysis and MSEA to identify potential key significantly altered metabolic pathways (ACO vs asthma and ACO vs COPD) [36, 37]. First, to explore the metabolic pathways that are potentially dysregulated in ACO, a global metabolic pathway analysis was carried out. The default ‘global test’ and ‘relative-betweenness centrality’ for pathway enrichment and pathway topological analyses were selected, respectively. The “current 2019” Kyoto Encyclopedia of Genes and Genomes (KEGG) version pathway library was also used.

Each batch consisting of six test and two QC samples was thawed on ice, before metabolite extraction and derivatization procedures, as described previously [34] with minor modifications. In brief, 50 μl of serum sample was thawed on ice and 10 μl of freshly prepared isopropyl maleic acid (1 mg/ml) was added as an internal standard. Next, 800 μl of ice-cold methanol was mixed with the sample and vortexed for 30 s. The suspension was then centrifuged at 15,000×g for 10 min at 4 °C and supernatant dried in a vacuum evaporator at 40 °C for 30 min. Dried samples were then treated with 2% methoxyamine HCl in pyridine (MOX) reagent at 60 °C for 2 h followed by a silylation step with N,O-Bis (trimethylsilyl) trifluoroacetamide (BSTFA) at 60 °C for 1 h. After derivatization, the sample tubes were centrifuged at 10,000×g for 5 min and the supernatant transferred into a glass vial insert kept inside a 2 ml screw capped glass GC vial. Metabolomics standards initiative (MSI) guidelines were followed while performing all the metabolomics experiments [35]. The detailed methodology of GC-MS data acquisition, pre-processing and analysis is given in the Supplementary Materials section.

Coded samples were randomized using a web-based tool (www.randomizer.org) to process these samples for metabolite extraction and derivatization in batches followed by GC-MS data acquisition within 24 h. In metabolomics, use of QC samples in the quality assurance procedure provides a mechanism to assess the analytical variance of the data. The QC sample qualitatively and quantitatively represents pooled samples of equal volume obtained from all enrolled participants. These samples provide an average of all of the metabolomes analysed in the study and ensure data reproducibility [31–33].

Five ml of venous blood samples were collected from subjects post confirmation of their disease status. Samples were incubated at room temperature for 45 min to allow clotting and centrifuged at 1500×g at 4 °C for 15 min. The serum fraction was separated, aliquoted, and stored immediately at − 80 °C. All samples were collected following a minimum of 12 h overnight fasting.

The pilot metabolomic study was conducted on two different patient cohort, comprising of the discovery and validation phase. The discovery phase patient cohort consisted of (i) controls = 33 (ii) asthma = 34 (iii) COPD = 30 and (iv) ACO = 35. For the validation phase, (i) asthma = 32 (ii) COPD = 32 and (iii) ACO = 40 patients were considered. Only the discovery cohort was considered for immunological profiling. Both, the discovery and validation cohort of patients had the same inclusion and exclusion criteria.

All patients were recruited at the Institute of Pulmocare and Research (IPCR) Kolkata, India. The Institutional Human Ethics Committee of IPCR, Kolkata approved this study. Written informed consent was obtained from all participants who volunteered to participate in this study. The detailed inclusion and exclusion criteria are discussed elsewhere [27]. Briefly, the recruited subjects were assigned to four groups: (a) moderate or severe cases of asthma, diagnosed based on the GINA guidelines (GINA 2014) [28] (b) stage II and III, i.e. moderate and severe COPD patients diagnosed according to the GOLD criterion (GOLD 2014) [29] (c) major criteria used for ACO diagnosis were (i) persistent airflow limitation (post-bronchodilator FEV1/FVC <  0.70) in individuals 40 years of age or older (ii) ≥ 10 pack-years of tobacco smoking (iii) documented history of asthma before 40 years of age, or bronchodilator response (BDR) of > 400 mL in FEV1; minor criteria were (i) documented history of atopy or allergic rhinitis (ii) BDR of FEV1 ≥ 200 mL and 12% from baseline values on two or more visits (iii) peripheral blood eosinophil count of ≥ 300 cells/μL; all major criteria and at least one minor criterion was considered for inclusion of subjects into the ACO cohort [5, 30] (d) age-matched healthy male smokers as controls having normal lung function. Only current or former male smokers were recruited in this study to avoid gender and smoking induced bias. Patients who have had history of exacerbations, active respiratory infections, had received oral corticosteroid treatment or antibiotics/antiviral drugs during the previous 3 months were excluded. Patients having other comorbidities including metabolic diseases were also excluded from this study.

Significant negative correlations (− 0.336 to − 0.794, p ≤ 0.05) were observed between serine, ethanolamine, threonine, glucose, cholesterol and succinic acid with TNF-α (not significant with ethanolamine, threonine), IL-1β, NGAL (not significant with threonine), MCP-1 (not significant with serine, cholesterol), YKL-40, IFN-γ, and IL-6. Also, a significant negative correlation was observed between succinic acid and IL-18. Mannose too showed negative correlation with IL-1β, GM-CSF, IL-5, YKL-40, IFN-γ and IL-6. In contrast, stearic acid and linoleic acid positively correlated (0.390 to 0.604, p ≤ 0.05) with TNF-α (not significant with linoleic acid), IL-1β, NGAL, IL-5, IFN-γ (not significant with linoleic acid), IL-6 (not significant with linoleic acid) and YKL-40. Lactic acid showed a significant positive correlation with TNF-α and IL-1β (Fig. ). The energy metabolites and cholesterol negatively correlated with the immunological mediators, whereas the fatty acids and lactic acid showed a positive correlation.

Pairwise Pearson’s correlation analysis was used to assess the association of the significantly altered metabolites with immunological mediators in serum of ACO patients. Correlation coefficients (r) ranged from 1.0 (maximum positive correlation) to − 1.0 (maximum anticorrelation), with a value of 0 representing no correlation in a heatmap. The red coloured cells represent negative correlations while the blue coloured cells represent positive correlations. The size of the squares indicates the magnitude of the correlation.

The expression level of Th1 mediated cytokines, such as TNFα, and IL-1β were significantly higher in ACO cases with respect to asthma and controls. The highest expression of these cytokines was noted in patients with COPD. The expression of IL-5, a pro-inflammatory Th2 cytokine, was highest in patients with ACO as compared to asthma, COPD and controls. The anti-inflammatory cytokine, IL-10 was least expressed in asthma; the levels in ACO cases were significantly less when compared with COPD and controls. Key immunological markers, such as IFN-γ, IL-6, TGF-β and IL-17E/IL-25 showed significantly altered expression profiles in ACO. Immune system-related proteins and chemokines such as NGAL, YKL-40, MCP-1, GM-CSF etc. some of which have already been explored in ACO subjects also exhibited differential expression patterns (Fig. ). The expression level of IFN-α could not be determined. Except IL-10, Pearson’s correlation analysis showed negative correlation between the dysregulated mediators and lung function parameters. However, the correlation was observed to be significant only for GM-CSF, IL-6, IFN-γ, YKL-40, IL-1β, NGAL and IL-5 (Table ).

Pearson’s correlation analysis between each of the significantly altered metabolites and lung function parameters of ACO subjects is provided in Table . Significant positive correlation was observed between metabolites including serine, threonine, glucose, cholesterol, D-mannose, succinic acid and FEV1 and FEV1/FVC. A negative correlation was observed between stearic acid, lactic acid and linoleic acid; however, all correlations were not statistically significant.

ROC diagram plots the true positive rate (sensitivity) of a test on the y-axis against the false positive rate (100-specificity) on the x-axis, thus producing the AUC. An AUC is a measure of the accuracy of a diagnostic test, where 1.0 indicates a perfect test and a 0.5 shows that the test is no better than random chance, and therefore has no diagnostic or prognostic value. ROC curves were generated for the common set of significantly altered metabolites of the two groups (ACO vs. asthma and ACO vs. COPD). Six metabolites with the highest AUCs (AUC > 0.7) were taken into consideration for both the groups to establish a predictive model that could well differentiate ACO from asthma and also from COPD (Fig. ). Four metabolites viz. glucose, 2-palmitoylglycerol, D-mannose and succinic acid were found to be common between the two ROC models.

MSEA also showed multiple metabolic pathways that were significantly dysregulated in serum of ACO patients (p <  0.05). Supplementary Fig. 7 highlights the fold enrichment obtained when using peak areas of all the identified metabolites. The colour intensity denotes the level of statistical significance and the length of each bar represents the fold enrichment of the pathway. Supplementary Tables 3 and 4 present all the perturbed biochemical pathways with number of metabolite hits, p-values, Holm-adjusted p-values and FDR. Pathways with the highest number of hits and significant Holm p-values are considered to be significantly perturbed in ACO patients. The p-values of the pathways is determined by the difference of peak area data and the number of participating metabolites. Some of the most significantly altered pathways in ACO included fructose and mannose metabolism, glycine-serine-threonine metabolism, valine-leucine-isoleucine biosynthesis and glycolysis/gluconeogenesis.

All the 85 identified and quantified metabolites in the discovery cohort were considered for pathway analysis using MetPA. All matched pathways are displayed as circles. The colour and size of each circle is based on the p-value and pathway impact value, respectively. Various common significantly altered pathways were observed in ACO vs. asthma and ACO vs. COPD, including starch and sucrose metabolism, linoleic acid metabolism, glycolysis / gluconeogenesis, citrate cycle (TCA cycle), glycine, serine and threonine metabolism and aminoacyl-tRNA biosynthesis (Supplementary Fig. 6a, b). A few less significant pathways could also be identified (Supplementary Table 1 and 2).

The significantly altered 11 metabolites, including serine, threonine, ethanolamine, glucose, cholesterol, 2-palmitoylglycerol, stearic acid, lactic acid, linoleic acid, D-mannose and succinic acid identified in the discovery phase were further validated in an independent fresh cohort of subjects using UVA (Table ). Similar trends in expression with significant changes were seen as observed earlier in the discovery cohort.

To cross-validate MVA, peak area matrix of all the metabolites were subjected to UVA using ANOVA (Dunnett’s post hoc test) or Kruskal–Wallis test (Dunn’s post hoc test), as applicable. UVA refers to statistical analyses that involve only one dependent variable and which are used to test hypotheses. UVA with multiple testing correction is an attractive approach since it is relatively simple to implement and provides a measure of statistical significance for each covariate that is easy to interpret. Common variables with VIP score > 1.3, ANOVA p-value ≤0.05 and adjusted FDR <  0.01 were selected as the major metabolites responsible for differentiating ACO from both, asthma and COPD. Metabolites including serine, threonine, ethanolamine, glucose, D-mannose and succinic acid were found to be down-regulated in ACO as compared to both, asthma and COPD. Cholesterol, 2-palmitoylglycerol and lactic acid were also down-regulated in ACO, but only with respect to COPD. Interestingly, these metabolites were found to be upregulated when compared with asthma. Also, two metabolites, i.e. stearic acid and linoleic acid were found to be upregulated in ACO as compared to COPD and downregulated with respect to asthma (Table ).

The discovery phase samples were assigned into two groups, obstructive lung diseases (asthma, COPD and ACO) and healthy controls. MVA is a statistical technique which involves the simultaneous observation and analysis of more than two variables. Multivariate methods may be supervised or unsupervised. While unsupervised methods such as clustering are exploratory in nature and help in identification of patterns, supervised methods use some type of response variable to discover patterns associated with the response. Both PLS-DA and OPLS-DA are supervised MVA tools. PLS-DA is a chemometrics technique used to optimise separation between different groups of samples, which is accomplished by linking two data matrices X (i.e., metabolite peak areas) and Y (i.e., groups). PLS aim to differentiate between classes in highly complex data sets, despite within class variability. Initially, PLS-DA models were generated which showed class separation between the disease groups and controls (Supplementary Fig. 2). OPLS-DA is often used in lieu of PLS-DA to disentangle group-predictive and group-unrelated variation in the measured data. In doing so, OPLS-DA constructs more parsimonious and easily interpretable models compared to PLS-DA. Next, OPLS-DA models were generated for optimized separation between the two groups (Fig. a). The permutation test evidenced significantly higher R2 and Q2 values than that of 200 permutated models, indicating a good predictive ability of the model (Supplementary Fig. 3a).

The baseline clinical characteristics of all subjects recruited is tabulated in Table . Following spectral annotation with NIST 14 library, a total of 145 consistent metabolites could be identified. Out of these metabolites, 85 had an occurrence frequency of at least 80% among all samples and were considered for further analysis. A representative GC–MS spectrum is shown in Supplementary Fig. 1. Both multivariate analysis (MVA) and univariate analysis (UVA) were performed on the constant sum normalized, log transformed and mean scaled peak area data matrix of the final 85 annotated metabolites.

In our earlier study, NMR based metabolomics provided new insights into the altered pathways which could be contributing to the higher mortality and morbidity in ACO in comparison to asthma/COPD [27]. Since MS is known to provide complementary information to NMR [18], serum metabolome of the same patients are analysed using GC-MS. In addition, metabolomic data has been integrated with a wide range of inflammatory mediators to improve the understanding of ACO and effectively differentiate it from asthma, COPD and healthy controls.

A total of 11 metabolites and 13 inflammatory mediators were found to be important in distinguishing ACO from both asthma and COPD. Changes in plasma levels of glucogenic amino acids like serine and threonine have been reported by different groups in both asthma [23] and COPD [38, 39]. Serine and threonine may get converted to pyruvate which in turn enters the TCA cycle to compensate the energy demand [40]. We hypothesize that enhanced cellular demand due to upregulation of glycolysis is responsible for the decrease in the expression levels of these metabolites in ACO.

Sugars such as glucose, mannose and succinate, a TCA cycle intermediate were significantly down-regulated in ACO patients as compared to asthma and COPD. Numerous reports on glucose down-regulation in COPD and asthma exist [22, 41, 42]. A similar trend was also observed in our previous NMR study where significant decrease in glucose level in ACO was observed [27]. Mannose is an intermediate metabolite of galactose metabolism pathway and may be converted to fructose-6-phosphate which enters the glycolytic pathway [43]. The significantly low level of mannose in ACO is attributed to increased glycolytic activity and higher energy demand [44]. Significant decrease in the expression of succinate in ACO cases is in good agreement with the reports of other researchers in COPD and asthma where dysregulated succinate levels have been linked with energy metabolism, hypoxemic stress, or prolonged exertion [45–47].

Ethanolamine, a metabolite of glycerophospholipid metabolism, was also observed to be down-regulated in ACO. Its involvement in the synthesis of phosphatidyl ethanolamine, a central intermediate of lipid metabolism and link with cellular respiration is reported [48]. However, the role of ethanolamine in glucose metabolism per se still remains unclear and warrants further investigation.

The levels of 2-palmitoylglycerol and cholesterol were significantly down-regulated in ACO vs COPD whereas a reverse trend was seen in ACO vs asthma. 2-palmitoylglycerol is a monoacylglycerol and can be associated with lipid cycle disruption with subsequent input to energy cycles [49]. It is also documented that monoacylglycerols are not merely intermediate lipid molecules, but may act as signalling molecules in various inflammatory and other immune system related processes [50]. Cholesterol, on the contrary, has been linked to inflammation, and is reported to decrease in serum of patients with asthma and increase in patients with very severe COPD [51, 52]. The expression pattern of cholesterol in ACO cases may be linked to comprehensive changes in lipid and sterol metabolism.

The increased expression of lactate is attributed to increased glycolysis due to an imbalance in oxygen supplement and demand, as explained by the “Warburg effect”. The concentration levels of lactate have been extensively studied in asthma and COPD and similar mechanisms in ACO are also suggested [24, 52, 53]. Stearic acid (18:0) is a saturated fatty acid (SaFA) which was found to be up-regulated in ACO vs COPD and down-regulated in ACO vs asthma. Numerous reports indicate the involvement of SaFAs with the inflammasome, such as proinflammatory cytokines, VEGF, IL-6, IL-1β etc. [54, 55]. Linoleic acid (9,12- Octadecadienoic acid) also exhibited a similar expression pattern. It plays a critical role in cellular metabolism, signaling and is also the precursor of arachidonic acid, which is actively involved in proinflammatory response and Th2 differentiation in asthma patients [56]. Fatty acid levels are generally related to the metabolic status and diet of the subjects; however, none of the participants in the present study were obese or suffering from any other metabolic disorder. It was also ensured that all subjects followed a similar dietary pattern.

We have also studied a wide range of immunological mediators which have been mostly explored in either asthma and/or COPD. However, no reports exist on the comprehensive immunological profile of ACO. Th1 mediated cytokines such as IFN-γ, IL-12 and IL-2 were estimated in ACO. Only IFN-γ showed significantly altered levels in ACO with respect to asthma, COPD and controls. While IFN-γ is believed to inhibit Th2-mediated inflammation, studies among asthmatic patients have yielded conflicting results, including its association with lung function and disease severity [57–61]. TNFα, at increased levels leads to the development of heightened inflammatory responses in asthma and COPD [14, 62]. It was observed to be higher in ACO subjects in comparison to asthma and controls. COPD cases exhibited the highest circulating levels of TNFα.

The Th2-type cytokines, IL-4, 5, and 13, which are associated with the promotion of IgE and eosinophilic responses mostly in atopy and asthma, and IL-10, characterized by anti-inflammatory response, were explored in ACO [14, 62]. Though not significant, these cytokines were found to be upregulated in ACO, with IL-5 and 10 showing significant changes. Our findings open up the possibility of using anti-IL-5 monoclonal antibodies for the management of ACO, similar to that of the treatment suggested for severe asthma [63].

IL-25, also known as IL-17E, is evidenced to be involved in airway inflammation in asthma. It promotes and augments allergic Th2 inflammation via production of IL-4, IL-5, and IL-13 [64]. The expression trend of IL 25 in ACO was similar to that of asthma, which may be attributed to its role in systemic inflammation. However, the primary mediators of Th17 cells such as IL-17A and IL-17F were not estimated which restricts us from generating any conclusive ideas regarding Th17 status in ACO cases.

IL-1β has been associated with systemic inflammation in asthma, COPD as well as exacerbations in both the diseases [65, 66]. It is also suggested that raised IgE levels induce IL-1β expression in monocytes which leads to its increased level in blood [67]. ACO patients were found to have significantly higher levels of IL-1β than controls; however, levels were not as high as observed in asthma or COPD. TGF-β is implicated in several aspects of fibrosis, including deposition of extracellular matrix proteins such as collagens and fibronectin [68]. TGF-β levels were highest in COPD patients followed by ACO which is most likely due to the structural changes in the airway epithelium of these patients.

IL-6 plays a key role in acute phase response and is associated with a variety of clinical and biological parameters in asthma, COPD as well as ACO [69, 70]. We found the IL-6 expression level to be the highest in subjects with ACO. This mediator could be useful in clinics for the identification of patients with high systemic inflammation [71]. We also suggest that anti-IL-6 therapies warrant attention as a possible therapeutic strategy for ACO.

IL-18 is known to enhance Th1 response and has a synergistic effect on IL-12 in inducing IFN-γ release and inhibiting Th2 inflammation. We found the highest expression level of IL-18 in COPD patients. This is in good agreement with the findings of Imaoka et al. (2008) [72] where serum levels of IL-18 in COPD patients and smokers were observed to be significantly higher than that of non-smokers. Furthermore, the group also reported a significant negative correlation of IL-18 with FEV₁ (%) in these patients, which is also in accordance with our observations. IL-18 also has autoimmune regulatory effects on both Th1 and Th2 cytokines [73] and several studies have demonstrated increase in IL-18 activity in Th2 type diseases, such as asthma exacerbations and allergic rhinitis [74].

Other immunological markers such as MCP-1, GM-CSF, YKL-40 and NGAL were also assessed. MCP-1 is evidenced to be higher in blood of both asthma and COPD cases and is strongly related to smoking [75, 76]. GM-CSF is another pleiotrophic and pro-inflammatory cytokine that promotes leucocyte survival and activation, and regulates mucosal immunity and inflammation. YKL-40 a secreted glycoprotein, produced by various cell types, including macrophages, neutrophils, and airway epithelium is reported to be involved in the pathogenesis of COPD, including bronchial neutrophilic airway inflammation and remodelling [77]. It has also been studied in ACO with a few conflicting reports. MCP-1, GM-CSF and YKL-40 were found to be upregulated in ACO with respect to asthma and controls. However, highest expression of these three markers was seen in patients with COPD. NGAL is attributed to activated neutrophils in response to smoke related airway inflammation as well as reactive oxygen species. It is one of the most extensively explored markers of ACO [13, 77, 78]. We found the level to be highest in ACO as compared to asthma, COPD and controls. Our results are in good agreement with reports suggesting higher levels of NGAL in ACO as compared to asthma [13, 78]. Further, a significant negative correlation was observed between serum NGAL and lung function in ACO patients. Our findings are supported by the work of Gao et al. (2016), where an increased NGAL expression in sputum has been independently correlated with degree of airflow limitation in ACO [13]. Other important markers which are frequently studied in asthma and COPD such as periostin, eotaxin and TSLP, though not significant, exhibited an altered profile in ACO.

In recent years, there has been an increasing interest in understanding the group of individuals having features of both asthma and COPD. As the complexity of ACO as a disease entity is gradually unravelled and better understood, a further revision in ACO definition could likely be required [5, 13]. This study in not without limitations. First, owing to the existence of numerous ACO defining guidelines [6–8], the findings of this study are restricted to ACO patients diagnosed as per GINA/GOLD and ATS roundtable diagnostic criteria. Second, this is a single-center study. Genetic variability needs to be accounted for before generalization of the study findings. Similar studies across different countries/continents are, therefore, recommended so that findings on variable ethnicity can be correctly compared. Third, the present findings are limited to patients without active respiratory infections. Alterations in metabolomic profile with lung infections is well realized [79–81]. Since patients with ACO are susceptible to infection [82, 83], it would be worthwhile to investigate ACO metabolomic profiles with and without infections. Fourth, owing to the use of GC-MS platform, this study is limited to the analysis of volatiles with restricted resolution and sensitivity. LC-MS/MS based serum metabolic profiling in the same patient cohort is presently underway. It is envisioned that combining our earlier NMR findings with the complementary GC-MS and LC-MS/MS data will enrich the metabolome coverage and overall improve the data quality. Fifth, in conjunction with serum which reflects analytes in the systemic circulation, it would be useful to analyse the markers at a cellular level in the more proximal biofluid bronchoalveolar lavage fluid (BALF) and lung tissue. However, owing to ethical constraints, this study could not be implemented. Last, our limited sample size may justifiably raise concern regarding robustness of metabolomic data analysis; however, the validation cohort included in this study demonstrates that the dysregulated expression pattern of the metabolites is reproducible and a characteristic of the disease state.

Recently, ACO has been under intensive research. Although several studies have focused on the symptoms and clinical characteristics of ACO, the pathophysiology of this disease has been poorly investigated. In the present study, the levels of sputum HMGB1, HSP70, S100A8, and LL-37 were elevated, while those of sputum Gal-3 were decreased, in patients with ACO compared to healthy controls. Interestingly, the distinct release of these DAMPs between ACO and asthma or COPD indicated that the specific pathophysiology of ACO might differ from that of asthma and COPD. Moreover, sputum HMGB1 may serve as a biomarker for the differentiation of ACO patients from NS, HS, and asthma patients, while sputum LL-37 may differentiate asthmatics from those with COPD. Finally, consistent with previous studies^8,9,11, we found that the levels of these DAMP molecules in the asthma, COPD and ACO patients were significantly different from those in the normal control groups in our study, indicating the important role of DAMPs in the pathogenesis of COPD.

Although an accurate prevalence of ACO was not detected, 13–73% of all COPD patients and 2% of severe asthmatics fulfilled the criteria for ACO^11,16. In this study, 29.8% of patients with ACO were similar to that demonstrated previously¹⁶. In addition, the study¹⁷ demonstrated that the patients with ACO had a lower lung function parameter, while those in our study had lower smoking pack-years and higher FEV1 and FEV1/FVC% compared to the COPD patients.

The role of DAMPs in the pathophysiology of COPD has been reviewed and discussed by Pouwels et al.⁸. Reportedly, HMGB1 levels were elevated in patients with asthma and COPD and independently correlated with the pulmonary function parameters⁹. In a previous study, we also demonstrated that treatment with anti-HMGB1 antibody significantly reversed the development of airway remodeling in a murine asthma model¹⁸. The elevated levels of HMGB1 in ACO suggested its role in the pathogenesis of ACO as well as asthma and COPD. Moreover, the interaction of HMGB1 with TLR-2, -4, -9, and RAGE can promote the activation of neutrophils, macrophages, and dendritic cells, followed by the release of inflammation mediators¹⁹. In addition, the sputum HMGB1 levels in the ACO group were negatively correlated with the FEV1%, indicating that HMGB1 in induced sputum may be a useful biomarker for the disease severity of ACO. Strikingly, the current study stated that sputum HMGB1 had a high sensitivity and specificity in distinguishing ACO from NS, HS, and asthma patients and in distinguishing asthmatics from COPD. The sputum HMGB1 has been recognized as a marker reflecting airway neutrophilic inflammation^9,20. Consistent with these studies, sputum HMGB1 levels were positively correlated with sputum neutrophils in ACO patients. In agreement with the study by Iwamoto et al.²¹, the present results suggested that enhanced airway neutrophilic inflammation may be a characteristic feature of ACO.

HSPs are a family of highly conserved proteins in all cells as well as chaperone proteins. Some studies have found that sputum and plasma HSP70 levels in asthmatics were increased and correlated significantly with clinical parameters, such as FEV1 and FEV1/FVC^10,22. Another study reported that serum HSP70 levels are upregulated in COPD patients compared to healthy controls²³. In agreement with these studies, the current study revealed that patients with asthma, COPD, and ACO had higher sputum HSP70 levels than healthy never-smokers and smokers. Additionally, neither the role of HSP70 in asthma and COPD nor the mechanism of HSP70 underlying the pathogenesis of ACO has been fully elucidated.

S100A8 has been recognized as a DAMP molecule as extracellular S100 proteins can bind to receptors, such as RAGE and TLR4, and both lead to NF-κB activation²⁴. Recently, some studies explored the role of S100A8 in the pathogenesis of asthma and COPD. Mass spectrometry (MS) indicated that S100A8 and S100A9 were elevated in the BALF of COPD patients compared to healthy smokers and never-smokers¹¹. Furthermore, a recent computationally intensive analysis of induced sputum proteome demonstrated reduced levels of S100A8/9 in the induced sputum of asthmatic patients compared to healthy subjects²⁵. To the best of our knowledge, this is the first report comparing sputum S100A8 in patients with asthma, COPD, and ACO. In the present study, the ACO group showed a higher sputum S100A8 level than asthmatic patients but a lower S100A8 level than COPD patients. Since S100A8 is constitutively expressed in neutrophils, the distinct sputum neutrophil counts may partially account for these differences in the levels of S100A8. However, the association between S100A8 and airway inflammation in these diseases remains unclear.

LL-37/hCAP-18 (one of the antimicrobial peptides), primarily secreted by airway epithelial cells, possesses antimicrobial activity against bacteria, fungi, and viruses²⁶. LL-37 also displays a DAMP function as it can provoke a proinflammatory response by binding to TLR7, TLR9, and RAGE and induce necrosis in airway epithelial cells²⁷. Consistent with the present study, some studies revealed that the increased LL-37 concentrations in COPD patients negatively correlated with lung function parameters^28,29. Importantly, ROC curve analysis showed that sputum LL-37 had a high specificity and sensitivity and thus may be a useful and novel biomarker for the differentiation of asthma and COPD. Furthermore, we demonstrated that the sputum LL-37 levels in ACO were significantly higher compared to the non-smoking controls. Based on the previous literature and the current results, we speculated that LL-37 might participate in the inflammatory process underlying ACO.

Gal-3 is a β-galactoside-binding lectin with several physiological functions. Gal-3 is released into the extracellular space and can display a proinflammatory function, termed DAMP³⁰. Only a few studies have investigated the role of Gal-3 in asthma and COPD. Gal-3 levels were significantly decreased in the BALF of patients with COPD and healthy smokers compared to controls³¹. Gao et al.³², for the first time, showed that the concentration of Gal-3 in sputum was significantly reduced in neutrophilic asthma compared to eosinophilic and paucigranulocytic asthma. However, the differences in levels of Gal-3 between asthmatics and healthy controls have not yet been determined. To the best of our knowledge, this is the first report to detect the sputum Gal-3 levels in COPD and ACO patients and to compare the level of Gal-3 in patients with asthma, COPD, and ACO. Interestingly, some studies^31,33 showed that Gal-3 could remove apoptotic neutrophils, avoiding the amplification of inflammation, and that the decreased level of Gal-3 could be associated with defective efferocytosis of macrophages in COPD.

Notably, the relatively large sample size in the present study was beneficial for minimizing the statistical error. The asthmatics were first diagnosed without the use of medicine; the patients with asthma, COPD, and ACO neither suffered from nor had exacerbations during the month before admission, and this excluded the impact of infection. Furthermore, the participants were selected cautiously, especially the ACO patients, who were diagnosed according to the guidelines; the diagnosis criteria of ACO were based on both clinical history and objective parameters. Nonetheless, the present study has several limitations. Demographic differences were noted between the five groups with respect to age, smoking status, and use of drugs. In addition, comparisons between our results and those of other studies might be challenging due to the lack of consensus on the diagnostic criteria for ACO.

In conclusion, the current data provide a novel approach to understanding the pathophysiological features of ACO. Interestingly, compared to asthma or COPD, ACO exhibited a distinct biomarker profile in sputum DAMP levels. Importantly, sputum HMGB1 is a novel biomarker for differentiating ACO from healthy controls, and sputum LL-37 may also serve as a valuable biomarker for differentiating asthma and COPD.

Purpose: To analyse the one-month change in subfoveal choroidal thickness (SFChT) of myopic children treated with 0.01 % atropine, orthokeratology (OK), or their combination.

Methods: This is a prospective, randomized controlled trial. One hundred fifty-four children aged between 8 and 12 years with a spherical equivalent (SE) of -1.00 to -6.00 diopters were enrolled. Subjects were randomly assigned to receive 0.01 % atropine and orthokeratology (ACO, n = 39), 0.01 % atropine and single vision glasses (atropine, n = 42), orthokeratology and placebo (OK, n = 36), or placebo and single vision glasses (control, n = 37). SFChT was assessed using optical coherence tomography (OCT). Ocular parameters, including axial length (AL), were measured using a Lenstar LS 900.

Results: SFChT significantly increased in the ACO (14.12 ± 12.88 μm, p < 0.001), OK (9.43 ± 9.14 μm, p < 0.001) and atropine (5.49 ± 9.38 μm, p < 0.001) groups, while it significantly decreased in the control group (-4.81 ± 9.93 μm, p = 0.006). The one-month change in SFChT was significantly different between the control and treatment groups (p < 0.001). The results of pairwise comparisons among the treatment groups showed that the magnitude of the SFChT change was larger in the ACO group than in the atropine group (p = 0.002). The changes in the ACO and OK groups were not significantly different (p = 0.326).

Conclusion: The combination of OK and atropine induced a greater increase in SFChT than monotherapy with atropine, which might indicate a better treatment effect for childhood myopia control.

Ant colony optimization (ACO) is an efficient heuristic algorithm for combinatorial optimization problems, such as clustering. Because the search strategy of ACO is similar to those of other well-known heuristics, the probability of searching particular regions will be increased if better results are found and kept. Although this kind of search strategy may find a better approximate solution, it also has a high probability of losing the potential search directions. To prevent the ACO from losing too many potential search directions at the early iterations, a novel pheromone updating strategy is presented in this paper. In addition to the “original” pheromone table used to keep track of the promising information, a second pheromone table is added to the proposed algorithm to keep track of the unpromising information so as to increase the probability of searching directions worse than the current solutions. Several well-known clustering datasets are used to evaluate the performance of the proposed method in this paper. The experimental results show that the proposed method can provide better results than ACO and other clustering algorithms in terms of quality.

Partitional clustering is a classical NP-hard problem [1–3]. The goal of this problem is to divide a set of patterns into several different groups based on their features. Moreover, the basic concept of clustering is simply to put similar data together. For example, consider four flowers with different colors: red, yellow, blue, and purple. They can be split up into warm color group with red and yellow flowers and cool color group with blue and purple flowers. Clustering is important because technologies relevant to it can be applied to a large number of practical applications, such as bioinformation [4, 5], document analysis [6, 7], human face recognition [8, 9], and search engine [10–12].

Jain and Dubes [1] divided the clustering process into several steps: pattern representation, definition of a pattern proximity measure appropriate for the data domain, clustering or grouping, data abstraction, and assessment of output. Pattern representation refers to the number of data, groups, and other features available to the clustering algorithm. Pattern proximity uses objective function (or fitness function) to measure the difference of patterns. Two kinds of measurements are usually used for evaluating the results of clustering. The first kind considers the degree of closeness between patterns. For instance, one of the well-known measurements is to calculate the sum of distances between patterns and centroids to which the patterns belong. The second kind considers the difference among the groups. One widely used measurement is to compute the sum of distances among centroids. A large sum implies that groups are quite different from each other.

Since clustering technologies can be used in our daily life, a large number of search methods were presented to solve the partitional clustering problem of which

-means [13, 14] is one of the most widely used, for it is simple and easy to implement. Initially, -means generates a set of centroids, the number of which is predefined. Then, the patterns will be assigned to the nearest groups (centroids). At this step, patterns assigned to the same group are regarded as in the same group. After the assignment process, -means will then recalculate the centroid of each group by averaging the positions of patterns in that group. The assignment and update processes will be repeated until the positions of all centroids do not change or the changes are less than a predefined threshold. Although -means are simple and easy to implement, the result is extremely sensitive to the initial solution. Another drawback is that -means is easily falling into local optima, and there is simply no way to escape from the local optima.

Therefore, many researches attempt to solve the clustering problem using different metaheuristic algorithms. Tabu search (TS) [15] uses a tabu list to keep track of solutions that have been tried recently so as to avoid searching for the same solutions repeatedly in the near future. Genetic algorithm (GA) [16, 17] uses the crossover and mutation operations to share information available between chromosomes to enlarge the search region. Recently, another promising research trend has been using particle swarm optimization (PSO) [18, 19] that takes into account both the population experience and the individual experience to obtain a better clustering result than traditional clustering algorithms do. Another swarm intelligence, ant colony optimization (ACO) [20], has also been applied to the clustering problem, by having the ants put pheromone on the paths they passed through so as to share the information learned by each ant. This strategy can also lead the algorithm to find good solutions.

Generally speaking, ACO is a powerful search method, especially for combinatorial optimization problems, because all the search information can be shared and accumulated on the pheromone table. However, for some optimization problems, ACO also suffered from the problem of falling into local optima at early iterations even though it has a better chance to select different search directions. Also, paths with higher pheromone concentration do not always lead to the final goal. Hence, we present in this paper an ACO-based clustering algorithm, called multiple pheromone table based ant colony optimization (MPTACO), to solve this problem (this is basically an extended version of our previous study [21]). A second pheromone table is added to the ACO to keep track of the unpromising information so that the ACO can try to find better clustering results by using this unpromising information. To evaluate its performance, the proposed method is compared with four state-of-the-art algorithms: standard -means [14], genetic -means algorithm [16], ant colony system (ACS) [22], and ACODPT [21].

The remainder of the paper is organized as follows. Section 2 gives a brief introduction to the ACO-based algorithms and improvement on them. Section 3 describes in detail the proposed algorithm. Performance evaluation of the proposed algorithm is presented in Section 4, and the conclusion is drawn in Section 5.

The basic idea of ant colony optimization (ACO) comes from the foraging behavior of ant colony [23]. The foraging behavior of an ant colony is that ants will leave pheromone on the paths they passed through for the other ants to find the food. The ants tend to select the path with higher pheromone concentration. For this reason, a path with higher pheromone concentration will lead more ants to go through the path, implying that the path has a larger chance to find food.

Algorithm 1 gives an outline of the ACO algorithm. In the solution construction process, the ants need to decide which paths to take. To realize this concept, the probability of choosing

as the next path is defined as follows:where denotes the set of candidates which ant can reach from the current position ; is the pheromone value on the path ; is the inverse of the distance of the path ;

and

are the weight of exploration and exploitation, respectively. The operator for updating the pheromone concentration on each path is defined as follows:where denotes the pheromone decay which is a value in the range ;

is the number of ants; is the length of the tour created by ant .

In a later research [22], Dorigo et al. presented another algorithm called ant colony system (ACS) which is an extended version of [23]. ACS enhances the convergence of the original ACO. The algorithm gives ants a certain chance to directly select the path with the highest pheromone concentration as follows:where denotes a random value and

is a threshold for determining the strategy to be used.

The first strategy selects the best next partial solution in the case while the second strategy selects the next partial solution determined by the following probability distribution:For this reason, the ant will not always check the probability of each path. Therefore, ACS can speed up the convergence process, thus reducing some of the computation time.

Shelokar et al. [24] presented an ACO-based clustering algorithm which uses pheromone trails. The main idea is transferring the meaning of information in the solutions. Different from the original version of ACO for the traveling salesman problem (TSP) in which the solutions record the order of the cities the salesman travels, for the clustering problem, the solutions [24] record the cluster numbers to which the patterns are assigned, by using an

by

pheromone matrix where is the number of nodes and is the number of centroids. The matrix records the pheromone concentration on the path between each pattern and each centroid. As illustrated in Figure 1, unlike the pheromone table which is represented by a matrix, each solution is encoded as a one-dimensional array, the indices of which correspond to the pattern number while the value of each element corresponds to the cluster to which that pattern is assigned. The example shows that pattern 1 is assigned to centroid 3, pattern 2 is assigned to centroid 5, pattern 3 is assigned to centroid 8, and so on. By keeping track of to which cluster each pattern is assigned as illustrated in Figure 1, ACO can be applied to the clustering problem. Moreover, the local search method is also added to speed up the convergence.

In [25], different strategies are added to the ACO to decide the solution. One of the strategies is that ants only check part of the paths instead of all the paths. Another strategy is using a threshold of pheromone to split patterns into different groups and checking the groups at the end. If only a few patterns are in a group and the centroid of that group is close to the centroid of another group, the two groups are merged. In [26], several additional steps are used in the ACO. First, the root cluster, which contains all the patterns, is split into several smaller clusters, and each pattern is assigned to a suitable cluster. Then, clusters which are close to each other are merged. Finally, dissimilar patterns are removed from their clusters, and new clusters for these patterns are constructed.

More recently, Tiwari et al. [27] used two strategies to enhance the performance of ACO. One is setting the current value of pheromone to the initial value once every 50 iterations while the other is setting the pheromones on all paths to the initial value if the pheromones on the paths remain intact for the past 10 iterations in a row. Akarsu and Karahoca [28] used sequential backward selection (SBS) to reduce the dimensions of patterns. They use Manhattan distance as the fitness function and consider the pheromone only when the ants select paths. Jiang et al. [29] split the structure of solutions into two parts. The first part is used to decide which patterns will be included in the following computation. The second part is used to record the relationship between patterns and clusters. Some researches are focused on self-adaptive clustering. For instance, Liu and Fu [30] presented a method which uses Jaccard index to decide a suitable cluster number. Niknam and Amiri [31] presented a method which combines fuzzy adaptive particle swarm optimization (FAPSO), ACO, and -means algorithm. The experiment results show that these methods get a better performance.

Since the clustering problem is a traditional optimization problem, a large number of clustering algorithms have been presented for several years [1]. As we mentioned in Section 1, -means is one of the most well-known clustering algorithms, but its search process may fall into local minima because it is very sensitive to the randomly generated initial solution and it does not have any mechanism to escape the local optima [32]. Metaheuristic algorithms provide an alternative way to prevent the search process from falling into local minima. One of them is the simulated annealing [33] which has a small chance to accept a solution that is worse than the current solution as the new search direction. In [34], tabu search used the so-called tabu list to avoid the search process from falling into local minima. Different from simulated annealing and tabu search, genetic algorithm (GA) [35] used multiple search directions to keep the search direction from falling into local minima or searching the same region repeatedly. Also because GA has a mutation operator, a small percentage of subsolutions can be disturbed. This implies that GA has a chance to escape the local minima. Particle swarm optimization [36] and ant colony optimization [23] also used multiple search directions to search for the solution; they will not easily fall into local minima at early iterations. In summary, most clustering algorithms will confront the dilemma of local minima; therefore, how to mitigate this problem has become an active research topic in recent years. Using the metaheuristic algorithm alone and combining it with traditional clustering algorithms are two promising solutions to keep the search process from falling into local minima at early iterations [3].

In this section, we will describe in detail the proposed algorithm MPTACO in three steps: first the concept, then the proposed algorithm, and finally an example.

In general, most ACO-based algorithms, even most metaheuristic algorithms, focus on the positive information. For example, for the ACO, ants move along the trail with higher pheromone; for GA, fitter chromosomes survive; for PSO, particles move toward the best particle. However, searching the so-called good directions does not always lead to good results. In fact, the trajectory of search is tortuous when dealing with complex problems. To prevent the ACO from choosing only paths that have high pheromone values and remove paths that have low pheromone values so as to increase the search diversity on the convergence process, the proposed algorithm adds a second pheromone table to keep track of negative information. The consequence is that the proposed algorithm has a better opportunity to select worse partial solutions during the search process so as to find solutions better than those the original ACO can find. As we mentioned in Section 2.3, although most metaheuristic algorithms have mechanisms to avoid their search process from falling into local minima at early iterations, they do not guarantee that their search process will not fall into local minima. Because the search direction of metaheuristic algorithms typically is toward a result that is better in terms of the fitness value (i.e., the value returned by a fitness or objective function), if the algorithm cannot find a better result for a long while before it finds the global optima, the search process might get stuck at a particular position or region. This phenomenon is called falling into local optima. Most of the metaheuristic algorithms have some strategies to escape the local optima.

According to our observation, the search process of ACO will move toward particular solutions (or paths), because the pheromone values associated with these solutions will be strengthened if they are better than the other solutions ACO finds. One of the reasons is that ACO keeps only the promising solutions. To avoid the search process from moving toward particular search directions, a different way to construct the solutions is considered in this paper, that is, taking into account the information of unpromising solutions ACO finds. As shown in Figure 2,

denotes the

th pattern,

is the

th centroid, and a solid line is the assignment of a pattern to a centroid. This example shows that pattern is assigned to centroid

, pattern

is assigned to centroid , and pattern is assigned to centroid . The ant picks first path and then path . Afterwards, the proposed algorithm eliminates the path , based on values in the negative pheromone table and the selection strategy, and then selects path . The dashed line denotes the virtual path, which is used to show that the ant moves continuously. In the following subsection, we will discuss briefly the concept first and then explain how the ants choose paths.

As shown in Figure 3, the major difference between the proposed method and the ACO is in that MPTACO employs two tables, which we refer to as the positive pheromone table and the negative pheromone table, in the transition phase to decide which path an ant is supposed to take. The positive pheromone table is the only pheromone table in ACO, which is used to select paths in the candidate list, whereas the negative pheromone table is used by MPTACO to eliminate paths from the candidate list.

In the initialization phase, that is, initially, the value of each path in the positive pheromone table is set equal to while the value of each path in the negative pheromone is set equal to 1. Moreover, since the negative pheromone table is used to keep track of the probability of selection, all the paths will be in the candidate list at the very beginning. For each ant, the initial partial solution is set randomly. In other words, each pattern is assigned to a cluster randomly, and then the centroids are calculated. The proposed method represents solutions in such a way that it is the same as described in [24], that is, to which centroid each pattern belongs. Instead of binary encoding, integer encoding is used for two reasons: (1) it is easier to understand the meaning of the solutions, and (2) it is shorter to integer encode the solutions. But, on the other side, integer encoding usually takes more memory space than binary encoding in storing the information.

This phase decides the paths in which ants move. In other words, this phase decides the relationship between patterns and centroids. Two kinds of strategies are used for the transition of the proposed algorithm. Equation (5) shows how the decision is made in the proposed algorithm for the transition phase:where is the number used to decide which strategy to choose, and are the same as defined in ACS, is the weight to control the influence of distance, and is a random variable that is used to select the next partial solution based on the following probability distribution:with representing the probability of pattern of ant belonging to centroid in the next partial solution, the set of paths which ant can reach from pattern , and , generated by (7), the set of paths which will not be selected:where is a random number uniformly distributed in the range of 0 to 1. It will be regenerated every time the algorithm checks the negative pheromone.

Like ACS, the ants have some chances to select the path with the highest positive pheromone concentration. The negative pheromone used in the second strategy is the probability to determine whether a path will be added to the candidate list or not. Then, the ants will select the paths in terms of the positive pheromone from the candidate list. This implies that the proposed algorithm has an opportunity to eliminate the paths which are really bad. The remaining paths will be chosen with a higher probability. Moreover, the main difference between MPTACO and ACODPT [21] is in the transition phases of them. For ACODPT, three search strategies are used by each ant. For MPTACO, the search strategies for each ant have been simplified to two, and , which is defined asis used to increase the search diversity at early iterations of the proposed algorithm.

Figure 4 gives a simple example to illustrate how the transition phase works. First, the example shows that a number of paths can be selected, each of which is associated with a pair of numbers denoted by

where denotes the positive pheromone and is the negative pheromone. For instance, the positive pheromone of path is while the negative pheromone is . The first strategy will select the path with the most positive pheromone, so path is selected. The second strategy assumes that path

has been eliminated so that the probability of choosing the remaining paths increases. Here we assume that path

is selected in this example.

In general, the proposed algorithm will search more broadly at early iterations and more deeply at later iterations. For this reason, the proposed algorithm has a larger chance to select the first strategy at early iterations and the second strategy at later iterations. Which strategy is taken depends on the threshold defined above.

The proposed algorithm will then update the positions of centroids by using the current partial solution for each ant as follows:where denotes the th pattern of th centroid and is the number of patterns belonging to centroid .

Same as described in [23], the positive pheromone table will be updated by using the following equation:where denotes the pheromone value associated with the path , is the evaporation degree, and is the initial pheromone on each path. This phase updates the positive pheromone on the paths which are taken by the ants. If a path is taken by more ants, the positive pheromone on the path will be higher.

This phase will update both the positive and the negative pheromone tables. The algorithm will choose the best ant so far to update the positive pheromone table. The positive pheromone on the paths passed through by the best ant will be enhanced. The best ant means the ant with the best fitness value. For example, if the sum of squared errors (SSE) is used as the fitness value, the ant with the smallest SSE will be the best. The pheromone update on all the paths of the best ant is defined aswhere denotes the influence degree of the best ant and is the total distance which the best ant has moved. As for the updating of negative pheromone table, the proposed algorithm will select the worst ant so far to update the negative pheromone on the paths through which it passed. Same as the best ant, the worst ant means the ant with the worst fitness value. For every path the worst ant has passed through, the negative pheromone will be updated as follows:where is a constant in the range

, indicating the evaporation rate of negative pheromone. Besides, the proposed algorithm will also update the negative pheromone on all the paths which are passed through by the best ant so far as follows:Because is used as a probability to decide whether the path should be put into the candidate list or not, the value of has to be in the range . That is, if is larger than 1, it will be set to 1. A larger negative pheromone of the path means that the path has a higher probability of being put into the candidate list. The path passed through by the best ant means that it may be the best choice of the final solution. So the selection probability of the path will increase.

The termination criteria can vary. For instance, the search process can be terminated if the total distance of any ant is shorter than a threshold or if the difference between ants is less than a threshold. In this paper, the search process will be terminated if the number of iterations has reached a predefined number.

In ACO, only one kind of pheromone is used. All the ants use the same kind of pheromone to decide the search directions. However, using only one kind of pheromone to decide the search directions may neglect some search directions which have the potential to lead to better results, and it may also lead the search process to fall into local optima easily. For this reason, ACODPT adds the second kind of pheromone, which is referred to as the negative pheromone, to improve the quality of the end result. The negative pheromone is used to eliminate the search directions, thus increasing the chance the remaining search directions are selected. This way, the proposed method can have a higher probability to search for all the potential directions. Compared to ACODPT and MPTACO, the number of decision strategies in the transition operator of ACODPT is three, but it has been reduced to two in MPTACO. More important, the transition operator of MPTACO can select the decision strategies automatically according to the search time instead of the probability.

Comparing to the hill-climbing method, ACO, ACODPT, and MPTACO do not always search for better solutions. The selection strategies give them chances to select worse directions. If the algorithm always accepts a better solution just like the hill-climbing method does, it will fall into local optima at early iterations because it does not search for other possibilities in a large search space. Moreover, the pheromone tables and the corresponding strategies make the proposed method capable of mitigating the problem of falling into local optima quickly.

The experiments are conducted on a PC with 2.67 GHz Intel Core i7-920 CPU and 4 GB of memory running Fedora 12 with Linux 2.6.31.5-127. Also, the programs are written in C and compiled by gcc version 4.4.2 20091027 (Red Hat 4.4.2-7).

The datasets used in the experiments are taken from UCI [37], the details of which are as described in Table 1.

To evaluate the performance of MPTACO, we compare it with four state-of-the-art clustering algorithms: standard -means [14], genetic -means algorithm (GKA) [16], ant colony system (ACS) [22], and ACODPT [21]. The reasons that these algorithms are chosen for the purpose of comparison are in order. Standard -means is a simple, easy to implement, and widely used algorithm for the clustering problem. It is a single-solution-based algorithm and has the characteristic of not being able to escape the local optima. For this reason, -means can be used to compare the difference when using strategies that can escape the local optima. GKA for clustering leverages the strength of genetic algorithm (GA) and -means; thus, it is not only capable of a deep search, but also capable of escaping the local optima. Because ACODPT is based on ACS and MPTACO is based on ACODPT, this explains why ACS and ACODPT are used to evaluate the performance of MPTACO.

The parameter settings of each algorithm are as follows. Except -means, the population sizes of GKA, ACS, and MPTACO are fixed at 30. For the ACS and MPTACO, the population size is the number of ants. For the GKA, the crossover operator is replaced by a 2-iteration -means operator, the mutation probability is 0.01, and the number of generations is 1,000. For the ACS and MPTACO, is set to 0.1, to 11.0, to 0.1, and to 0.01. For the ACS, is set to 1. All the algorithms are run for 30 times, and the averages are taken as the experimental results. Tables 2 and 3 show, respectively, the quality and the computation time. The numbers in bold indicate which algorithm gives the best result for the dataset evaluated. The quality is measured by the sum of squared errors (SSE) defined asand the numbers after

are the standard deviation. This implies that the smaller the value, the better the result.

Table 2 shows that the proposed method outperforms the other algorithms evaluated in most cases in terms of the quality. -means gets the worst results for all datasets because it has no strategy to escape the local optima. As for GKA, it combines GA and -means to get rid of the weakness of -means. This is why GKA can find a better result, even the best result for some of the datasets. ACS has a strong global search capability, as is reflected in complex datasets like abalone. Nevertheless, the local search capability of ACS is not as powerful as that of GKA. As such, for some simple datasets, the result is worse than that of GKA. The proposed method beats ACS not only for complex datasets but also for simple datasets. For some datasets, the proposed method provides a worse result than that of GKA and ACODPT. This can be easily justified by observing that the proposed method lacks the local search strategy to fine-tune the result.

Table 3 shows the computation time of each algorithm. Undoubtedly, -means takes the least time because it only searches toward the best direction. The performance of the proposed method is the worst in terms of the computation time. This is basically a trade-off between quality and computation time. For applications the computation time of which is not at a premium, but the quality is a major concern, then the proposed algorithm is a good choice.

Besides the previous experiments, we are also interested in the scales of the values in the datasets. It is obvious that attributes of datasets may influence the clustering result because of the different scales of the values in the datasets. Table 4 shows the results after the attributes of all the datasets are normalized to be in the range . As can be easily seen from the results, the best value of some datasets has been changed. This implies that the scale of attributes can influence the clustering result.

In this experiment, we test the influence of the parameter , which is used to control the proportion between positive pheromone and distance. Figure 5 shows the results. The four datasets tested are abalone, yeast, normalized abalone, and normalized yeast. The table shows that a larger implies a better quality. This means that the distance has a stronger influence on the result than the positive pheromone. This is because SSE is used as the fitness function, which considers only distances between patterns and centroids. If a different fitness function is used, the result may be different.

in (8) is used to automatically decide the search strategy according to the total number of iterations and the current iteration number. At the beginning of the search procedure, the diversification is respected to discover more regions that may have better local optima. The algorithm will have a higher chance to choose the search direction which is not the best at the moment. As the time passes by, the search procedure will focus on the intensification. In the end of the search procedure, the algorithm tries to converge to the local optima of the regions. In other words, this strategy makes the algorithm search widely at the beginning but deeply in the end.

In this paper, we proposed an efficient algorithm, called MPTACO, to enhance the quality of the ACO-based algorithms for clustering. The idea is adding a second table to record the negative pheromone and to eliminate the most impossible paths so as to increase the probability of choosing worse paths in the table. It can prevent the ants from falling into local optima again and again. In addition, MPTACO can automatically adjust the convergence process according to the number of iterations carried out so far. The experimental results show that the proposed method can get a better result in most cases, especially for large-scale and complex datasets. Our future goal is to reduce the computation time of the proposed method while at the same time enhancing its quality. Also, another concern is how to use the negative pheromone table more efficiently.

Pseudocode 1 gives the pseudocode of the proposed algorithm MPTACO. As the pseudocode shows, the positive and negative pheromones of each path will be initialized to and , respectively, before the main loop is performed repeatedly until the termination criterion is met, as follows. First, all the ants are transited to the new states in the transition phase, as lines (10)–(12) show. Next, in the means update phase, MPTACO will update the centroids of each ant, as lines (14)–(16) show. Then, in the local pheromone update phase, the positive pheromone values will be updated, as lines (18)–(21) show. Finally, in the global pheromone update phase, the best and worst ants will be found and the positive and negative pheromone values on the paths they passed through will be updated, as lines (23)–(28) depict. Once the termination criterion is met, MPTACO will output the best result (line (30)) and terminate the search process.

The authors declare that there is no conflict of interests regarding the publication of this paper.

The authors would like to thank the editors and anonymous reviewers for their valuable comments and suggestions on the paper that greatly improve the quality of the paper. This work was supported in part by the Ministry of Science and Technology of Taiwan, under Contracts MOST103-2221-E-110-061, MOST103-2221-E-197-034, and MOST103-2221-E-006-145-MY3.